Multi-modal fusion data processing method and system based on AI modeling

Abstract

The application discloses a multi-modal fusion data processing method and system based on AI modeling, and relates to the technical field of electric digital data processing. The method collects multi-modal data of a specified urban renewal area, performs collaborative fusion processing to obtain global unified data, performs compliance preliminary screening based on an AI-driven data consistency verification model, then performs demand adaptability verification, generates three-dimensional coupled data in combination with a specified AI algorithm, simultaneously performs scene compliance screening, finally performs decision dimension fusion processing in combination with a decision fusion algorithm, obtains decision dimension fusion data, and performs compliance rechecking, realizes AI-driven whole-process data processing and decision support closed loop, helps to capture the connection relationship between CAD drawing engineering properties and smart city dynamic management requirements, and effectively improves the collaboration and timeliness of multi-modal data fusion, and meets the fine and dynamic application requirements of smart cities.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, and in particular to a multimodal fusion data processing method and system based on AI modeling. Background Technology

[0002] As engineering design, smart governance, and spatial optimization become more refined and intelligent, multimodal data fusion has become a core focus of cross-sectoral integration in smart cities. Breakthroughs in artificial intelligence (AI) modeling technology have provided core support for multimodal fusion, leveraging the powerful feature extraction and nonlinear mapping capabilities of deep learning frameworks such as Transformer to construct an efficient data processing core. Meanwhile, the widespread adoption of IoT and cloud computing technologies has generated massive amounts of multimodal heterogeneous data, and these two technologies have synergistically driven the research and development of AI-driven multimodal fusion data processing technologies.

[0003] Taking Chinese invention patent CN118012977B as an example, it discloses a two-dimensional and three-dimensional multimodal data processing method based on the integration of AI and GIS. Specifically, it includes using several sets of sensors to capture and record remote sensing images of the earth's surface and dividing them into two-dimensional geographic data information and three-dimensional spatial data information; enhancing the comprehensive understanding of the two types of data by using multimodal data captured by multiple sensors; introducing a data balance assessment index to comprehensively evaluate the balance of geographic data by integrating multiple dimensions; and adopting corresponding processing strategies according to different levels of the data balance assessment index by pre-setting assessment thresholds.

[0004] The technological logic presented in this patent is a typical practice of multimodal data fusion in specific scenarios: it achieves accurate analysis and balanced optimization of multi-source data through AI algorithms, and builds a bridge connecting two-dimensional and three-dimensional data with the help of Geographic Information System (GIS) technology. Its core idea is highly consistent with the technological needs of cross-border integration in smart cities. As a core hot topic in smart city construction, multimodal data fusion, based on the technology of the aforementioned patent, further enhances the intelligence and scenario adaptability of data processing by combining AI modeling technology, forming a higher-dimensional technological barrier. It can not only meet the engineering needs of geographic data processing, but also accurately adapt to the demands of diverse scenarios such as urban renewal and building management, providing key support for smart cities to move from technological exploration to practical application.

[0005] Existing technologies extract core features from different modalities such as images, text, and sensors using models like Transformer. They achieve feature semantic alignment through contrastive learning and cross-modal attention mechanisms, mapping multi-source data to a unified shared embedding space. Feature-level or decision-level fusion strategies are employed to integrate complementary information, and a quantitative evaluation mechanism is introduced simultaneously to optimize the consistency between data and actual technical specifications. This effectively resolves conflicts between multimodal data and actual application scenarios, strengthens the data's support for construction implementation, and provides key technical support for the intelligent upgrading of smart city-related scenarios.

[0006] However, while existing technologies have achieved the fusion of two-dimensional geographic data and three-dimensional spatial data, they often fail to fully address the practical needs of engineering design scenarios such as urban renewal and building construction. For example, in the renovation of old buildings during urban renewal, core engineering attributes recorded in computer-aided design drawings (CAD) such as building structural load-bearing parameters and pipeline layout topology are often difficult to accurately correlate with real-time building settlement data captured by sensors and surrounding geospatial planning data. This leads to temporal and logical conflicts in the fusion of engineering attributes of CAD data with other modal data during actual renovation design and construction simulation, resulting in the loss or distortion of key information in some scenarios. Fusion models built based on such insufficiently coordinated data are unable to effectively reflect the cross-influence of dynamic factors such as dynamic spatial changes in urban renewal and equipment status fluctuations in building management.

