A system and method for port yard design scheme question and answer checking and optimization

Abstract

The application discloses a system and method for port yard design scheme question and answer checking and optimization, the system comprises: a question and answer analysis unit for obtaining design task information, analyzing the design task information through a natural language processing model to obtain a structured design task; a multi-objective evaluation unit for setting a function loss value based on the structured design task, and constructing a multi-objective balanced loss function according to the function loss value; an iterative optimization unit for iteratively optimizing and calculating the multi-objective balanced loss function through an improved Bayesian algorithm, and outputting a preliminary design scheme; and a feedback adjustment unit for obtaining user feedback information, analyzing the user feedback information through a natural language model to obtain feedback quantization results, and updating the multi-objective balanced loss function according to the feedback quantization results. The system and method provided by the application can accurately output a design scheme for a port project by processing and quantizing expert feedback opinions.

Description

Technical Field

[0001] This invention relates to the field of intelligent optimization technology for port engineering design, and in particular to a system and method for question-and-answer verification and optimization of port yard design schemes. Background Technology

[0002] Port engineering environments possess significant unique characteristics, and their design optimization cannot simply apply general-purpose models. Existing design processes heavily rely on human experience, resulting in low efficiency and inconsistent solution quality. While some intelligent auxiliary tools have emerged in recent years, they generally suffer from two major technical bottlenecks: first, they struggle to adaptively optimize weights among conflicting objectives such as cost and safety; second, existing systems are unable to process and quantify qualitative and vague feedback from experts (e.g., "the design feels too conservative" or "the economic viability of the pile foundation scheme is questionable"), making it impossible to use this feedback as direct input to the optimization algorithm. This leads to a severe disconnect between valuable expert knowledge and the intelligent computing process. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a system and method for question-and-answer verification and optimization of port yard design schemes, which can accurately output design schemes for port projects by processing and quantifying expert feedback.

[0004] In a first aspect, embodiments of the present invention provide a system for question-and-answer verification and optimization of port yard design schemes, comprising: a question-and-answer parsing unit, a multi-objective evaluation unit, an iterative optimization unit, a feedback adjustment unit, and a result output unit;

[0005] The question-answering parsing unit is used to obtain design task information and parse the design task information through a natural language processing model to obtain a structured design task.

[0006] The multi-objective evaluation unit is used to set the function loss value based on the structured design task and construct a multi-objective balanced loss function based on the function loss value.

[0007] The iterative optimization unit is used to perform iterative optimization calculations on the multi-objective balance loss function by improving the Bayesian algorithm, and to output a preliminary design scheme before the numerical convergence of the multi-objective balance loss function.

[0008] The feedback adjustment unit is used to obtain user feedback information based on the preliminary design scheme, and to parse the user feedback information through a natural language model to obtain the feedback quantification result, and to update the multi-objective balance loss function according to the feedback quantification result.

[0009] The result output unit is used to output the target design scheme after the numerical convergence of the multi-objective balanced loss function.

[0010] Furthermore, the design task information is textual information of the design task expressed in natural language; the structured design task includes design objectives and design parameter space, and the design objectives include specification constraint objectives, economic objectives, and performance objectives.

[0011] Furthermore, the function loss values ​​include specification compliance loss, economic loss, performance loss, and user consistency loss;

[0012] The expression for the multi-objective balanced loss function is:

[0013] L = W1L1 + W2L2 + W3L3 + W4L4;

[0014] In the formula, L is the overall loss; L1 is the specification compliance loss; L2 is the economic loss; L3 is the performance loss; L4 is the user consistency loss; W1 is the first weight parameter; W2 is the second weight parameter; W3 is the third weight parameter; and W4 is the fourth weight parameter.

[0015] Furthermore, the specification compliance loss L1 is used to represent the specification compliance loss of the design scheme, and its expression is:

[0016] ;

[0017] In the formula, S i S represents the safety criticality of the i-th design specification. i The value of S is [0, 10]. i A higher value for V indicates a higher level of safety criticality; i V represents the degree to which the design scheme violates the i-th design specification. i The value of l is [0, 1]; i This represents the mutual influence factor between design specifications; N is the number of design specifications.

