A curation method and system based on parameter coupling and iterative simulation evaluation optimization

Abstract

The application discloses a curation optimization method and system based on parameter coupling and iterative simulation evaluation optimization, and is used for solving the problem that a prediction model and a physical reality lack fitting in the prior art, and comprises the following steps: an audience portrait containing quantified behavior parameters and a weighted exhibit knowledge graph containing semantic correlation strength are created, and a candidate scheme containing exhibit layout and narrative sequence is generated; a portrait-simulation parameter direct mapping mechanism of an intelligent agent driven by multi-dimensional parameters of the audience portrait is adopted, and the semantic correlation strength of the graph is coupled to an optimization objective function; in each round of iteration of a multi-objective optimization algorithm, a high-fidelity intelligent agent simulation is forced to be called to evaluate a current scheme, prediction data based on physical rules are obtained, and iteration optimization is performed according to the prediction data; and the application avoids the approximation error introduced by using a proxy model by constructing a forced closed-loop architecture combining data, prediction and optimization in depth, thereby guaranteeing the reliability of the optimization result in the physical world and improving the scientificity and predictability of curation.

Description

Technical Field

[0001] This invention belongs to the field of smart museum curation technology, and relates to a curation method and system that dynamically couples and optimizes physical space simulation and content narrative logic. Background Technology

[0002] In museum curatorial work, precise exhibition space planning and rigorous logical organization are key to improving the efficiency of knowledge transmission and enhancing the visitor experience.

[0003] The curator's experience plays a decisive role in traditional curating. Due to the limitations of the curator's subjective experience, the actual operation of the exhibition often deviates from the initial design plan, such as the problem of visitor congestion in specific areas of the museum exhibition hall.

[0004] To address the shortcomings of traditional curatorial methods, some curatorial organizations have developed artificial intelligence-assisted curatorial approaches. In some existing solutions (such as application number CN202511423977.9), multimodal exhibit data is integrated to construct a comprehensive knowledge graph. Based on this graph, the correlation between each exhibit and a specific exhibition theme is quantitatively calculated. This approach combines historical user behavior data to optimize the ranking of all exhibits to be displayed. Ultimately, according to a pre-set narrative framework and the actual physical scope, a curatorial plan encompassing the exhibit display order and spatial configuration is generated. However, most existing solutions still focus on content planning, using data mining and knowledge graph technologies to build a model linking exhibits, themes, and user profiles.

[0005] The existing system architecture has technical flaws, including the system architecture of the technical solution described in application number CN202511423977.9. It generally suffers from "feedforward" and "unidirectional" technical deficiencies. In this flawed system, each independent technical module must be executed strictly in a fixed order, and the coupling between modules is weak. This results in technical decoupling between the "digital planning system" that ultimately produces the curatorial scheme and the "physical reality environment" that ultimately applies it.

[0006] This deep-seated decoupling effect directly results in shortcomings in existing technologies in the following two aspects:

[0007] The first technical challenge lies in the information attenuation during the parameterization process of audience behavior model construction. When using existing technologies to build audience behavior models, in-depth mining and analysis of user behavior data produces audience profiles with a certain degree of accuracy. However, when mapping the statistical characteristics reflected in these audience profiles into model parameters that can be directly used for spatial layout optimization, it still relies on simplified rules set manually. This transformation process reduces the dimensionality of information in the original data that can truly reflect the diversity of individual audience behaviors, converting it into discrete and deterministic parameters. This result directly leads to a significant reduction in the accuracy of subsequent exhibition planning in the stage of fitting the behavior patterns of real audience groups.

[0008] The second technical challenge lies in addressing the distortion and fragmentation of evaluation metrics during the optimization process. An effective curatorial plan must simultaneously optimize both the "smoothness of the physical experience" and the "effectiveness of the narrative." Existing technologies, when performing pre-optimization, suffer from inherent technical barriers in their evaluation models.

[0009] The first aspect is the distortion of the physical experience evaluation model.

[0010] The "exhibition physical constraints" in the technical solution of application number CN202511423977.9 are based on resource allocation using static rules, rather than on accurate prediction and simulation of the dynamic interactive behavior of the audience.