[0007] Furthermore, scenarios such as regional planning adjustments in urban renewal and updates to equipment operation and maintenance data in building management can cause correlation fluctuations between basic engineering data such as building layout, component dimensions, and pipeline routing in CAD drawings and dynamic scene data collected in real time by sensors. Existing models lack an adaptive adjustment mechanism for parameter weights based on dynamic scene changes, failing to match the correspondence between CAD engineering data and dynamic scene data in real time, resulting in model outputs often lagging behind actual scenario requirements. Simultaneously, the complexity of multimodal data transmission links and differences in device response exacerbate the timing deviations in data fusion. The high demands of smart city refined management for data collaboration, coupled with the inability of CAD drawings to adapt engineering baseline data (such as building layout and component dimensions) to real-time dynamic scene conditions, further amplify these issues. Ultimately, this leads to a significant decrease in the synergy and timeliness of multimodal data fusion, making it difficult to support the precise connection between the engineering attributes of CAD drawings and the dynamic management needs of smart cities. Summary of the Invention

[0008] To address the technical problems of low collaboration and timeliness in existing multimodal data fusion technologies, this invention provides a multimodal data fusion processing method and system based on AI modeling. The technical solution is as follows: On the one hand, a multimodal fusion data processing method based on AI modeling is provided. This method includes: Step 1, taking the CAD drawings of a designated urban renewal area as the benchmark, the collected multimodal data is collaboratively fused to obtain unified data across the entire domain, and data consistency verification and initial compliance screening are performed based on an AI-driven data consistency verification model; Step 2, the unified data across the entire domain that has passed the initial compliance screening is subjected to demand adaptability verification, and three-dimensional coupled data is generated by combining a specified AI algorithm for scenario compliance screening; Step 3, based on the three-dimensional coupled data that has passed the scenario compliance screening, decision dimension fusion processing is performed by combining a decision fusion algorithm to obtain decision dimension fusion data, and compliance review is performed.

[0009] On the other hand, an AI-based multimodal fusion data processing system is provided. This system includes: a multimodal data processing and compliance screening module, which is used to collaboratively fuse collected multimodal data to obtain unified data across the entire domain, and to perform data consistency verification and compliance screening based on an AI-driven data consistency verification model; an adaptability verification and scenario compliance screening module, which is used to perform requirement adaptability verification on the unified data across the entire domain after compliance screening, and to generate three-dimensional coupled data in combination with a specified AI algorithm for scenario compliance screening; and a decision dimension fusion processing and compliance review module, which is used to perform decision dimension fusion processing based on the three-dimensional coupled data after scenario compliance screening, combined with a decision fusion algorithm to obtain decision dimension fusion data, and to perform compliance review.

[0010] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. This invention uses CAD drawings of designated urban renewal areas as a benchmark, collects multimodal data from these areas, and collaboratively integrates and processes it into unified data across the entire region. Combined with a data consistency verification model, it performs consistency verification and initial compliance screening, achieving precise mapping between core CAD engineering attributes such as building structural load-bearing parameters and pipeline topology and real-time building settlement data and geospatial planning data. This effectively avoids the loss or distortion of key information caused by temporal logic conflicts. After passing the initial compliance screening, the unified data across the entire region undergoes requirement adaptation verification, and a specified AI algorithm is used to generate three-dimensional coupled data for further processing. The scenario compliance screening reduces the conflict between CAD drawing planning logic and actual management needs, and improves the fit between data and specific scenarios such as old building renovation and construction simulation through scenario-based adaptation mechanism. Based on the fusion of decision dimensions and compliance review of three-dimensional coupled data, it can accurately capture the cross-effects of dynamic spatial changes in urban renewal and fluctuations in building equipment status, match the correspondence between CAD engineering data and actual dynamic scenario data in real time, effectively eliminate the problems of time sequence deviation and feedback lag, significantly improve the synergy and timeliness of multimodal data fusion, and provide solid decision support for the refined and dynamic management of smart cities.

