Industrial park innovation atmosphere evaluation method and system
By constructing a multi-level evaluation system and a core MLLM model, combined with expert scoring and role simulation, the problem of insufficient subjectivity and comprehensiveness in the evaluation of innovation atmosphere in existing technologies has been solved, and efficient and accurate evaluation of innovation atmosphere in parks and planning support have been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN URBAN PLANNING & DESIGN INST
- Filing Date
- 2025-12-16
- Publication Date
- 2026-04-14
AI Technical Summary
Existing methods for assessing the innovation atmosphere in industrial parks suffer from difficulties in capturing subjective perceptions, insufficient dynamic monitoring, inadequate semantic correlation, and challenges in large-scale application, resulting in insufficient scientific rigor and comprehensiveness in the assessment.
A multi-level evaluation system for the innovation atmosphere of industrial parks is constructed, including data preprocessing, construction of a multi-level evaluation system, determination of the core MLLM evaluation model, and construction of an automated evaluation framework. The core model is selected through expert scoring and multi-dimensional model verification, and quantitative evaluation is carried out by combining role simulation and entropy weight method.
It enables a comprehensive and accurate assessment of the innovation atmosphere in industrial parks, improves the scientific nature and efficiency of the assessment, provides precise decision support for park planning, and meets the actual needs of different core stakeholders.
Smart Images
Figure CN121860199A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the interdisciplinary field of artificial intelligence and environmental perception, and more specifically, to a method and system for evaluating the innovation atmosphere of industrial parks. Background Technology
[0002] Industrial parks, as core spaces for regional economic development and technological innovation, directly determine their attractiveness to innovative talent, enterprise R&D efficiency, and overall innovation vitality based on their innovation atmosphere. Therefore, establishing a scientific evaluation system is crucial for park development. However, existing methods for evaluating the innovation atmosphere of industrial parks still have several limitations, such as:
[0003] (1) Traditional assessment methods rely heavily on manual questionnaire surveys, field interviews or on-site inspections. They not only have the problems of strong subjectivity and high manpower and time costs, but are also limited by the sample size, making it difficult to capture subtle differences in small and medium-scale spaces, and even more difficult to achieve dynamic monitoring and real-time updates of the park atmosphere.
[0004] (2) Although 3S technologies (GIS, RS, GPS) can improve the efficiency of spatial data acquisition, this technology can only process objective geographic information and cannot analyze subjective perception indicators such as "spatial enclosure" and "environmental comfort".
[0005] (3) Although deep learning technology can extract physical features (such as building density and green area) from images, it cannot establish semantic relationships between "physical features - subjective perception - innovation effectiveness", making it difficult to support the transformation of evaluation results into actual planning.
[0006] (4) Although biosensors (such as skin conductance sensors and EEG devices) can directly capture human sensory feedback, they are difficult to apply on a large scale in daily park scenarios due to their invasiveness.
[0007] Therefore, in order to overcome the limitations of existing industrial park innovation atmosphere assessment methods in terms of subjective perception capture, dynamic monitoring, semantic association construction, and large-scale application, and to comprehensively improve the scientific nature and comprehensiveness of the assessment, it is necessary to study an industrial park innovation atmosphere assessment tool that combines low cost, automation, and multi-subject perception capture capabilities, so as to provide accurate and effective assessment support and decision-making basis for the high-quality development of industrial parks. Summary of the Invention
[0008] The technical problem to be solved by the present invention is to provide a method and system for evaluating the innovation atmosphere of industrial parks, addressing the shortcomings of the prior art.
[0009] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for evaluating the innovation atmosphere of an industrial park, comprising the following steps:
[0010] S1. Perform data preprocessing on the acquired initial historical real-scene image set to obtain the target historical real-scene image set;
[0011] S2. Construct a multi-level industrial park innovation atmosphere evaluation system that includes a target layer, a criterion layer, and an indicator layer. The target layer is used to clarify the overall goal of industrial park innovation atmosphere evaluation. The criterion layer is used to decompose the overall goal clarified in the target layer into two intermediate dimensions: "intuitive perception" and "element characteristics". The indicator layer is used to determine specific indicators for each dimension of the criterion layer.
[0012] S3. Based on the multi-level industrial park innovation atmosphere evaluation system, score the target historical real-scene image set to determine the evaluation benchmark, and determine the core MLLM evaluation model according to the evaluation benchmark;
[0013] S4. Based on the core MLLM evaluation model, an automated innovation atmosphere evaluation framework is built by organically integrating the core driving entity identity definition, evaluation perspective and corresponding priority evaluation indicators.
[0014] Furthermore, in step S1, the data preprocessing of the acquired initial historical real-scene image set to obtain the target historical real-scene image set includes:
[0015] S11. Remove interfering images containing irrelevant elements and / or blurry images with substandard image quality from the initial set of historical real-scene images to obtain the core set of historical real-scene images.
[0016] S12. Denoise and tilt angle correction are performed on each image in the core historical real-scene image set to obtain an optimized historical real-scene image set;
[0017] S13. Convert each image in the optimized historical real-scene image set into a unified image format to obtain the target historical real-scene image set.
[0018] Furthermore, in step S2, the overall objectives of the industrial park innovation atmosphere assessment include improving focus, regulating negative emotions, and enhancing mental activity.
[0019] Furthermore, in step S2, specific indicators determined based on the intermediate dimension of intuitive feeling, for the overall goal of improving focus, include tranquility, comfort, and cleanliness;
[0020] To achieve the overall goal of improving focus, specific indicators determined based on the intermediate dimension of element characteristics include noise level, facility quality, greening rate, and enclosure degree.