[0018] Economic loss L2 is used to represent the economic loss of the design scheme. The economic loss mainly includes direct engineering cost loss, and its expression is:

[0019] L2 = (C1 - C0) / C0;

[0020] In the formula, C1 is the estimated direct engineering value, and C0 is the target direct engineering value;

[0021] The performance loss L3 is used to represent the safety loss of the design scheme, and its expression is:

[0022] ;

[0023] In the formula, R k P represents the risk level of the k-th key safety indicator. kE represents the probability coefficient of the occurrence of the k-th key safety indicator. k This represents the environmental impact factor, where P is the total number of key safety indicators.

[0024] User consistency loss L4 is used to represent the loss caused by inconsistencies between the initial design and user feedback. The initial value of L4 is 0.

[0025] Furthermore, the first weight parameter W1 is the weight parameter for specification compliance loss; the second weight parameter W2 is the weight parameter for economic loss; the third weight parameter W3 is the weight parameter for performance loss; and the fourth weight parameter W4 is the weight parameter for user consistency loss. The first weight parameter W1, the second weight parameter W2, the third weight parameter W3, and the fourth weight parameter W4 satisfy the following expression:

[0026] W1+W2+W3+W4=1;

[0027] In the formula, W1, W2, W3, and W4 are all adjustable constants, and their values ​​are all in the range of [0, 1].

[0028] Furthermore, the iterative optimization calculation of the multi-objective equilibrium loss function by improving the Bayesian algorithm includes:

[0029] Based on the engineering scenario, set prior probability distributions for the first, second, third, and fourth weight parameters in the multi-objective balance loss function;

[0030] A proxy model is established based on combined kernel functions;

[0031] Candidate optimal design parameters are obtained by predicting and simulating the multi-objective equilibrium loss function using a surrogate model, and the actual loss value of the candidate optimal design parameters is calculated.

[0032] The actual loss value under the candidate optimal design parameters is evaluated to obtain the evaluation result. The surrogate model is iteratively optimized using the evaluation result until the multi-objective balance loss function converges.

[0033] Furthermore, the expression for the combined kernel function is as follows:

[0034] K=K g ×K p ×K s ×C;

[0035] In the formula, K g As the basic core, it is used to capture the general nonlinear relationships between design parameters; K p A periodic kernel, used to handle the effects of periodic loads; K s C is the spatial kernel, used to handle spatial correlation; C is the constraint kernel, used to handle the boundary effects of rigid constraints.

[0036] Furthermore, updating the multi-objective balance loss function based on the feedback quantification results includes: updating the user consistency loss of the multi-objective balance loss function using a penalty function based on the feedback quantification results, while dynamically increasing or decreasing any one or more of the first weight parameter, second weight parameter, third weight parameter, and fourth weight parameter.

[0037] Furthermore, the question-and-answer parsing unit and the feedback adjustment unit are equipped with human-computer interaction modules, on which users can input design task information; users can also review the preliminary design scheme and input user feedback information.

[0038] Secondly, embodiments of the present invention provide an application method for a system for question-and-answer verification and optimization of port yard design schemes, including:

[0039] S1. Use the question-answering parsing unit to obtain design task information, and use a natural language processing model to parse the design task information to obtain a structured design task.

[0040] S2. Using a multi-objective evaluation unit, set the function loss value based on the design objective, and construct a multi-objective balanced loss function based on the function loss value.

[0041] S3. The iterative optimization unit is used to perform iterative optimization calculation of the multi-objective balance loss function by improving the Bayesian algorithm, and a preliminary design scheme can be output before the numerical convergence of the multi-objective balance loss function.

[0042] S4. Use the feedback adjustment unit to obtain user feedback information based on the preliminary design scheme, and use a natural language model to parse the feedback information to obtain the feedback quantification result. Update the multi-objective balance loss function based on the feedback quantification result.

[0043] S5. Utilize the result output unit to output the target design scheme after the numerical convergence of the multi-objective balance loss function.

[0044] The embodiments of the present invention bring the following beneficial effects:

[0045] 1. This invention achieves the quantification and dynamic balancing of multiple objectives, improving the overall optimality of the design scheme. Addressing the pain point of existing technologies struggling to balance multiple objectives such as economy, safety, and standardization, this invention constructs a unified multi-objective balancing loss function, placing all design considerations within a unified and quantifiable mathematical framework. Simultaneously, the optimization process is no longer a compromise to a single objective, but a systematic search for comprehensive losses, ensuring that the final solution is the globally optimal solution under all constraints and objectives, thus avoiding the "weakest link" effect.