[0011] 2. By classifying and grouping CAD drawing data, building equipment operation data, and environmental perception data, data adaptability is quantitatively evaluated from three dimensions: spatial overlap, temporal synchronization rate, and semantic fit. Quantitative calculations using overlapping area ratio, timestamp alignment ratio, and cosine similarity ensure data consistency across spatial, temporal, and semantic levels, effectively addressing the issue of loose multimodal data association. Valid data is filtered according to corresponding reference values, while invalid data is removed. After standardization, a full-domain data collaborative fusion process is executed. First, data is classified by spatial partitions. Then, timestamp alignment eliminates association biases caused by time differences in data collection. Spatial overlap is combined to form a spatially associated dataset. Data from each partition is then stitched together to generate unified data across the entire domain, achieving precise association between CAD drawing engineering attributes and measured data. This effectively avoids temporal logic conflicts and key information distortion, providing standardized and highly reliable data support covering spatial geometry, equipment operation, and environmental status for subsequent demand adaptation and decision generation, thus improving the collaboration and practicality of multimodal data fusion.

[0012] 3. First, verification units are divided according to functional zones, constructing a two-dimensional verification dataset of CAD drawing design parameters and measured data. The consistency deviation value quantifies the degree of data matching, accurately identifying differences between drawings and on-site data. Simultaneously, the latency of data transmission and acquisition is integrally calculated to quantify timeliness performance with a data real-time index, effectively avoiding feedback lag issues in dynamic scenarios. Correction plans are generated for abnormal units, and processing prompts are provided, ensuring data compliance and accuracy from the source. In the requirement adaptability verification phase, a data item list is compiled based on actual application requirements. Comprehensive data support is quantified through dimensional coverage, and the mapping and matching of requirement adaptability performance benchmarks and application requirement satisfaction indicators are combined to accurately determine the adaptability verification results. The entire process resolves conflicts between CAD drawings and on-site data through consistency verification and improves data timeliness through real-time evaluation. Requirement adaptability verification ensures a high degree of data compatibility with building operation and maintenance, spatial planning, and other scenarios, providing compliant, adaptable, and highly reliable data input for subsequent 3D coupled data generation.

[0013] 4. First, dimensional correlation verification is performed on the adaptable, unified data across the entire domain: a multivariate nonlinear mapping function is constructed, and the three-dimensional linkage coupling coefficient is obtained by solving the first-order partial derivatives. This accurately quantifies the strength of bidirectional linkage and cross-influence between the three types of data. Subsequently, data that meets the requirements of the three-dimensional linkage coupling coefficient is time-series aligned and organized into a three-dimensional matrix, forming three-dimensional coupled data containing quantified coupling parameters, providing deep correlation data support for scenario adaptation. Scenario compliance screening involves constructing a feature correlation matrix to calculate the Pearson correlation coefficient, and performing gradient correction on abnormal correlation items to ensure the rationality of data association logic. Then, the verified data is matched item by item with the management requirement list to ensure that the data meets the scenario constraints. The entire process not only achieves accurate capture of the cross-influence of dynamic factors through coupling coefficient quantification, but also improves the accuracy of data scenario adaptation through dual compliance verification, providing logically consistent and compliant high-quality data for subsequent decision fusion, significantly improving the reliability of data processing and decision support.