[0021] For the overall goal of regulating negative emotions, specific indicators determined based on the intermediate dimension of intuitive feelings include tranquility, comfort, safety, beauty, and relaxation;
[0022] To address the overall goal of regulating negative emotions, specific indicators determined based on the intermediate dimension of element characteristics include noise levels, greening rate, color matching, and sense of belonging to the environment.
[0023] With the overall goal of improving mental agility, specific indicators determined based on the intermediate dimension of intuitive feeling include ease, novelty, and rich color.
[0024] To achieve the overall goal of improving the activity of thinking, specific indicators determined based on the intermediate dimension of element characteristics include greening rate, color matching, scene diversity, and field of vision.
[0025] Furthermore, in step S3, the step of scoring the target historical real-scene image set according to the multi-level industrial park innovation atmosphere evaluation system to determine the evaluation benchmark, and determining the core MLLM evaluation model according to the evaluation benchmark, includes:
[0026] S31. Using an expert scoring model, the innovation atmosphere is scored based on the target historical real-scene image set according to the multi-level industrial park innovation atmosphere evaluation system, so as to determine the evaluation benchmark that can measure the reliability of the model evaluation results.
[0027] S32. Select multiple candidate MLLM models and unify the input content and temperature parameter of each candidate MLLM model.
[0028] S33. After inputting the target historical real-scene image set and the corresponding benchmark prompt words into each candidate MLLM model, the predicted value output by each candidate MLLM model and the corresponding evaluation benchmark are respectively substituted into the multi-dimensional model verification index system to calculate the evaluation index values, so as to obtain the specific quantitative results of each candidate MLLM model under the preset verification index.
[0029] S34. Compare the quantitative values of each candidate MLLM model under the same validation metrics, and select the candidate MLLM model that can show comprehensive advantages under each validation metric as the core MLLM evaluation model.
[0030] Furthermore, in step S33, the multi-dimensional model validation index system includes the mean absolute error used to measure the average deviation between the model's predicted values and the expert scores, the Pearson correlation coefficient used to measure the linear correlation between the model's predicted values and the expert scores, the maximum error used to measure the maximum difference between the model's predicted values and the expert scores in all samples, and the coefficient of determination used to measure the model's ability to explain the variance of the expert scores.
[0031] Furthermore, in step S4, the construction of an automated innovation atmosphere evaluation framework based on the core MLLM evaluation model, according to the organic integration of the core driving entity identity definition, evaluation perspective, and corresponding priority evaluation indicators, includes:
[0032] S41. Key participants in innovation activities in industrial parks should be the core driving force.
[0033] S42. For each core driving entity, create a corresponding role profile based on the core needs it focuses on in the park's innovation activities;
[0034] S43. Determine the priority evaluation indicators corresponding to each role profile from the indicator layer of the multi-level industrial park innovation atmosphere evaluation system, and set structured prompt words for each type of core driving entity based on the definition of the driving entity's identity, evaluation perspective and corresponding priority evaluation indicator description, scoring requirements and scoring basis.
[0035] S44. For each type of core driving entity, construct an automated innovation climate assessment framework as follows:
[0036] (1) The set of real-world images to be evaluated and the corresponding structured prompts are used as inputs to the core MLLM evaluation model;
[0037] (2) The core MLLM evaluation model calculates the sub-scores corresponding to each specific indicator in the indicator layer based on the input data;
[0038] (3) The core MLLM evaluation model uses the entropy weight method to assign weights to the scores of each item according to the difference in the degree of attention to demand and sum them up, and outputs the total score of the park's innovation atmosphere.
[0039] Furthermore, after step S4, the method further includes:
[0040] S5. During the actual evaluation, the total innovation atmosphere score corresponding to different core driving entities output by the automated innovation atmosphere evaluation framework is associated with the spatial coordinates of the industrial park, and an innovation atmosphere spatial distribution map is generated accordingly.
[0041] Secondly, this application discloses an industrial park innovation atmosphere evaluation system, which includes a data preprocessing module, an innovation atmosphere evaluation system construction module, an MLLM evaluation model screening module, and an automated evaluation framework construction module, wherein:
[0042] The data preprocessing module is used to preprocess the acquired initial historical real-scene image set to obtain the target historical real-scene image set.
[0043] The innovation atmosphere assessment system construction module is used to construct a multi-level industrial park innovation atmosphere assessment system that includes a target layer, a criterion layer, and an indicator layer. The target layer is used to clarify the overall goal of the industrial park innovation atmosphere assessment. The criterion layer is used to break down the overall goal clarified in the target layer into two intermediate dimensions: "intuitive perception" and "element characteristics". The indicator layer is used to determine specific indicators for each dimension of the criterion layer.
[0044] The MLLM evaluation model screening module is used to score the target historical real-scene image set according to the multi-level industrial park innovation atmosphere evaluation system to determine the evaluation benchmark, and to determine the core MLLM evaluation model according to the evaluation benchmark.
[0045] The automated evaluation framework building module is used to build an automated innovation atmosphere evaluation framework based on the core MLLM evaluation model, according to the organic integration of the core driving subject identity definition, evaluation perspective and corresponding priority evaluation indicators.
[0046] Thirdly, this application discloses a readable storage medium, which includes a method program for evaluating the innovation atmosphere of an industrial park. When the method program is executed by a processor, it implements the steps of the method described in any of the preceding claims.