[0046] 2. A closed-loop learning and evolutionary capability based on human-machine collaboration has been established, enabling the quantitative transfer of expert experience. Addressing the challenge of effectively integrating qualitative expert feedback in existing technologies, this invention, through user consistency loss and natural language processing quantification techniques, has for the first time established a pathway from "qualitative feedback" to "quantitative model input." This allows valuable expert experience and engineering intuition to directly guide the algorithm's iteration direction, forming an intelligent closed loop of "algorithm recommendation -> expert correction -> model evolution." The system not only solves the current problem but also continuously learns and evolves, becoming increasingly aligned with engineering practice.

[0047] 3. Significantly improves the scenario adaptability and convergence efficiency of the optimization algorithm, achieving rapid response. Addressing the issue that general algorithms cannot be used in specific engineering scenarios, this invention introduces "scenario-aware prior distribution" and "customized combined kernel functions," enabling the algorithm to possess "prior knowledge" about the specific engineering project (such as soft soil foundations or heavy-duty storage yards) before optimization begins. This allows it to more quickly locate high-value areas in complex parameter spaces, significantly shortening the optimization time and improving design efficiency.

[0048] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0050] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0051] Figure 1 This is a schematic diagram of a system for question-and-answer verification and optimization of port yard design schemes provided in an embodiment of the present invention;

[0052] Figure 2 This is a flowchart illustrating an application method for a system for question-and-answer verification and optimization of port yard design schemes, provided in an embodiment of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Example 1

[0055] To facilitate understanding of this embodiment, in conjunction with Figure 1 This invention provides a detailed description of a system for question-and-answer verification and optimization of port yard design schemes, as disclosed in an embodiment of the present invention.

[0056] This embodiment discloses a system for question-and-answer verification and optimization of port yard design schemes, including: a question-and-answer parsing unit, a multi-objective evaluation unit, an iterative optimization unit, a feedback adjustment unit, and a result output unit.

[0057] The question-answering parsing unit is used to obtain design task information and parse the design task information through a natural language processing model to obtain a structured design task.

[0058] The design task information is textual information of the design task expressed in natural language; the structured design task includes design objectives and design parameter space, and the design objectives include specification constraint objectives, economic objectives, and performance objectives.

[0059] The multi-objective evaluation unit is used to set the function loss value based on the structured design task and to construct a multi-objective balanced loss function based on the function loss value.

[0060] The function loss values ​​include specification compliance loss, economic loss, performance loss, and user consistency loss. Specification compliance loss, economic loss, and performance loss are parameterized variables calculated based on specification constraint objectives, economic objectives, and performance objectives, respectively.

[0061] The expression for the multi-objective balanced loss function is:

[0062] L = W1L1 + W2L2 + W3L3 + W4L4

[0063] In the formula, L is the overall loss; L1 is the specification compliance loss; L2 is the economic loss; L3 is the performance loss; L4 is the user consistency loss; W1 is the first weight parameter; W2 is the second weight parameter; W3 is the third weight parameter; and W4 is the fourth weight parameter.

[0064] Furthermore, the specification compliance loss L1 is used to represent the specification compliance loss of the design scheme, and its expression is:

[0065] The specification compliance loss L1 is used to represent the specification compliance loss of the design scheme, and its expression is:

[0066] ;

[0067] In the formula, S i S represents the safety criticality of the i-th design specification. i The value of S is [0, 10]. i A higher value for V indicates a higher level of safety criticality; i V represents the degree to which the design scheme violates the i-th design specification. i The value of l is [0, 1]; i This represents the mutual influence factor between design specifications; N is the number of design specifications.

[0068] Economic loss L2 is used to represent the economic loss of the design scheme. The economic loss mainly includes direct engineering cost loss, and its expression is:

[0069] L2 = (C1 - C0) / C0;

[0070] In the formula, C1 is the estimated direct engineering value, and C0 is the target direct engineering value.

[0071] Specifically, when the estimated direct engineering value C1 exceeds the target direct engineering value C0, the economic loss L2 is positive. The greater the difference between the estimated direct engineering value C1 and the target direct engineering value C0, the greater the economic loss L2.

[0072] The performance loss L3 is used to represent the safety loss of the design scheme, and its expression is:

[0073] ;

[0074] In the formula, R k P represents the risk level of the k-th key safety indicator. k E represents the probability coefficient of the occurrence of the k-th key safety indicator. k This represents the environmental impact factor, where P is the total number of key safety indicators.