[0011] To improve computational efficiency, existing technical methods generally use static rules, surrogate models, or simplified mathematical functions to approximate key physical indicators such as pedestrian flow and audience density. In fact, surrogate models are essentially low-dimensional fittings to complex physical processes. Our reliance on models with distorted problems means that the curatorial optimization process is actually carried out in a virtual "digital twin" space that deviates from the real physical environment. The effectiveness of the curatorial solutions produced by this optimization process in real-world scenarios is questionable.

[0012] The second aspect reveals a lack of a coupling evaluation mechanism between content logic and physical experience.

[0013] Current technological solutions have demonstrated the ability to effectively optimize and refine the narrative logic of content. However, this optimization is conducted under idealized conditions without fully considering the constraints of the physical environment. The current technological architecture does not provide an effective mechanism to quantitatively assess how congestion and resource constraints in the actual physical environment can negatively impact and reduce the previously planned narrative flow and experience coherence. This problem stems from its optimization objective function, which does not quantitatively represent the interaction between physical and content metrics. This may cause the algorithm to converge to a Pareto suboptimal solution.

[0014] Such suboptimal solutions may manifest in the following ways: the physical space experience may be smooth, but the logical structure of the content is disjointed and fails to effectively convey the curatorial idea. Another example is that the narrative is coherent and engaging, but the physical experience feels cramped, which can affect the audience's immersion and comfort.

[0015] This convergence result is fundamentally at odds with the ultimate goal of the curatorial activity.

[0016] Existing technical solutions face multiple technical challenges when carrying out museum curatorial optimization work, as they are limited by a single feedforward architecture design.

[0017] The first technical challenge is the audience behavior model. In its parameterization process, the conversion between raw data and model parameters was not done well, which caused information attenuation and affected the model's ability to accurately fit real audience behavior.

[0018] The second technical challenge is the evaluation of the curatorial scheme, which faces two challenges: the physical experience evaluation model has low fidelity, and the model cannot predict complex nonlinear crowd interaction behavior; the effectiveness of the content narrative is not coupled with the evaluation of the physical space experience, which prevents the algorithm from finding the globally optimal solution.

[0019] The reason for the above technical difficulties is that the existing system has not built a mandatory closed-loop feedback architecture, which was originally intended to deeply integrate audience behavior characteristics, high-fidelity simulation of physical environment dynamics, and curatorial content optimization.

[0020] A continuous feedback and iteration mechanism can achieve this deep binding, which can bridge the decoupling between digital planning and physical reality, thereby preventing the "Pareto suboptimal solution" problem.

[0021] Therefore, there is an urgent need in this field to develop a system and methodology that enables curatorial decisions to be scientific, forward-looking, and effective, which has become a key technical challenge. Summary of the Invention

[0022] The present invention aims to solve the technical problems in the prior art, which are caused by subjective errors in parameter transmission and the reliance on distorted proxy models for optimization evaluation, resulting in biased curatorial scheme predictions, unreliable optimization results, and a final mismatch with the actual audience experience.

[0023] To address the aforementioned technical problems, this invention provides a museum curatorial optimization method based on parameter coupling and iterative simulation evaluation, the steps of which include:

[0024] Acquire historical visitor behavior data and multimodal information data of exhibits to be curated in museums;

[0025] Construct a multi-dimensional interest profile of the target audience and a weighted knowledge graph that includes the entities and semantic relationships of the exhibits;

[0026] Based on the portrait and map, generate candidate curatorial schemes that include the layout of exhibits and the narrative order;

[0027] In the virtual exhibition hall space model, the intelligent agent simulation algorithm is used to simulate the visiting behavior of different profiles of visitors under the candidate curatorial scheme, and generate quantitative behavioral prediction data.

[0028] Based on the behavioral prediction data, a multi-objective optimization algorithm is used to iteratively adjust the candidate curatorial schemes and output the optimal curatorial scheme.