[0047] The beneficial effects of this invention are:
[0048] (1) In the constructed multi-level industrial park innovation atmosphere evaluation system, the target layer clarifies the overall direction of the evaluation, the criteria layer breaks down the overall target into two intermediate dimensions: "intuitive perception" and "element characteristics", and the indicator layer further refines the specific indicators. This hierarchical design makes the evaluation system logically clear and hierarchical, and can comprehensively and systematically cover all aspects of the industrial park innovation atmosphere, ensuring the comprehensiveness and accuracy of the evaluation;
[0049] (2) By evaluating and comparing multiple candidate MLLM models according to the evaluation benchmark, the core MLLM evaluation model with the best overall performance under various validation metrics can be selected from among many models. This model can more accurately understand and process the information in the image set, providing strong technical support for subsequent automated evaluation and improving the accuracy and efficiency of the evaluation;
[0050] (3) Through role simulation mechanism, the perception differences of core subjects are quantified. For example, users focus on micro-space quality, while service providers focus on macro value, which solves the defect of the traditional assessment of "one-size-fits-all" and makes the assessment results more in line with actual needs.
[0051] (4) It can provide “full-cycle” decision-making tools for park planning, stock renewal and regional coordination, help build a high-quality innovation ecosystem that “adapts to spatial needs” and enhance the park’s ability to support innovation activities. Attached Figure Description
[0052] Figure 1 This is a flowchart illustrating a method for evaluating the innovation atmosphere of an industrial park, as disclosed in this invention.
[0053] Figure 2 This is a schematic diagram of the structure of an industrial park innovation atmosphere assessment system disclosed in this invention;
[0054] Figure 3 This is a schematic diagram of the structure of a readable storage medium disclosed in this invention. Detailed Implementation
[0055] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.
[0056] like Figure 1 As shown, this application discloses a method for evaluating the innovation atmosphere of an industrial park, which includes the following steps:
[0057] Step S1: Perform data preprocessing on the acquired initial historical real-scene image set to obtain the target historical real-scene image set.
[0058] Step S2: Construct a multi-level industrial park innovation atmosphere evaluation system that includes a target layer, a criterion layer, and an indicator layer. The target layer is used to clarify the overall goal of industrial park innovation atmosphere evaluation. The criterion layer is used to break down the overall goal clarified in the target layer into two intermediate dimensions: "intuitive perception" and "element characteristics". The indicator layer is used to determine specific indicators for each dimension of the criterion layer.
[0059] Step S3: Based on the multi-level industrial park innovation atmosphere evaluation system, score the target historical real-scene image set to determine the evaluation benchmark, and determine the core MLLM evaluation model according to the evaluation benchmark.
[0060] Step S4: Based on the core MLLM evaluation model, an automated innovation atmosphere evaluation framework is built by organically integrating the core driving entity identity definition, evaluation perspective and corresponding priority evaluation indicators.
[0061] As can be seen from the above, the industrial park innovation atmosphere evaluation method disclosed in this application constructs a multi-level industrial park innovation atmosphere evaluation system. The objective layer clarifies the overall direction of the evaluation, the criterion layer breaks down the overall objective into two intermediate dimensions: "intuitive perception" and "element characteristics," and the indicator layer further refines specific indicators. This hierarchical design makes the evaluation system logically clear and hierarchically distinct, comprehensively and systematically covering all aspects of the industrial park innovation atmosphere, ensuring the comprehensiveness and accuracy of the evaluation. By evaluating and comparing multiple candidate MLLM models according to the evaluation benchmark, the core MLLM evaluation model with the best overall performance under various verification indicators can be selected from numerous models. This model can more accurately understand and process information in image sets, providing strong technical support for subsequent automated assessments and improving the accuracy and efficiency of assessments. Through role simulation mechanisms, it quantifies the perceptual differences of core subjects. For example, users focus on micro-level spatial quality, while service providers focus on macro-level value, solving the shortcomings of traditional "one-size-fits-all" assessments and making assessment results more aligned with actual needs. It can provide "full-cycle" decision-making tools for park planning, stock renewal, and regional coordination, helping to build a high-quality innovation ecosystem that "adapts to spatial needs" and enhancing the park's ability to support innovation activities.
[0062] In one embodiment, step S1, which involves preprocessing the acquired initial historical real-scene image set to obtain the target historical real-scene image set, includes:
[0063] Step S11: Remove interfering images containing irrelevant elements and / or blurry images with substandard image quality from the acquired initial historical real-scene image set to obtain the core historical real-scene image set.
[0064] Specifically, this application combines manual review with machine screening to fully leverage the efficiency and speed of machine screening, enabling a preliminary screening of a large number of initial historical real-world images in a short period of time to filter out images that clearly do not meet the requirements; at the same time, it utilizes the accuracy and flexibility of manual review to conduct a detailed review of the machine screening results, avoiding misjudgments due to the limitations of machine algorithms.
[0065] Step S12: Denoise and tilt angle correction are performed on each image in the core historical real-scene image set to obtain an optimized historical real-scene image set.
[0066] Specifically, this application employs Gaussian filtering and Hough transform to denoise and correct the tilt angle of each image in the core historical real-scene image set. The Gaussian filtering method aims to remove Gaussian noise (such as light interference during shooting and noise from equipment sensors) from the image while preserving key spatial structure information such as building facade lines and street boundaries. During the tilt angle correction process, the dominant straight lines in the image (such as building eaves lines and street edges) are detected by Hough transform, and the tilt angle is calculated based on the slope distribution of these dominant straight lines. The image is then rotated according to the calculated tilt angle to ensure the accurate representation of spatial structures (such as the layout relationship between the R&D area and the public area).
[0067] Step S13: Convert each image in the optimized historical real-scene image set into a unified image format to obtain the target historical real-scene image set.