[0075] Among them, the environmental impact factor is calculated by comprehensively considering the impact of environmental factors on the safety risks of the design scheme. Environmental factors include, but are not limited to, geological factors, hydrological factors, and climate factors.

[0076] User consistency loss L4 is used to represent the loss caused by inconsistencies between the initial design and user feedback. The initial value of L4 is 0.

[0077] The first weight parameter W1 is the weight parameter for specification compliance loss; the second weight parameter W2 is the weight parameter for economic loss; the third weight parameter W3 is the weight parameter for performance loss; and the fourth weight parameter W4 is the weight parameter for user consistency loss. The first weight parameter W1, the second weight parameter W2, the third weight parameter W3, and the fourth weight parameter W4 satisfy the following expression:

[0078] W1+W2+W3+W4=1;

[0079] In the formula, W1, W2, W3, and W4 are all adjustable constants, and their values ​​are all in the range of [0, 1].

[0080] The iterative optimization unit is used to perform iterative optimization calculations on the multi-objective balance loss function by improving the Bayesian algorithm, and can output a preliminary design scheme before the numerical convergence of the multi-objective balance loss function.

[0081] The iterative optimization calculation of the multi-objective equilibrium loss function by improving the Bayesian algorithm includes:

[0082] Step 1: Set the prior probability distribution for the first, second, third, and fourth weight parameters in the multi-objective balance loss function according to the engineering scenario.

[0083] In step 1, the engineering scenarios are typical port engineering design scenarios, including soft soil foundation scenarios, highly corrosive environment scenarios, port area scenarios with strong tidal difference and large tides, liquefied soil site scenarios, and hazardous materials operation area scenarios; the prior probability distribution can guide the search starting point of the improved Bayesian algorithm to high-value design areas.

[0084] In this embodiment, when the engineering scenario is a soft soil foundation scenario, the second weight parameter W2 and the third weight parameter W3 are assigned higher initial weights; when the engineering scenario is a highly corrosive scenario, the third weight parameter W3 is assigned a higher initial weight; when the engineering scenario is a port area with strong tidal difference, the third weight parameter W3 is assigned a higher initial weight; when the engineering scenario is a liquefiable soil site scenario, the third weight parameter W3 is assigned the highest initial weight; and when the engineering scenario is a hazardous materials operation area scenario, the first weight parameter W1 and the third weight parameter W3 are assigned high initial weights.

[0085] In soft soil foundation scenarios, assigning higher initial weights to W2 and W3 can minimize project costs while meeting settlement control requirements. In highly corrosive environments, assigning a higher initial weight to W3 can effectively balance initial investment with the durability and maintenance costs of the building structure throughout its lifecycle. In tidal port areas with strong tidal differences, assigning a higher initial weight to W3 can effectively ensure the building structure can withstand strong periodic hydrodynamic loads under complex water flow conditions, ensuring its stability and durability. In liquefied soil scenarios, assigning the highest initial weight to W3 can prioritize the stability of the foundation even when it instantly loses its bearing capacity. In hazardous materials operation areas, assigning the highest initial weights to W1 and W3 ensures that the design scheme complies with design specifications, further guaranteeing safety.

[0086] Step 2: Establish a proxy model based on the combined kernel function.

[0087] In step 2, the combined kernel function can capture complex patterns between data during the iterative optimization calculation of the improved Bayesian algorithm. The expression of the combined kernel function is as follows:

[0088] K=K g ×K p ×K s ×C;

[0089] In the formula, K g As the basic core, it is used to capture the general nonlinear relationships between design parameters; K p A periodic kernel, used to handle the effects of periodic loads; K s C is the spatial kernel, used to handle spatial correlation; C is the constraint kernel, used to handle the boundary effects of rigid constraints.

[0090] Furthermore, space core K s When handling spatial correlation, physical distance can be transformed into the correlation strength between parameters, making the engineering properties of two physically close points more similar than those of two distant points. A spatial kernel K is set in this combined kernel function. s The advantage lies in the fact that when the improved Bayesian algorithm predicts the engineering attributes of an unknown location, it will refer more to the engineering attributes of its neighboring known points, thereby constructing a surrogate model that conforms to the laws of spatial continuity.

[0091] When dealing with the boundary effects of rigid constraints, constraint kernel C can introduce penalized boundaries into the surrogate model. The advantage of setting constraint kernel C in this combined kernel function is that when the combination of design parameters attempts to approach or cross the boundary of a rigid constraint, the combined kernel function will cause the values ​​of the various losses predicted by the surrogate model to increase sharply, forming a "cliff effect," thereby guiding the optimization algorithm to automatically avoid infeasible design regions.