[0029] To achieve the above method, this invention proposes an innovative core architecture of "ternary coupling and iterative simulation evaluation and optimization", characterized by:

[0030] 1. Image-Simulation Coupling: This coupling mechanism includes two aspects:

[0031] The quantitative distribution parameters that characterize the audience's behavioral characteristics, such as the mean and standard deviation of the moving speed μ, output from the portrait construction step, are directly passed through the program interface and used to drive the sampling of the moving behavior parameters of the agent in the simulation model.

[0032] The interest parameters that represent the audience's preferences for different exhibits, such as the multidimensional interest quantity R, output from the profile construction step, can be directly used as an input parameter in the simulation model to calculate the agent's behavior in front of the exhibits (such as the probability of staying or the time of staying).

[0033] By coupling the above two aspects, the manual interpretation and experience setting links are eliminated, and the direct mapping from the statistical features of the audience profile to the behavioral parameters of the intelligent agent is realized, thereby improving the simulation accuracy.

[0034] 2. Knowledge Graph – Optimizing Coupling: The edge weights (W) calculated in the exhibit knowledge graph, representing the strength of semantic relationships between exhibits, are used to optimize coupling. total This directly serves as the core calculation basis for the "knowledge acquisition efficiency" indicator in the multi-objective optimization algorithm, ensuring that the optimization objective is always deeply bound to the inherent logic of the curatorial content.

[0035] 3. Forced closed loop: In each iteration of the multi-objective optimization algorithm, the physical rule-based agent simulation module is forcibly and completely invoked to simulate the real behavior of all current candidate solutions in order to obtain high-fidelity fitness evaluation values, thus avoiding the use of any form of proxy model or approximate evaluation function from the architecture.

[0036] Through the aforementioned architecture, this invention constructs a technological closed loop that deeply integrates logic and physics, data and behavior, and prediction and optimization. Its beneficial effects are specifically reflected in:

[0037] 1. Improved simulation accuracy and eliminated errors caused by subjective experience: Unlike the approach of manually setting parameters in the background technology, the portrait-simulation coupling mechanism of this invention directly maps the physical movement characteristic parameters and interest preference parameters in the audience portrait to drive the movement behavior and dwelling decisions of the intelligent agent. This fundamentally ensures that the diversity of simulated intelligent agent group behavior, physical movement characteristics and content interaction decisions are mathematically consistent with the statistical characteristics of the real audience group, thereby significantly improving the accuracy of behavior prediction and solving the problem of simulation model distortion caused by manual interpretation and parameter conversion.

[0038] 2. Enhanced the correlation between optimization goals and curatorial content: Existing technologies often neglect the logical consistency of content due to an overemphasis on spatial accessibility. The optimization coupling mechanism of this invention in terms of the graph directly incorporates the semantic correlation strength between exhibits into the fitness function. It forces the optimization algorithm to weigh "physical accessibility" against "logical fluency," preventing the algorithm from converging to local optima that have reasonable spatial layouts but chaotic narrative logic, thus enabling the knowledge dissemination effectiveness of the final solution.

[0039] 3. Ensuring the reliability of optimization results in the physical world: The core of this invention lies in replacing the surrogate model commonly used in the background technology with a "forced closed-loop" mechanism. The surrogate model, in essence, performs dimensionality reduction fitting on complex physical processes, inevitably introducing and accumulating errors, and is particularly difficult to capture behaviors emerging from nonlinearities such as congestion. This invention ensures that each evaluation returns to high-fidelity physical simulation, guaranteeing that each iterative optimization is based on the most realistic actual response of the current solution in the physical world. Although this mechanism has a higher computational cost, it ensures that the performance of the final curatorial solution (such as pedestrian flow, space utilization, and other related performance) has been evaluated and verified using high-fidelity intelligent agent simulation based on physical rules.

[0040] 4. Balancing optimization accuracy and computational feasibility: Although this invention employs high-fidelity simulation with higher computational costs, by using techniques such as time scaling in specific implementations, it can effectively shorten the computation cycle while ensuring the accuracy of physical simulation, making the forced closed-loop optimization method feasible for practical engineering applications. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings will be briefly described below.

[0042] Figure 1 This is a flowchart of a method according to an embodiment of the present invention.