[0068] Specifically, this application will convert each image in the optimized historical real-scene image set into a unified JPEG image format and compress the images to less than 2MB to adapt to the model input efficiency.
[0069] In one embodiment, in step S2, the overall objectives of the industrial park innovation atmosphere assessment include improving focus, regulating negative emotions, and increasing mental activity.
[0070] In one embodiment, in step S2, for the overall goal of improving concentration, specific indicators determined based on the intermediate dimension of intuitive feeling include tranquility, comfort, and cleanliness; for the overall goal of improving concentration, specific indicators determined based on the intermediate dimension of element characteristics include noise level, facility quality, greening rate, and enclosure degree; for the overall goal of regulating negative emotions, specific indicators determined based on the intermediate dimension of intuitive feeling include tranquility, comfort, safety, beauty, and relaxation; for the overall goal of regulating negative emotions, specific indicators determined based on the intermediate dimension of element characteristics include noise level, greening rate, color matching, and sense of belonging to the environment; for the overall goal of improving mental activity, specific indicators determined based on the intermediate dimension of intuitive feeling include relaxation, novelty, and rich colors; for the overall goal of improving mental activity, specific indicators determined based on the intermediate dimension of element characteristics include greening rate, color matching, scene diversity, and field of vision.
[0071] Specifically, the target layer (S1 level) is the top-level evaluation target, which focuses on the three core psychological mechanisms by which the spatial environment affects innovation effectiveness: enhancing focus, regulating negative emotions, and improving mental activity. The criteria layer (S2 level) is the intermediate dimension. This level breaks down the S1 level targets into two operational dimensions: "intuitive feelings" and "element characteristics." Intuitive feelings reflect the user's subjective experience of the space, while element characteristics correspond to the objective physical attributes of the space. The indicator layer (S3 level) consists of specific indicators. These are selected through a structured questionnaire survey (recipients include enterprise R&D personnel, research institution employees, and park operation managers, with an effective response rate of no less than 80%) to identify specific indicators significantly related to the S1 level mechanism. Ultimately, eight intuitive perception indicators (P1: tranquil, P2: comfortable, P3: clean, P4: safe, P5: beautiful, P6: relaxing, P7: novel, P8: colorful) and eight element characteristic indicators (C1: noise level, C2: facility quality, C3: greening rate, C4: enclosure degree, C5: color matching, C6: sense of environmental belonging, C7: scene diversity, C8: openness of view) are determined. By clearly defining the connotation of each indicator, an evaluation indicator library that can be directly converted into model prompts is formed. Among them:
[0072] P1 Quiet: No obvious sources of noise in the space;
[0073] P2 Comfortable: The scale of the building is in good proportion to the sidewalks and greenery, without creating a sense of oppression, and the visual elements are clean and free of clutter.
[0074] P3 Clean: The street environment is well maintained, and the building facades are clean and tidy;
[0075] P4 safe: The park has a wide field of vision, no safety hazards, and the separation of pedestrians and vehicles on the roads is relatively clear;
[0076] P5 Beautiful: The overall environment is clean and orderly, meaning the architectural design is unique or aesthetically pleasing, and the green landscape is carefully planned and maintained;
[0077] P6 Relaxed: The park has a relaxed pace of life, a low plot ratio, and is equipped with green spaces, leisure trails and other recreational facilities.
[0078] P7 Novel: The park layout or internal architecture adopts a unique, modern spatial form and design language;
[0079] P8 is rich in color: the park space features a variety of colors in strong contrast or harmonious combination, forming a vibrant visual focus;
[0080] C1 Noise Level: Whether there is a wide and dense green belt that separates the sidewalks, office areas and main roads (the green belt serves as a physical sound insulation facility within the park).
[0081] C2 Facility Quality: The physical condition and quality of the park's hardware facilities, reflecting the level of facility maintenance and the park's investment costs;
[0082] C3 Greening Rate: The visual area ratio of green space in the park;
[0083] C4 Enclosure: The sense of spatial enclosure formed by buildings or natural barriers within the park, such as continuous building facades and semi-enclosed corridors;
[0084] C5 Color Scheme: Color coordination between building facades, public facilities, and natural landscapes;
[0085] C6 Environmental Belonging: Whether the scale of the park space is appropriate, whether there are public spaces for people to stay and communicate, and whether psychological identification is enhanced through spatial enclosure and interactive design;
[0086] C7 Scene Diversity: Does it have functional zoning to meet the multi-dimensional needs of production, living, and leisure, forming a composite park space?
[0087] C8 Openness of View: Whether the park has a low-density layout and whether the building spacing is spacious to form visual corridors, thereby reducing the feeling of visual oppression.
[0088] In one embodiment, step S3, which involves scoring the target historical real-scene image set according to the multi-level industrial park innovation atmosphere evaluation system to determine the evaluation benchmark, and determining the core MLLM evaluation model according to the evaluation benchmark, includes:
[0089] Step S31: Using an expert scoring model, the innovation atmosphere is scored based on the target historical real-scene image set according to the multi-level industrial park innovation atmosphere evaluation system, so as to determine an evaluation benchmark that can measure the reliability of the model evaluation results.
[0090] Specifically, this application will first establish an expert review panel composed of cross-disciplinary experts, with approximately 40% from urban planning, 30% from architecture, 20% from industrial development strategy, and 10% from environmental psychology. Next, each expert will score the innovation atmosphere based on the target historical real-scene image set constructed in step S1, according to a pre-determined multi-level industrial park innovation atmosphere evaluation system. Specifically, the expert scoring will use a 10-point scale, where 1 point corresponds to the lowest evaluation level and 10 points to the highest. Before scoring, all experts will receive standardized training to clarify the indicator definitions and scoring criteria. The scoring process will employ an independent scoring system. Finally, after completing the scoring, extreme outliers deviating from the group mean ± 2 standard deviations will be removed, and the average of the remaining scores will be used as the expert benchmark score to ensure the authority and stability of the benchmark data.