[0092] In this embodiment, the periodic load includes tidal load and traffic load.

[0093] Spatial correlations include differences in foundation parameters at different locations, differences in wave heights along the coastline, planar distribution of pollutant concentrations within the harbor basin, wind load distribution on the surfaces of large structures, distribution of the water flow velocity field beneath the wharf, and non-uniformity of siltation thickness in the channel area.

[0094] Rigid constraints include design specifications, dimensional limitations of building structures, physical limits of building materials, and mandatory indicators of environmental regulations. In this embodiment, dimensional limitations of building structures include, but are not limited to, the maximum manufacturable size of components and the minimum width of waterways; physical limits of building materials include, but are not limited to, the yield strength of reinforcing steel; and mandatory indicators of environmental regulations include, but are not limited to, the maximum permissible turbidity of dredging projects.

[0095] Step 3: Use a surrogate model to predict and simulate the multi-objective balance loss function to obtain candidate optimal design parameters, and calculate the actual loss value of the candidate optimal design parameters.

[0096] In step 3, candidate optimal design parameters are obtained by using a surrogate model to predict and simulate the multi-objective balance loss function, and the actual loss value of the candidate optimal design parameters is calculated. This includes obtaining multiple sets of predicted loss values ​​by using a surrogate model to predict and simulate the multi-objective balance loss function. Each set of predicted loss values ​​corresponds to a set of design parameters. In the current iteration round, one or more sets of design parameters corresponding to the minimum predicted loss values ​​are selected as candidate optimal design parameters, and the actual loss value of the candidate optimal design parameters under the true function is calculated.

[0097] Step 4: Evaluate the actual loss value under the candidate optimal design parameters to obtain the evaluation result, and use the evaluation result to iteratively optimize the surrogate model until the multi-objective balance loss function converges.

[0098] The feedback adjustment unit is used to obtain user feedback information based on the preliminary design scheme, parse the user feedback information through a natural language model to obtain the feedback quantification result, and update the multi-objective balance loss function based on the feedback quantification result.

[0099] The preliminary design scheme is a phased optimization design scheme presented to users, including candidate optimal design parameters under the current iteration round, as well as the preliminary estimated performance, cost, and preliminary comprehensive loss corresponding to the candidate optimal design parameters.

[0100] User feedback information consists of text messages expressed in natural language, representing user feedback on the preliminary design scheme after review.

[0101] Furthermore, updating the multi-objective balance loss function based on the feedback quantification results includes: updating the user consistency loss of the multi-objective balance loss function using a penalty function based on the feedback quantification results, while dynamically increasing or decreasing any one or more of the first weight parameter, second weight parameter, third weight parameter, and fourth weight parameter.

[0102] The penalty function is used to quantify the deviation between design parameters and user expectations. In practical applications, the penalty function is manually set based on the actual situation. Its core idea is that the greater the deviation between the design parameters and the expected value reflected in user feedback, the greater the loss value. Based on this, the penalty function has various forms, including but not limited to exponential functions, piecewise functions, or higher-order polynomial functions.

[0103] The result output unit is used to output the target design scheme after the numerical convergence of the multi-objective balanced loss function.

[0104] When the multi-objective balance loss function converges numerically, the iterative optimization unit stops iterative optimization calculation. At this time, the result output unit can output the target design scheme, which is the final optimized design scheme shown to the user. Its composition structure is similar to that of the preliminary design scheme. The target design scheme includes the optimal design parameters, the estimated performance and cost corresponding to the optimal design parameters, and the comprehensive loss. The optimal design parameters are the candidate optimal design parameters corresponding to the minimum actual loss value obtained in the last iteration.

[0105] In a preferred embodiment, the question-and-answer parsing unit and the feedback adjustment unit are equipped with a human-computer interaction module, on which the user can input design task information; the user can also review the preliminary design scheme and input user feedback information on the human-computer interaction module.

[0106] Example 2

[0107] Based on Example 1, combined with Figure 2 This embodiment discloses an application method for a system used for question-and-answer verification and optimization of port yard design schemes, including:

[0108] S1. Use the question-answering parsing unit to obtain design task information, and use a natural language processing model to parse the design task information to obtain a structured design task.

[0109] S2. Using a multi-objective evaluation unit, set the function loss value based on the design objective, and construct a multi-objective balanced loss function based on the function loss value.