[0043] Figure 2 This is a schematic diagram of the core architecture of "ternary parameter coupling and iterative simulation evaluation and optimization" according to an embodiment of the present invention. The diagram concisely and intuitively illustrates the forced closed-loop architecture of deep fusion of data, prediction, and optimization according to the present invention. Wherein:

[0044] The “Data Acquisition and Model Building” module (corresponding to step a in claim 1) is responsible for building a viewer profile that includes quantitative parameters of the viewer’s physical movement characteristics, and a weighted exhibit knowledge graph that includes the semantic association strength between adjacent exhibits.

[0045] The "Candidate Scheme Generation" module (corresponding to step b of claim 1) generates candidate curatorial schemes that include exhibit layout and narrative order based on the above-mentioned portrait and map.

[0046] The "Agent Simulation Evaluation" module (corresponding to step c of claim 1) is one of the key components of this invention. It realizes multi-dimensional "portrait-simulation coupling," namely: on the one hand, it directly maps and drives the agent's movement behavior by mapping the quantitative parameters (such as the mean speed μ and standard deviation σ) representing physical movement characteristics in the audience profile; on the other hand, it uses the interest parameters representing content preferences in the audience profile as key factors to determine the agent's dwell behavior. Through the above mechanism, high-fidelity behavior prediction data (including congestion risk and space utilization) is generated.

[0047] The "Iterative Optimization" module (corresponding to step d in claim 1) forms a forced closed-loop iteration together with the "Agent Simulation Evaluation" module. Its core is that in each iteration, the "Agent Simulation Evaluation" module is forcibly invoked to obtain real-time behavioral evaluation indicators. At the same time, it realizes "graph-optimization coupling", that is, the semantic association strength between exhibits in the knowledge graph is directly incorporated into the fitness function to calculate the knowledge acquisition efficiency index. By deeply integrating the above-mentioned indicators, the iterative optimization algorithm continuously adjusts the candidate schemes and finally generates the optimal curatorial scheme.

[0048] The diagram, using a simple combination of boxes and arrows, clearly illustrates the three core innovations of this invention: "image-simulation coupling," "map-optimization coupling," and "simulation evaluation forced closed loop," solving the technical problem of decoupling digital planning and physical reality in the prior art.

[0049] Figure 3 This is an example of a heatmap of audience behavior based on intelligent agent simulation according to an embodiment of the present invention.

[0050] Figure 4This is a schematic diagram illustrating the construction principle of a navigation network in a virtual exhibition hall according to an embodiment of the present invention. The diagram shows an exhibition hall space model, obstacle areas, and a visitor-accessible navigation grid generated based on this model. Detailed Implementation

[0051] The technical solutions of the embodiments 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, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0052] like Figure 1 and Figure 2 As shown, this embodiment provides a complete curatorial optimization method.

[0053] S100: Data Acquisition and Preprocessing

[0054] The system obtains historical visitor behavior data from the museum ticketing system, WiFi location data, and guide APP logs of historical exhibitions. It uses automated data processing scripts to clean the data, extract features such as visit duration, frequency of stay, and movement trajectory points for each visitor, and imports multimodal data of exhibits for the exhibition to be curated, including text descriptions, images, and three-dimensional size data.

[0055] S200: Constructing Multidimensional Interest Profiles

[0056] The K-Means clustering algorithm, which is related to machine learning, is used to divide the historical audience into two categories: "professional research type" and "family leisure type". The core of this step is to output quantitative parameters that can be directly used for simulation, rather than qualitative labels. This embodiment uses the K-Means clustering algorithm to divide the audience, but those skilled in the art can also choose other clustering algorithms, such as DBSCAN, hierarchical clustering, etc., but this does not constitute a limitation of the present invention.

[0057] Professional research-oriented profile: Output moving speed mean μ=0.5m / s, standard deviation σ=0.1m / s, corresponding speed distribution N(0.5,0.01).

[0058] Family and leisure type profile: Output moving speed mean μ=0.8m / s, standard deviation σ=0.15m / s, corresponding speed distribution N(0.8,0.0225).