[0091] Step S32: Select multiple candidate MLLM models and unify the input content and temperature parameter of each candidate MLLM model.
[0092] Specifically, based on the cutting-edge nature of MLLM (Multimodal Large Language Model) technology and its adaptability to industrial scenarios, this application selected six mainstream models for comparison (including Gemini-2.5-pro, Gemini-2.5-flash, GPT-4o, Doubao-Seed-1.6-thinking, QvQ-Max, and Claude 3.5Sonnet).
[0093] It should be noted that all candidate MLLM models must be run in a unified testing environment. The hardware of this environment can be configured as a GPU Tesla V100 with 64GB of memory, and the software environment is Python 3.9 and PyTorch 2.0 to eliminate the interference of environment differences on the test results (the environment configuration here is not unique; as long as it meets the basic hardware resource requirements for model operation and the software version is compatible with model code execution, it can be used as an alternative configuration).
[0094] In one embodiment, this application unifies the input content of each candidate MLLM model to "image dataset + benchmark prompt words".
[0095] In one embodiment, this application unifies the temperature parameter of each candidate MLLM model to 0.2, thereby reducing output randomness and ensuring result stability.
[0096] Step S33: After inputting the target historical real-scene image set and the corresponding benchmark prompt words into each candidate MLLM model, the predicted value output by each candidate MLLM model and the corresponding evaluation benchmark are respectively substituted into the multi-dimensional model verification index system to calculate the evaluation index values, so as to obtain the specific quantitative results of each candidate MLLM model under the preset verification index.
[0097] Specifically, the preset verification indicators include: Mean Absolute Error (MAE), which measures the average deviation between the model's predicted value and the actual value of expert evaluation by calculating the average absolute difference between the two. A smaller MAE value indicates a more accurate model prediction. The Pearson correlation coefficient (r) quantifies the linear correlation between the model's predicted value and the expert rating. Its value ranges from -1 to 1; the closer to 1 or -1, the stronger the linear correlation, and the closer to 0, the weaker the correlation. In the evaluation, we expect this coefficient to be as close to 1 as possible. The Max Error reflects the model's robustness in handling extreme cases, i.e., the situation where the difference between the model's predicted value and the actual value is largest. A smaller Max Error indicates a stronger ability of the model to cope with extreme cases. The Coefficient of Determination (R²) reflects the model's ability to explain the variance of expert ratings. The closer the R² value is to 1, the higher the proportion of expert rating variance that the model can explain, and the better the model's fit.
[0098] These quantitative results enable a comprehensive and in-depth evaluation of the merits and demerits of each candidate MLLM model, providing a solid basis for the subsequent selection of the core evaluation model most suitable for assessing the innovation atmosphere of industrial parks.
[0099] Step S34: Compare the quantitative values of each candidate MLLM model under the same validation metrics, and select the candidate MLLM model that can show comprehensive advantages under each validation metric as the core MLLM evaluation model.
[0100] In the above embodiments, an evaluation benchmark is first determined based on a set of historical real-world images of the target through expert scoring, providing a reliable reference for subsequent model evaluation. Then, a multi-dimensional model validation index system is used to calculate the quantitative results of each candidate model to comprehensively measure model performance. Finally, the core MLLM evaluation model with outstanding comprehensive advantages is selected by comparing the quantitative values to ensure that the selected model has high accuracy, reliability, and comprehensiveness in evaluating the innovation atmosphere of industrial parks, laying a solid technical foundation for accurately assessing the innovation atmosphere of industrial parks.
[0101] In one embodiment, in step S33, the multi-dimensional model validation index system includes the mean absolute error used to measure the average deviation between the model prediction value and the expert score, the Pearson correlation coefficient used to measure the linear correlation between the model prediction value and the expert score, the maximum error used to measure the maximum difference between the model prediction value and the expert score in all samples, and the coefficient of determination used to measure the model's ability to explain the variance of the expert score.
[0102] It should be noted that this application does not innovate the calculation formulas for each indicator. The calculation methods of the above indicators fall within the scope of existing technology, and this application will not elaborate further on them.
[0103] In one embodiment, step S4, which involves building an automated innovation atmosphere evaluation framework based on the core MLLM evaluation model, according to the organic integration of the core driving entity identity definition, evaluation perspective, and corresponding priority evaluation indicators, includes:
[0104] Step S41 identifies key participants in the innovation activities of the industrial park as the core driving force.
[0105] Specifically, this application identifies the following four key stakeholders as the core driving forces in the innovation ecosystem of industrial parks: company founders, researchers, investors, and park managers.
[0106] Step S42: For each core driving entity, create a corresponding role profile based on the core needs it focuses on in the park's innovation activities.
[0107] Specifically, for each type of core driving entity, this application will develop a detailed role profile based on its core needs in the park's innovation activities, including:
[0108] (1) The role profile of the enterprise founder focuses more on “team efficiency improvement” and “attracting core talents”. This application will specify that the priority evaluation indicators are “novel” spatial form and “scenario diversity”.
[0109] (2) The role profile of researchers focuses more on “concentration protection” and “cognitive fatigue relief”. This application will specify that the priority evaluation indicators are a “quiet” environment and a “comfortable” space.