[0110] S3. The iterative optimization unit is used to perform iterative optimization calculation of the multi-objective balance loss function by improving the Bayesian algorithm, and a preliminary design scheme can be output before the numerical convergence of the multi-objective balance loss function.

[0111] S4. Use the feedback adjustment unit to obtain user feedback information based on the preliminary design scheme, and use a natural language model to parse the user feedback information to obtain the feedback quantification result. Update the multi-objective balance loss function based on the feedback quantification result.

[0112] S5. Utilize the result output unit to output the target design scheme after the numerical convergence of the multi-objective balance loss function.

[0113] Based on the above steps, as an example, the following content will further illustrate the method provided in this embodiment using the road surface design of a newly built container yard as a design task. The specific content includes:

[0114] S1. Use the question-answering parsing unit to obtain design task information, and use a natural language processing model to parse the design task information to obtain a structured design task.

[0115] In this embodiment, the design task information is input by the user through the question-and-answer parsing unit via the human-computer interaction module. The design task information is: "A road surface structure needs to be designed for a newly built container yard in a port area. This area has a typical soft soil foundation, and the designed vehicles are heavy-duty container trucks with an axle load of 50 tons. According to the 'Port Roads and Container Yard Design Code,' the foundation settlement of the container yard pavement should not exceed 0.3m within its design service life. Based on this, the owner hopes that the long-term settlement can be strictly controlled within 0.1m, while the unit cost should ideally be controlled at around 300 yuan / ㎡. Please provide the optimal design for the pavement structure layer thickness."

[0116] In the structured design task, the design objectives include: regulatory constraints: the estimated long-term settlement should not exceed 0.3m; economic objectives: the estimated direct engineering cost should be close to 300 yuan / m²; and performance objectives: the ideal value of the estimated long-term settlement should not exceed 0.1m.

[0117] The design parameter space includes: the thickness of the asphalt concrete surface layer, with a value range of [8, 15] cm; the thickness of the cement-stabilized crushed stone base layer, with a value range of [25, 40] cm; and the thickness of the graded crushed stone subbase layer, with a value range of [20, 35] cm.

[0118] S2. Using a multi-objective evaluation unit, set the function loss value based on the design objective, and construct a multi-objective balanced loss function based on the function loss value.

[0119] In this embodiment, the roles of each function loss value in the multi-objective balance loss function are as follows:

[0120] The specification compliance loss L1 can assess whether a design scheme conforms to the design specifications. When the estimated long-term settlement of a design scheme exceeds the specification limit of 300mm, the specification compliance loss L1 will be activated and reach a very large value, causing the design scheme to be immediately abandoned during the iterative optimization process.

[0121] The economic loss L2 quantifies the difference between the estimated direct engineering cost and the target direct engineering cost of 300 yuan / m². When the estimated direct engineering cost exceeds the target direct engineering cost, the economic loss L2 increases with the amount of the excess, and the process of iterative optimization seeks a lower-cost design solution.

[0122] Performance loss L3 quantifies the difference between the estimated long-term settlement and the target long-term settlement of 0.1m. When the long-term settlement value is greater than 0.1m, performance loss L3 will be activated, and the greater the long-term settlement value exceeds this value, the larger the performance loss L3 will be, thereby driving the design scheme to be optimized towards a higher safety performance standard.

[0123] S3. The iterative optimization unit is used to perform iterative optimization calculation of the multi-objective balance loss function by improving the Bayesian algorithm, and a preliminary design scheme can be output before the numerical convergence of the multi-objective balance loss function.

[0124] In this embodiment, since the scenario is explicitly defined as a soft soil foundation, weights are assigned to the loss values ​​of each function, specifically, higher initial values ​​are given to the economic loss L2 and the performance loss L3. This step reflects that when the scenario is a soft soil foundation, controlling economic costs and long-term settlement are the primary considerations.

[0125] During the iterative optimization process, this embodiment provides a preliminary design scheme for the output of a certain iteration, including:

[0126] "1. Candidate optimal design parameters: asphalt concrete surface layer thickness: 11cm; cement stabilized crushed stone base layer thickness: 32cm; graded crushed stone subbase layer thickness: 26cm."

[0127] 2. Preliminary performance and cost estimates: The estimated long-term settlement is 75mm over 30 years, which meets the design service life requirements, and the unit cost is 312 yuan / m².

[0128] 3. Preliminary total loss: 28.