[0014] 5. Based on the 3D coupled data after scenario compliance screening, the median point derivative is solved using the dynamic interval median mapping method. This standardizes and maps multi-dimensional data with different dimensions and distributions to a unified decision space. Boundary constraint methods are used to limit the convergence range, ensuring the data approaches the optimal solution for dynamic management needs. Preliminary fusion data that meets spatial correlation constraints, equipment operation logic, and environmental adaptability requirements is selected, accurately capturing the dynamic cross-influence of space, equipment, and environment. By monitoring the absolute difference between adjacent results during iterative optimization, iterative optimization stops when the obtained absolute difference is less than a preset absolute difference, ensuring the consistency and dynamic adaptability of the decision data. In the compliance review stage, the fusion data of the decision dimensions is bidirectionally benchmarked against standard parameters of CAD drawings set based on planning and design specifications, building safety standards, etc. The degree of deviation is quantified by calculating the Euclidean distance between the spatial coordinate mapping value and the corresponding benchmark value. The entire process is supported by mathematical algorithms, which realizes the deep integration and dynamic adaptation of multimodal data. It not only suppresses the correlation fluctuation between CAD engineering data and dynamic scene data, but also improves data reliability through iterative optimization process monitoring and compliance review, providing high-quality decision-making basis for the refined management of smart cities. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart of a multimodal fusion data processing method based on AI modeling provided in an embodiment of the present invention; Figure 2 A flowchart for compliance screening and requirement adaptability verification provided in an embodiment of the present invention; Figure 3 This is an architecture diagram of the data consistency verification model provided in the embodiments of the present invention; Figure 4 A flowchart for generating three-dimensional coupled data and screening scene compliance provided in an embodiment of the present invention; Figure 5 A trend chart of the independent influence coefficient provided for embodiments of the present invention; Figure 6 A flowchart for decision-making dimension fusion data acquisition and compliance review provided in this embodiment of the invention; Figure 7 This is a schematic diagram of the structure of a multimodal fusion data processing system based on AI modeling provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0020] This invention provides a multimodal fusion data processing method based on AI modeling, such as... Figure 1 The flowchart shown is for a multimodal fusion data processing method based on AI modeling. The processing flow of this method may include the following steps: Step one involves using the CAD drawings of the designated urban renewal area as a benchmark to collaboratively integrate the collected multimodal data, obtain unified data across the entire area, and perform data consistency verification and initial compliance screening based on an AI-driven data consistency verification model. This aims to reduce conflicts between the design parameters of the CAD drawings and the actual conditions on-site in the designated urban renewal area, providing reliable data support processed by AI for subsequent demand adaptation and decision generation. The multimodal data includes CAD drawing data, building equipment operation data, and environmental perception data, all of which are collected from Internet of Things (IoT) sensing terminals and environmental monitoring sensors deployed in the building itself (equipment rooms, pipeline shafts, key component installation locations), public spaces (corridors, lobbies, outdoor plazas), and surrounding environment areas within the urban renewal area.

[0021] Step two involves verifying the adaptability of the unified data across the entire domain after the initial compliance screening, and generating three-dimensional coupled data using a specified AI algorithm. Simultaneously, scenario compliance screening is conducted to reduce conflicts between the planning logic of CAD drawings and the actual management needs of the designated urban renewal area, thereby improving the accuracy of AI adaptation of the data to specific application scenarios. The three-dimensional coupled data is used to quantify the bidirectional linkage effect and collaborative adaptation degree between spatial layout, building equipment operation status, and dynamic changes in environmental indicators within the designated urban renewal area. The specified AI algorithm here is gradient descent, used to solve for the independent and cross-influence coefficients of the spatial, equipment, and environmental dimensions, efficiently quantifying the intensity of three-dimensional linkage.

[0022] Step three involves using the 3D coupled data that has passed the scenario compliance screening as a foundation, and combining it with a decision fusion algorithm to perform decision dimension fusion processing to obtain decision dimension fusion data. Compliance verification is then conducted to reduce conflicts between the established framework of CAD drawings and the dynamic management needs of the designated urban renewal area, ensuring the feasibility and scenario adaptability of the decision data, and achieving a closed loop of AI-driven end-to-end data processing and decision support. The decision fusion algorithm here is a combination algorithm that uses the dynamic interval median mapping method to solve the median derivative of the data change intervals of each dimension, completing the standardization transformation and unified decision space mapping of data with different dimensions and distributions. Then, the boundary constraint method is used to limit the data convergence range and approximate the optimal solution for dynamic management needs. The decision dimension fusion data is used for core scenario decisions such as spatial layout optimization, equipment operation and maintenance strategy adjustment, environmental governance plan formulation, and engineering compliance verification in the designated urban renewal area. This typically includes optimal spatial coordinate mapping values, equipment operating parameter optimization schemes, environmental adaptation adjustment thresholds, and engineering constraint adaptation correction values.