[0110] (3) The investor role profile focuses more on “asset value” and “industrial vitality”. This application will specify that its priority evaluation indicators are “environmental belonging” and “facilities quality”.
[0111] (4) The role profile of park managers focuses more on “overall image maintenance” and “enhancing enterprise attractiveness”. This application will specify that the priority evaluation indicators are “noise level” and “clean” environment.
[0112] Step S43: Determine the priority evaluation indicators corresponding to each role profile from the indicator layer of the multi-level industrial park innovation atmosphere evaluation system, and set structured prompts for each type of core driving entity based on the definition of the driving entity's identity, evaluation perspective and corresponding priority evaluation indicator description, scoring requirements and scoring basis.
[0113] For example, the structured prompts for company founders should clearly state the subject identity of "technology company founders" and list the key evaluation indicators from the perspective of "team efficiency and talent attraction". They should also provide a score of 1-10 and the basis for the score, and output the total score.
[0114] It should be noted that these structured prompts provide highly customized input specifications for the core MLLM assessment model, guiding it to accurately focus on the key concerns of different core driving entities within the innovation atmosphere of industrial parks. By clearly defining the entity's identity, assessment perspective, priority assessment indicators, scoring requirements, and basis, the model can avoid ambiguity and bias in information processing, thereby generating more realistic, accurate, and reliable assessment results, providing a solid basis for optimizing the innovation atmosphere of industrial parks.
[0115] Step S44: For each type of core driving entity, build an automated innovation climate assessment framework as follows:
[0116] (1) The set of real-world images to be evaluated and the corresponding structured prompts are used as inputs to the core MLLM evaluation model.
[0117] (2) The core MLLM evaluation model calculates the sub-scores corresponding to each specific indicator in the indicator layer based on the input data.
[0118] Specifically, after receiving the input set of real-world images and structured prompts, the core MLLM assessment model utilizes its powerful multimodal understanding capabilities for comprehensive analysis. For real-world images, the model extracts feature information from the images using deep learning techniques such as convolutional neural networks, identifying elements related to the innovation atmosphere, such as innovation facilities and research personnel activity scenes. Simultaneously, the model combines the subject identity, assessment perspective, and priority assessment indicators from the structured prompts to conduct targeted analysis of the extracted image features. For example, when assessing the innovation atmosphere of a company founder, the model focuses on image features related to team efficiency and talent attraction, and calculates the sub-scores for each specific indicator according to a preset scoring standard based on the degree of matching between these features and the priority assessment indicators.
[0119] (3) The core MLLM evaluation model uses the entropy weight method to assign weights to the scores of each item according to the difference in the degree of attention to demand and sum them up, and outputs the total score of the park's innovation atmosphere.
[0120] Specifically, the core MLLM assessment model employs the entropy weight method to assign weights to each sub-item score based on differences in the degree of attention given to needs, and then performs a weighted sum. The entropy weight method has the advantage of objectively reflecting the amount of data information and the importance of indicators. In assessing the innovation atmosphere of industrial parks, since different core driving entities have varying degrees of attention to each indicator, the entropy weight method can determine the weight of each sub-item score based on the data dispersion. This approach allows for a more scientific and reasonable synthesis of the sub-item scores, resulting in an accurate overall score that reflects the park's innovation atmosphere.
[0121] In the above embodiments, by designating key participants in innovation activities within industrial parks as core driving entities and creating role profiles for them, prioritizing evaluation indicators and setting structured prompts based on these profiles, an automated innovation atmosphere evaluation framework is established. This series of steps achieves the following: starting with the accurate identification of key entities and their needs, utilizing a core MLLM evaluation model, combined with structured prompts and a set of real-world images, automatically calculating scores for each indicator, and then using the entropy weight method to rationally allocate weights and derive a total score. This process improves the relevance, objectivity, and automation of the evaluation, enabling a comprehensive, accurate, and efficient assessment of the innovation atmosphere in industrial parks.
[0122] In one embodiment, after step S4, the method further includes:
[0123] Step S5: During the actual evaluation, the total innovation atmosphere score corresponding to different core driving entities output by the automated innovation atmosphere evaluation framework is associated with the spatial coordinates of the industrial park, and an innovation atmosphere spatial distribution map is generated accordingly.
[0124] Specifically, this application matches the overall innovation atmosphere scores of different core driving entities (such as corporate R&D personnel, venture capitalists, and park managers) output by the automated innovation atmosphere assessment framework with corresponding spatial identifiers according to the specific areas of the park to which the assessment corresponds. For example, if the overall innovation atmosphere score for corporate R&D personnel in a certain R&D building is 8 points, this overall score is associated with the spatial identifier of that R&D building. Then, geographic information software is used to visualize the areas corresponding to the spatial identifiers of the associated score data on a map. Specifically, different colors, fill patterns, or transparency can be used to represent the scores, such as more vibrant colors and denser fills for higher scores. Additionally, annotation information can be added to each area to display the type of core driving entity and the corresponding overall innovation atmosphere score.
[0125] In the above embodiments, by associating the total score with spatial coordinates and presenting it in the form of a spatial distribution map, abstract data is transformed into intuitive graphics, allowing users to quickly understand the innovative atmosphere in different areas of the park, which greatly improves the efficiency of information transmission.
[0126] Please refer to Figure 2 This application discloses an industrial park innovation atmosphere evaluation system, which includes a data preprocessing module, an innovation atmosphere evaluation system construction module, an MLLM evaluation model screening module, and an automated evaluation framework construction module, wherein:
[0127] The data preprocessing module is used to preprocess the acquired initial historical real-scene image set to obtain the target historical real-scene image set.