[0129] The preliminary design scheme meets all design specifications, but its estimated direct construction cost is 312 yuan / m², which is slightly higher than the target direct construction cost.

[0130] It should be noted that, since the multi-objective balance loss function has not yet completed numerical convergence in the iterative optimization process, the preliminary comprehensive loss corresponding to this preliminary design scheme is only a phase result of the iterative optimization stage and does not represent the optimal solution that ultimately satisfies all design objectives and user requirements. Therefore, the preliminary comprehensive loss mentioned here has not yet incorporated user feedback information for calculation.

[0131] S4. Use the feedback adjustment unit to obtain user feedback information based on the preliminary design scheme, and use a natural language model to parse the user feedback information to obtain the feedback quantification result. Update the multi-objective balance loss function based on the feedback quantification result.

[0132] In this embodiment, the user reviews the preliminary design scheme through the human-computer interaction module, and inputs user feedback information after the review. The user feedback information is as follows:

[0133] "This scheme controls settlement very well (75mm) with sufficient safety margin, but the cost (312 yuan / m²) is slightly higher than the target. Settlement control can be relaxed appropriately to make it closer to the 100mm limit, thereby achieving a further reduction in cost."

[0134] The user feedback information is parsed using a natural language model to obtain the feedback quantification result. Based on the feedback quantification result, a penalty function is used to update the user consistency loss of the multi-objective balanced loss function, while a second weight parameter W2 is dynamically added. In this embodiment, the expression of the penalty function is:

[0135] L4 = max(0, (T-T') / T');

[0136] In the formula, T is the thickness of the base layer in the current preliminary design scheme; T' is the thickness threshold set manually based on user feedback.

[0137] S5. Utilize the result output unit to output the target design scheme after the numerical convergence of the multi-objective balance loss function.

[0138] In this embodiment, after 25 iterations of optimization, the multi-objective balance loss function converges numerically, and the output target design scheme includes:

[0139] "1. Optimal design parameters: asphalt concrete surface layer thickness: 9cm; cement stabilized crushed stone base layer thickness: 30cm; graded crushed stone subbase layer thickness: 24cm."

[0140] 2. Estimated performance and cost: Compliance with specifications: Meets the requirements (bearing capacity meets the standard, estimated settlement of 0.09m is less than the specification requirement of 0.3m); Estimated long-term settlement: 0.09m (meets the owner's requirement of 0.1m); Estimated cost: 289 yuan / m².

[0141] 3. Total loss: 35.0.

[0142] It should be noted that the preliminary design scheme and the target design scheme shown in this embodiment are only examples, and this embodiment does not limit the contents of the preliminary design scheme and the target design scheme.

[0143] It should also be noted that the application method for question-and-answer verification and optimization of port yard design schemes provided in this embodiment enables all the functions of the system for question-and-answer verification and optimization of port yard design schemes provided in Embodiment 1 to be realized.