[0059] To measure the degree of preference of different profiles for the content of exhibits, multidimensional interest parameters should be constructed. The exhibits to be curated should be classified in advance, such as "historical artifacts", "interactive installations", "art paintings" and so on. By analyzing the average dwell time of different profiles in front of various types of exhibits in historical data and performing normalization, a multidimensional interest vector R is constructed for each profile. The value represented by each dimension of this vector represents the degree of preference of the profile for the corresponding exhibit category. This value will be used to calculate the interest matching degree between the agent and specific exhibits in subsequent simulations.

[0060] S300: Constructing a weighted knowledge graph of exhibits

[0061] We used Natural Language Processing (NLP) techniques to uncover semantic relationships between exhibits.

[0062] Implicit association (W) implicit ): A pre-trained natural language processing model is used to map the text descriptions of each exhibit into high-dimensional semantic vectors, and the cosine similarity between the vectors is calculated.

[0063] Explicit association (W) explicit A matrix of association rules is defined by the curatorial experts. For example, the following rules could be defined:

[0064] Relationship Dimension Matching conditions Score era Belonging to the same dynasty 1.0 Material Made of the same material (e.g., both are bronze) 0.8 theme Belonging to the same theme series 1.2 process The craft has a lineage. 0.9

[0065] Based on the above rules, any two pairs of exhibits are scored to obtain explicit association weights.

[0066] Final weights: using a weighted fusion formula Calculate the weights of each edge in the graph.

[0067] S400: Virtual Exhibition Hall Construction and Path Mapping

[0068] Using a 3D modeling and simulation engine, the CAD drawings of the exhibition hall are imported and scaled to a specified ratio. Walls, columns, large display cases, etc., are marked as static obstacles. Figure 4 As shown, the radius of the agent is set to 0.3m, and a navigation mesh for path planning is obtained through baking. To balance the efficiency and realism of the simulation, the time step of the engine's physical update can be set to 0.02 seconds, and the time scaling ratio can be set to 10 to achieve a simulation speed of 10 times. This means that the speed at which time passes in the simulation world is 10 times that of the real world. Without changing the agent's behavioral logic and physical interaction rules, large-scale iterative calculations can be completed within an acceptable time.

[0069] S500: Agent-based simulation evaluation module

[0070] This is one of the core steps of this invention, achieved by using a programming language to call the intelligent agent application programming interface (API) provided by the 3D simulation engine. This module directly maps the multi-dimensional parameters in the audience profile to the intelligent agent's behavior, realizing deep coupling between the profile and the simulation model, specifically manifested in two aspects: movement behavior and dwelling behavior.

[0071] 1. Direct injection of movement behavior parameters: For each generated agent, based on the profile type assigned to it, a velocity value is sampled from the corresponding normal distribution (such as N(0.5,0.01)) and directly assigned to its velocity attribute. This step completely eliminates the step of converting human experience, ensuring a high degree of consistency between the simulated behavior and the profile settings.

[0072] 2. Dwelling Behavior Simulation: During each simulation step, the probability of the agent dwelling in front of the exhibit is calculated based on a multi-factor weighted decision model. Its general calculation formula is:

[0073]

[0074] in:

[0075] σ is the Sigmoid activation function or other functions that can map linear weights to probability values;

[0076] F i Let i be the i-th feature factor influencing the stay decision;

[0077] W i For the i-th characteristic factor F i The corresponding weighting coefficients;

[0078] n represents the total number of characteristic factors influencing the stay decision.

[0079] In a preferred embodiment of the present invention, the set of feature factors includes at least:

[0080] Profile Interest Matching Factor (R): Measures the matching degree between the agent profile and the exhibit content;

[0081] Intrinsic appeal factor (T) of exhibits: such as the recommendation level or historical importance of exhibits set by curatorial experts;

[0082] Environmental impact factors (S): such as real-time pedestrian density or space congestion at the location of the exhibits.

[0083] Based on the above three core factors, the formula for the probability of stay can be specified as follows:

[0084]

[0085] Among them, WR W T W S , where are weighting coefficients, corresponding to the weights of profile interest matching degree R, exhibit attractiveness T, and environmental crowding degree S, respectively.