[0128] The innovation atmosphere assessment system construction module is used to construct a multi-level industrial park innovation atmosphere assessment system that includes a target layer, a criterion layer, and an indicator layer. The target layer is used to clarify the overall goal of industrial park innovation atmosphere assessment. The criterion layer is used to break down the overall goal clarified in the target layer into two intermediate dimensions: "intuitive perception" and "element characteristics". The indicator layer is used to determine specific indicators for each dimension of the criterion layer.
[0129] The MLLM evaluation model screening module is used to score the target historical real-scene image set according to the multi-level industrial park innovation atmosphere evaluation system to determine the evaluation benchmark, and to determine the core MLLM evaluation model according to the evaluation benchmark.
[0130] The automated evaluation framework building module is used to build an automated innovation atmosphere evaluation framework based on the core MLLM evaluation model, according to the organic integration of the core driving subject identity definition, evaluation perspective and corresponding priority evaluation indicators.
[0131] In one embodiment, the above modules are also used to implement an industrial park innovation atmosphere assessment method as described in any of the foregoing method embodiments, and this application does not limit this.
[0132] As can be seen from the above, the industrial park innovation atmosphere evaluation system disclosed in this application constructs a multi-level industrial park innovation atmosphere evaluation system. The objective layer clarifies the overall direction of the evaluation, the criterion layer breaks down the overall objective into two intermediate dimensions: "intuitive perception" and "element characteristics," and the indicator layer further refines specific indicators. This hierarchical design makes the evaluation system logically clear and hierarchically distinct, comprehensively and systematically covering all aspects of the industrial park innovation atmosphere, ensuring the comprehensiveness and accuracy of the evaluation. By evaluating and comparing multiple candidate MLLM models according to the evaluation benchmark, the core MLLM evaluation model with the best overall performance under various verification indicators can be selected from numerous models. This model can more accurately understand and process information in image sets, providing strong technical support for subsequent automated assessments and improving the accuracy and efficiency of assessments. Through role simulation mechanisms, it quantifies the perceptual differences of core subjects. For example, users focus on micro-level spatial quality, while service providers focus on macro-level value, solving the shortcomings of traditional "one-size-fits-all" assessments and making assessment results more aligned with actual needs. It can provide "full-cycle" decision-making tools for park planning, stock renewal, and regional coordination, helping to build a high-quality innovation ecosystem that "adapts to spatial needs" and enhancing the park's ability to support innovation activities.
[0133] Please refer to Figure 3 This application discloses a readable storage medium, which includes a method program for evaluating the innovation atmosphere of an industrial park. When the method program is executed by a processor, it implements the steps of the method described in any of the preceding claims.
[0134] As can be seen from the above, the readable storage medium disclosed in this application constructs a multi-level industrial park innovation atmosphere evaluation system. The objective layer clarifies the overall direction of the evaluation, the criterion layer breaks down the overall objective into two intermediate dimensions: "intuitive perception" and "element characteristics," and the indicator layer further refines specific indicators. This hierarchical design makes the evaluation system logically clear and hierarchically distinct, comprehensively and systematically covering all aspects of the industrial park's innovation atmosphere, ensuring the comprehensiveness and accuracy of the evaluation. By evaluating and comparing multiple candidate MLLM models according to the evaluation benchmark, the core MLLM evaluation model with the best overall performance under various validation indicators can be selected from numerous models. This model can more accurately understand and process information in image sets, providing strong technical support for subsequent automated assessments and improving the accuracy and efficiency of assessments. Through role simulation mechanisms, it quantifies the perceptual differences of core subjects. For example, users focus on micro-level spatial quality, while service providers focus on macro-level value, solving the shortcomings of traditional "one-size-fits-all" assessments and making assessment results more aligned with actual needs. It can provide "full-cycle" decision-making tools for park planning, stock renewal, and regional coordination, helping to build a high-quality innovation ecosystem that "adapts to spatial needs" and enhancing the park's ability to support innovation activities.
[0135] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for evaluating the innovation atmosphere of an industrial park, characterized in that, Includes the following steps: S1. Perform data preprocessing on the acquired initial historical real-scene image set to obtain the target historical real-scene image set; S2. Construct a multi-level industrial park innovation atmosphere evaluation system that includes a target layer, a criterion layer, and an indicator layer. The target layer is used to clarify the overall goal of industrial park innovation atmosphere evaluation. The criterion layer is used to decompose the overall goal clarified in the target layer into two intermediate dimensions: "intuitive perception" and "element characteristics". The indicator layer is used to determine specific indicators for each dimension of the criterion layer. S3. Based on the multi-level industrial park innovation atmosphere evaluation system, score the target historical real-scene image set to determine the evaluation benchmark, and determine the core MLLM evaluation model according to the evaluation benchmark; S4. Based on the core MLLM evaluation model, an automated innovation atmosphere evaluation framework is built by organically integrating the core driving entity identity definition, evaluation perspective and corresponding priority evaluation indicators.
2. The method according to claim 1, characterized in that, In step S1, the data preprocessing of the acquired initial historical real-scene image set to obtain the target historical real-scene image set includes: S11. Remove interfering images containing irrelevant elements and / or blurry images with substandard image quality from the initial set of historical real-scene images to obtain the core set of historical real-scene images. S12. Denoise and tilt angle correction are performed on each image in the core historical real-scene image set to obtain an optimized historical real-scene image set; S13. Convert each image in the optimized historical real-scene image set into a unified image format to obtain the target historical real-scene image set.