[0144] Finally, it should be noted that in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0145] The above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and are not intended to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, 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 covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A system for question-and-answer verification and optimization of port yard design schemes, characterized in that, include: Question answering unit, multi-objective evaluation unit, iterative optimization unit, feedback adjustment unit, and result output unit; The question-answering parsing unit is used to obtain design task information and parse the design task information through a natural language processing model to obtain a structured design task. The multi-objective evaluation unit is used to set the function loss value based on the structured design task and construct a multi-objective balanced loss function based on the function loss value. The function loss values ​​include specification compliance loss, economic loss, performance loss, and user consistency loss; The expression for the multi-objective balanced loss function is: L = W1L1 + W2L2 + W3L3 + W4L4; In the formula, L is the overall loss; L1 is the specification compliance loss; L2 is the economic loss; L3 is the performance loss; L4 is the user consistency loss; W1 is the first weight parameter; W2 is the second weight parameter; W3 is the third weight parameter; and W4 is the fourth weight parameter. The specification compliance loss L1 is used to represent the specification compliance loss of the design scheme, and its expression is: ； In the formula, S i S represents the safety criticality of the i-th design specification. i The value of S is [0, 10]. i A higher value for V indicates a higher level of safety criticality; i V represents the degree to which the design scheme violates the i-th design specification. i The value of l is [0, 1]; i This represents the mutual influence factor between design specifications; N is the number of design specifications. Economic loss L2 is used to represent the economic loss of the design scheme. The economic loss mainly includes direct engineering cost loss, and its expression is: L2 = (C1 - C0) / C0; In the formula, C1 is the estimated direct engineering value, and C0 is the target direct engineering value; The performance loss L3 is used to represent the safety loss of the design scheme, and its expression is: ； In the formula, R k P represents the risk level of the k-th key safety indicator. k E represents the probability coefficient of the occurrence of the k-th key safety indicator. k This represents the environmental impact factor, where P is the total number of key safety indicators. User consistency loss L4 is used to represent the loss caused by inconsistency between the preliminary design and user feedback. The initial value of L4 is 0. The iterative optimization unit is used to perform iterative optimization calculations on the multi-objective balance loss function by improving the Bayesian algorithm, and to output a preliminary design scheme before the numerical convergence of the multi-objective balance loss function. Iterative optimization calculations of the multi-objective equilibrium loss function through improved Bayesian algorithms include: Based on the engineering scenario, set prior probability distributions for the first, second, third, and fourth weight parameters in the multi-objective balance loss function; The surrogate model is built based on the combined kernel function, the expression of which is as follows: K = Kg × Kp × Ks × C; In the formula, Kg is the basic kernel, used to capture the general nonlinear relationship between design parameters; Kp is the periodic kernel, used to handle the influence of periodic loads; Ks is the spatial kernel, used to handle spatial correlation; and C is the constraint kernel, used to handle the boundary effects of rigid constraints. Candidate optimal design parameters are obtained by predicting and simulating the multi-objective equilibrium loss function using a surrogate model, and the actual loss value of the candidate optimal design parameters is calculated. The actual loss value under the candidate optimal design parameters is evaluated to obtain the evaluation result. The surrogate model is iteratively optimized using the evaluation result until the multi-objective balance loss function converges. The feedback adjustment unit is used to obtain user feedback information based on the preliminary design scheme, and to parse the user feedback information through a natural language model to obtain the feedback quantification result. The multi-objective balance loss function is updated according to the feedback quantification result. The update of the multi-objective balance loss function according to the feedback quantification result includes: updating the user consistency loss of the multi-objective balance loss function with a penalty function according to the feedback quantification result, and dynamically increasing or decreasing any one or more of the first weight parameter, the second weight parameter, the third weight parameter, and the fourth weight parameter. The result output unit is used to output the target design scheme after the numerical convergence of the multi-objective balanced loss function.

2. The system for question-and-answer verification and optimization of port yard design schemes according to claim 1, characterized in that, The design task information is textual information of the design task expressed in natural language; the structured design task includes design objectives and design parameter space, and the design objectives include specification constraint objectives, economic objectives, and performance objectives.

3. The system for question-and-answer verification and optimization of port yard design schemes according to claim 1, characterized in that, The first weight parameter W1 is the weight parameter for specification compliance loss; the second weight parameter W2 is the weight parameter for economic loss; the third weight parameter W3 is the weight parameter for performance loss; and the fourth weight parameter W4 is the weight parameter for user consistency loss. The first weight parameter W1, the second weight parameter W2, the third weight parameter W3, and the fourth weight parameter W4 satisfy the following expression: W1+W2+W3+W4=1; In the formula, W1, W2, W3, and W4 are all adjustable constants, and their values ​​are all in the range of [0, 1].

4. The system for question-and-answer verification and optimization of port yard design schemes according to claim 1, characterized in that, The question-and-answer parsing unit and the feedback adjustment unit are equipped with human-computer interaction modules, on which users can input design task information; users can also review the preliminary design scheme and input user feedback information.

5. The application method of a system for question-and-answer verification and optimization of port yard design schemes according to any one of claims 1 to 4, characterized in that, include: S1. Use the question-answering parsing unit to obtain design task information, and use a natural language processing model to parse the design task information to obtain a structured design task; S2. Using a multi-objective evaluation unit, set the function loss value based on the design objective, and construct a multi-objective balanced loss function based on the function loss value; S3. The iterative optimization unit is used to perform iterative optimization calculation of the multi-objective balance loss function through an improved Bayesian algorithm, and a preliminary design scheme can be output before the numerical convergence of the multi-objective balance loss function. S4. Use the feedback adjustment unit to obtain user feedback information based on the preliminary design scheme, and use a natural language model to parse the user feedback information to obtain the feedback quantification result. Update the multi-objective balance loss function according to the feedback quantification result. S5. Utilize the result output unit to output the target design scheme after the numerical convergence of the multi-objective balance loss function.

Images