[0086] In practical implementation, to ensure the objectivity of each weighting coefficient and the reproducibility of the system as a whole, the determination of the weighting coefficients includes the following specific implementation methods:

[0087] Prior settings for the system's cold start phase: The weighting coefficients can be initially set by senior curatorial experts based on the characteristics of specific exhibitions (such as art exhibitions emphasizing the attractiveness of exhibits, and science exhibitions emphasizing interest matching).

[0088] Data-driven model training and extraction: To achieve accurate dynamic correction, machine learning algorithms are used to extract weights in a model-like manner. The specific steps are as follows:

[0089] (1) Data collection and label definition: Collect past exhibition data of real visitors, define the behavior of visitors staying in front of an exhibit for more than a preset time threshold (such as 5 seconds) as positive samples, and set the label Y to 1 (staying); define the behavior of not exceeding the time limit or passing through directly as negative samples, and set the label Y to 0 (not staying).

[0090] (2) Feature extraction and model construction: Construct a logistic regression model. For each collected behavioral sample, the quantitative values ​​of its corresponding on-site “profile interest matching degree R, exhibit attractiveness T, and environmental crowding degree S” are used as the input feature vector X=(R,T,S) of the model.

[0091] (3) Model training and weight extraction: The logistic regression model is iteratively trained using the sample set constructed above to minimize the cross-entropy loss between the predicted output and the true label Y. After the model training converges, the learned regression coefficients corresponding to the input features R, T, and S in the logistic regression model are extracted.

[0092] (4) Normalization: The three extracted regression coefficients are normalized so that their sum is 1. The normalized values ​​are then used as the weight coefficients W in the simulation model. R W T W S .

[0093] By combining the aforementioned multidimensional feature construction with binary classification prediction, the abstract "finding the optimal weights" is transformed into a technically clear and reproducible logistic regression coefficient extraction process, effectively eliminating subjective biases imposed by manual settings.

[0094] 3. Data Statistics: During the simulation, real-time statistics and recording of data such as pedestrian density in each area, average dwell time on each exhibit, and path congestion index are generated, ultimately producing data such as... Figure 3 The system displays a behavior heatmap and outputs quantified behavior prediction data for further use.

[0095] S600: Closed-loop optimization

[0096] A genetic algorithm is used to optimize the curatorial scheme. This embodiment uses a genetic algorithm as an example of a multi-objective optimization algorithm, but the invention is not limited to this. Other multi-objective optimization algorithms, such as particle swarm optimization and simulated annealing, are also applicable to the framework of this invention.

[0097] The iterative operation of the genetic algorithm. In each generation, the algorithm performs the following core operations:

[0098] Selection: Calculate the fitness score of each individual based on the fitness function F described in this section, and use common methods such as roulette wheel selection or tournament selection to prioritize individuals with high fitness for the next generation.

[0099] Crossover: To generate new candidate solutions, the selected parent individuals are paired. Since each individual uses a combination of "exhibit spatial coordinates" and "visit sequence" encoding, the following clear, segmented processing flow is adopted when performing the crossover operation: The encoding of the two parent individuals is logically split to obtain the spatial coordinate part and the visit sequence part respectively; simulated binary crossover is performed on the spatial coordinate parts of the two individuals to generate two new child coordinate parts; sequential crossover or partial matching crossover is performed on the visit sequence parts of the two individuals to generate two new child sequence parts; the newly generated coordinate parts and sequence parts are recombined to form two structurally complete child individuals. This process ensures that operations for different data types are correctly applied and that the legality of the child individuals is maintained.

[0100] Mutation: To enhance population diversity and prevent the algorithm from getting stuck in local optima, newly generated individuals are mutated with a relatively small probability (e.g., 0.05). The mutation operation also uses a similar partial processing logic, acting independently on different parts of the encoding: For the spatial coordinate part, polynomial-related mutation can be used, randomly selecting the coordinate values ​​of one or more exhibits and making a small random perturbation within their value range; For the part related to the visit sequence, exchange mutation is used, that is, randomly picking two positions in the sequence and exchanging the exhibits represented by these two positions. This strategy of independent mutation of each part can effectively explore the solution space without destroying the internal structure of each part of the encoding.