3. The method according to claim 1, characterized in that, In step S2, the overall objectives of the industrial park innovation atmosphere assessment include improving focus, regulating negative emotions, and enhancing mental activity.
4. The method according to claim 3, characterized in that, In step S2, specific indicators determined based on the intermediate dimension of intuitive feeling, with regard to the overall goal of improving focus, include tranquility, comfort, and cleanliness; To achieve the overall goal of improving focus, specific indicators determined based on the intermediate dimension of element characteristics include noise level, facility quality, greening rate, and enclosure degree. For the overall goal of regulating negative emotions, specific indicators determined based on the intermediate dimension of intuitive feelings include tranquility, comfort, safety, beauty, and relaxation; To address the overall goal of regulating negative emotions, specific indicators determined based on the intermediate dimension of element characteristics include noise levels, greening rate, color matching, and sense of belonging to the environment. With the overall goal of improving mental agility, specific indicators determined based on the intermediate dimension of intuitive feeling include ease, novelty, and richness of color; To achieve the overall goal of improving the activity of thinking, specific indicators determined based on the intermediate dimension of element characteristics include greening rate, color matching, scene diversity, and field of vision.
5. The method according to claim 1, characterized in that, In step S3, the scoring of the target historical real-scene image set based on the multi-level industrial park innovation atmosphere evaluation system to determine the evaluation benchmark, and the determination of the core MLLM evaluation model according to the evaluation benchmark, includes: S31. Using an expert scoring model, the innovation atmosphere is scored based on the target historical real-scene image set according to the multi-level industrial park innovation atmosphere evaluation system, so as to determine the evaluation benchmark that can measure the reliability of the model evaluation results. S32. Select multiple candidate MLLM models and unify the input content and temperature parameter of each candidate MLLM model. S33. After inputting the target historical real-scene image set and the corresponding benchmark prompt words into each candidate MLLM model, the predicted value output by each candidate MLLM model and the corresponding evaluation benchmark are respectively substituted into the multi-dimensional model verification index system to calculate the evaluation index values, so as to obtain the specific quantitative results of each candidate MLLM model under the preset verification index. S34. Compare the quantitative values of each candidate MLLM model under the same validation metrics, and select the candidate MLLM model that can show comprehensive advantages under each validation metric as the core MLLM evaluation model.
6. The method according to claim 5, characterized in that, In step S33, the multi-dimensional model validation index system includes the mean absolute error used to measure the average deviation between the model's predicted values and the expert scores, the Pearson correlation coefficient used to measure the linear correlation between the model's predicted values and the expert scores, the maximum error used to measure the maximum difference between the model's predicted values and the expert scores in all samples, and the coefficient of determination used to measure the model's ability to explain the variance of the expert scores.
7. The method according to claim 1, characterized in that, In step S4, the automated innovation atmosphere evaluation framework is built based on the core MLLM evaluation model, according to the organic integration of the core driving entity identity definition, evaluation perspective, and corresponding priority evaluation indicators. This includes: S41. Key participants in innovation activities within industrial parks should be the core driving force. S42. For each core driving entity, create a corresponding role profile based on the core needs it focuses on in the park's innovation activities; S43. Determine the priority evaluation indicators corresponding to each role profile from the indicator layer of the multi-level industrial park innovation atmosphere evaluation system, and set structured prompt words for each type of core driving entity based on the definition of the driving entity's identity, evaluation perspective and corresponding priority evaluation indicator description, scoring requirements and scoring basis. S44. For each type of core driving entity, construct an automated innovation climate assessment framework as follows: (1) The set of real-world images to be evaluated and the corresponding structured prompts are used as inputs to the core MLLM evaluation model; (2) The core MLLM evaluation model calculates the sub-scores corresponding to each specific indicator in the indicator layer based on the input data; (3) The core MLLM evaluation model uses the entropy weight method to assign weights to the scores of each item according to the difference in the degree of attention to demand and sum them up, and outputs the total score of the park's innovation atmosphere.
8. The method according to any one of claims 1-7, characterized in that, After step S4, the method further includes: S5. During the actual evaluation, the total innovation atmosphere score corresponding to different core driving entities output by the automated innovation atmosphere evaluation framework is associated with the spatial coordinates of the industrial park, and an innovation atmosphere spatial distribution map is generated accordingly.
9. An evaluation system for the innovation atmosphere of an industrial park, characterized in that, The system includes a data preprocessing module, an innovation atmosphere evaluation system construction module, an MLLM evaluation model selection module, and an automated evaluation framework construction module, wherein: The data preprocessing module is used to preprocess the acquired initial historical real-scene image set to obtain the target historical real-scene image set. The innovation atmosphere assessment system construction module is used to construct a multi-level industrial park innovation atmosphere assessment system that includes a target layer, a criterion layer, and an indicator layer. The target layer is used to clarify the overall goal of industrial park innovation atmosphere assessment. The criterion layer is used to decompose the overall goal clarified in the target layer into two intermediate dimensions: "intuitive perception" and "element characteristics". The indicator layer is used to determine specific indicators for each dimension of the criterion layer. The MLLM evaluation model screening module is used to score the target historical real-scene image set according to the multi-level industrial park innovation atmosphere evaluation system to determine the evaluation benchmark, and to determine the core MLLM evaluation model according to the evaluation benchmark. The automated evaluation framework building module is used to build an automated innovation atmosphere evaluation framework based on the core MLLM evaluation model, according to the organic integration of the core driving subject identity definition, evaluation perspective and corresponding priority evaluation indicators.
10. A readable storage medium, characterized in that, The readable storage medium includes a method program for evaluating the innovation atmosphere of an industrial park, which, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 8.