[0101] Fitness Function: In each iteration of the multi-objective optimization algorithm proposed in this invention, the evaluation result is obtained by simulating the current candidate curatorial scheme using the agent simulation algorithm described in step c). Based on the quantitative behavior prediction data generated by the simulation, and combined with the weighted exhibit knowledge graph constructed in step a), the system performs a comprehensive fitness evaluation of the current candidate curatorial scheme. For each candidate curatorial scheme, its comprehensive fitness score F is calculated according to the following formula:

[0102]

[0103] in:

[0104] E knowledge The knowledge acquisition efficiency index quantifies the effectiveness and coherence of knowledge acquisition by visitors under a specific curatorial scheme. Its purpose is to address the problem of "lack of content logic evaluation" in existing technologies. The calculation method involves, according to the exhibit viewing sequence planned by the current candidate curatorial scheme, accumulating the semantic association strength (W) of adjacent exhibits along the viewing path on the weighted exhibit knowledge graph constructed in step a), by adding the semantic association strength (W) of adjacent exhibits along the viewing path. total Calculations show that a higher E knowledge The value indicates that the narrative logic of the curatorial scheme is smoother and the knowledge delivery efficiency is higher.

[0105] C congestion The congestion probability risk index aims to address the shortcomings of existing technologies in "distorted physical experience assessment." Its calculation process is as follows: Several key monitoring areas are pre-defined in the virtual exhibition hall space model, such as entrances and exits, narrow passages, or areas surrounding core exhibits. After completing the agent simulation in step c), the system will count the highest number of agents simultaneously existing in each monitoring area during the entire simulation period. The peak values ​​of each monitoring area are weighted and summed to calculate a comprehensive congestion evaluation value.

[0106] The higher this value, the greater the potential risk of congestion. In the fitness function, this term has a negative contribution, guiding the optimization algorithm to find layout schemes that can effectively manage pedestrian flow.

[0107] U space The space utilization rate index is used to help optimize the spatial layout and prevent uneven distribution of space. The calculation process is as follows: the ground accessible to visitors in the entire exhibition hall is divided into a uniform grid. After the simulation, the system will count the number of grid cells crossed by at least one agent's trajectory. The ratio of this number to the total number of grid cells is used as the space utilization rate index. A higher utilization rate index means that visitors have effectively accessed and utilized most of the exhibition hall space, and the layout is more balanced and reasonable.

[0108] The weighting coefficients W1, W2, and W3 are used to balance the importance of different optimization objectives, and their determination method aims to ensure that the optimization direction is consistent with the actual needs of curatorial work.

[0109] In one implementation approach, curatorial experts can manually configure the weights based on the core objectives of the exhibition. For exhibitions that prioritize the transmission of knowledge, the weight of the knowledge acquisition efficiency indicator W1 can be increased; for popular exhibitions that prioritize ensuring a smooth visitor experience, the weight of the congestion risk indicator W2 needs to be increased.

[0110] In another preferred embodiment, in order to find the optimal balance, multiple sets of optimization tests with different weight configurations can be carried out to form a set containing multiple optimal trade-off solutions, and the weight combination with the best overall performance can be selected as the final configuration result.

[0111] Iteration and Output: The population size is set to 100, crossover probability to 0.8, mutation probability to 0.05, and iterations are performed for 200 generations. After multiple rounds of "generation-simulation-evaluation-selection" cycles, the optimal curation scheme that maximizes the fitness function F is finally output.

[0112] Through the above embodiments, the present invention constructs a complete technical closed loop from data input to scheme output, which includes direct parameter coupling and iterative simulation evaluation and optimization, and can scientifically and accurately assist museum curatorial work.

Claims

1. A museum curatorial optimization method based on parameter coupling and iterative simulation evaluation, comprising the following steps: a) Obtain historical audience behavior data and information on exhibits to be curated, and based on the data and information, construct a multi-dimensional audience interest profile and a weighted exhibit knowledge graph, wherein the profile contains at least two types of quantitative parameters: One type is behavioral parameters that characterize the physical movement of the audience; Another type is interest parameters that characterize audience preferences for different exhibits; b) Based on the multidimensional interest profile of the audience and the weighted exhibit knowledge graph, generate at least one candidate curatorial scheme that includes exhibit layout and narrative order; Its characteristic is that it further includes the following steps: c) In the virtual exhibition hall space model, an intelligent agent simulation algorithm is used to simulate the visiting behavior of visitors with different profiles under the candidate curatorial schemes; wherein, the simulation is achieved through at least two parameter coupling mechanisms: The first coupling mechanism is to use the quantitative distribution parameters representing the physical movement characteristics of the audience in the multi-dimensional interest profile of the audience constructed in step a) as the basis for sampling and assigning values ​​to the movement behavior parameters of the intelligent agent. The second coupling mechanism is to use the interest parameters representing the audience's preference for different exhibits in the multi-dimensional interest profile of the audience constructed in step a) as an input parameter to calculate the probability or duration of the agent staying in front of the exhibits. The above coupling mechanism can generate quantified behavioral prediction data. d) The candidate curatorial schemes are iteratively simulated, evaluated, and optimized using a "multi-objective optimization algorithm" to generate the optimal curatorial scheme; wherein, each iteration of the "multi-objective optimization algorithm" forcibly calls the intelligent agent simulation algorithm described in step c) to simulate the current candidate curatorial scheme, and optimizes it based on the evaluation results obtained from the simulation, wherein the calculation of the evaluation results is based on: The quantified behavioral prediction data generated in step c) is used to calculate multidimensional behavioral assessment indicators, including congestion risk indicators and space utilization indicators, wherein: The congestion risk index is calculated by statistically analyzing the peak number of intelligent agents simultaneously existing in the preset monitoring area, and performing a weighted summation of the peak values ​​to quantify the potential crowding level in the exhibition hall. The index value is positively correlated with the predicted congestion risk. The space utilization rate index is calculated by dividing the accessible ground in the exhibition hall into grid cells and counting the proportion of grid cells crossed by at least one intelligent agent's trajectory to the total number of grid cells. This is used to quantify the balance of the audience's use of the exhibition hall space, and the index value is positively correlated with the effective utilization rate of the exhibition hall space. The weighted exhibit knowledge graph constructed in step a) is used to calculate the knowledge acquisition efficiency index related to the narrative order of the candidate curatorial scheme. This index is obtained by summing the semantic association strength of all adjacent exhibit pairs in the visit sequence of the candidate curatorial scheme.

2. The method according to claim 1, characterized in that, In step a), the specific process of constructing the multidimensional interest profile of the audience is as follows: perform statistical analysis on the historical audience behavior data to obtain the mean (μ) and standard deviation (σ) of the movement speed of different audience groups, and use the mean (μ) and standard deviation (σ) as the quantitative behavior parameters.

3. The method according to claim 2, characterized in that, In step c), the movement speed of the agent is assigned a value based on the mean (μ) and standard deviation (σ) of the movement speed by random sampling to simulate the group distribution of the audience's movement speed.

4. The method according to claim 1, characterized in that, In step a), the semantic association strength is calculated using the following formula: in: W implicit These are implicit semantic association weights calculated based on natural language processing techniques. W explicit For explicit association weights defined based on expert rules; W1 and W2 are preset weighting coefficients to satisfy specific conditions (such as W1+W2=1).

5. The method according to claim 1, characterized in that... The candidate curatorial schemes mentioned in step b) are represented by a combination of encoding of "exhibit spatial coordinates" and "visitor sequence". The iterative optimization referred to in step d) optimizes both "exhibit spatial coordinates" and "visitor sequence".

6. The method according to claim 1, characterized in that, The "multi-objective optimization algorithm" mentioned in step d) includes, but is not limited to, genetic algorithm, particle swarm optimization algorithm and simulated annealing algorithm; and the optimization is based on a comprehensive fitness function, which is at least a weighted combination of knowledge acquisition efficiency index, space utilization index and congestion risk index, wherein the weights of knowledge acquisition efficiency index and space utilization index are positive and the weight of congestion risk index is negative.

7. A museum curatorial optimization system, characterized in that, include: Memory, used to store computer programs; A processor connected to the memory; wherein, when the processor executes the computer program, it implements the method of any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method of any one of claims 1 to 6.

Images