[0023] Specifically, CAD drawing data typically includes core engineering information such as building structure layout, component dimension annotations, equipment installation coordinates, pipeline routing planning, and spatial zoning boundaries. Building equipment operation data typically includes operating parameters, energy consumption statistics, fault alarm records, start / stop status, and maintenance logs for equipment such as elevators, air conditioners, and water supply and drainage. Environmental perception data typically includes temperature and humidity, air quality index (PM2.5 / PM10), noise intensity, illuminance, wind speed and direction, and soil moisture. The unified data representation across the entire domain, after collaborative fusion processing, integrates three dimensions: spatial geometry, equipment operation, and environmental status, achieving a comprehensive data set with standardized format, logical consistency, conflict resolution, and redundancy elimination. CAD drawing design parameters typically include core engineering indicators such as spatial coordinate reference values, equipment installation spacing requirements, component load-bearing limit parameters, pipeline laying slope standards, and building structure safety constraints.

[0024] In a specific embodiment, for a renovation project of an old residential community, the process involves three steps: Step 1: Integrating building structure data from CAD drawings with elevator operation data and community temperature and humidity data collected by sensors; AI verification is used to remove invalid data where equipment installation coordinates deviate from the drawings; Step 2: Generating 3D coupled data using gradient descent to clarify the linkage coefficients between elevator start / stop frequency and corridor temperature and humidity, as well as building layout; Step 3: Outputting an elevator maintenance frequency optimization plan and public area ventilation adjustment thresholds, which are then reviewed for compliance with CAD drawing safety constraints. This example uses AI algorithms throughout the entire process of data fusion, coupled analysis, and decision generation to accurately resolve conflicts between the static benchmarks of CAD drawings and the dynamic management of the area, effectively improving the synergy of multimodal data and the accuracy of scenario adaptation; a multi-level compliance verification mechanism throughout the process ensures that decision data strictly conforms to engineering specifications and drawing safety constraints, providing feasible and highly reliable AI decision support for urban renewal projects, and helping to achieve management goals such as optimizing the spatial layout of the area, refining equipment operation and maintenance, and environmental governance.

[0025] Furthermore, the specific process for collaborative fusion processing is as follows: After data cleaning, the multimodal data is classified and grouped to obtain groups to be collaboratively fused. These groups serve as the basis for the collaborative association and efficient fusion of multimodal data across spatial, temporal, and semantic dimensions. The spatial overlap, temporal synchronization rate, and semantic fit of each group to be collaboratively fused are obtained. Spatial overlap is calculated by comparing the total amount of building equipment operation data and environmental perception data within the spatial coordinate range corresponding to the CAD drawing data with the maximum allowable proportion within the spatial coordinate range. Temporal synchronization rate is calculated by the ratio of the number of aligned timestamps of equipment operation timeline data and environmental perception data collection timestamps within the same production cycle. In runtime sequence data and environmental perception data, the timestamp error is no greater than the ratio of the number of aligned data entries with the expected synchronous acquisition timestamp to the total number of data entries in the device runtime sequence data and environmental perception data. The timestamp error represents the absolute difference between the acquisition timestamp of the device runtime sequence data and the acquisition timestamp of the corresponding environmental perception data. The expected synchronous acquisition timestamp is set comprehensively based on the hardware response delay (such as IoT terminal data transmission delay) and sensor sampling frequency (such as 1Hz / 5Hz) in the historical urban renewal area multimodal data acquisition process, as well as the accuracy requirements for timing consistency (such as millisecond-level / second-level synchronization). Semantic fit is obtained by extracting the engineering semantics of CAD drawings and the business semantics of measured data to construct feature vectors and calculating the cosine similarity.

[0026] For a given group to be processed through collaborative fusion, data that does not meet the preset conditions is recorded as invalid data, and data that does meet the preset conditions is recorded as valid data. Invalid data in each group is removed, while valid data is retained. The preset conditions indicate that spatial overlap, temporal synchronization rate, and semantic fit are all greater than their corresponding reference values. The reference values ​​for spatial overlap, temporal synchronization rate, and semantic fit are comprehensively set based on the functional zoning characteristics of historical urban renewal areas, the technical parameters of data acquisition equipment, multimodal data association business requirements, and historical data fusion optimization experience. Based on the retained valid data, a full-domain data collaborative fusion process is executed, specifically as follows: According to the spatial zoning boundaries of the designated urban renewal area (such as the zoning standards for buildings, floors, public areas, etc.), the valid data retained in each group to be collaboratively integrated are spatially classified. That is, the spatial coordinate parameters of the data association are matched and compared with the coordinates of the zoning boundaries to ensure that the CAD drawing data, equipment operation data, and environmental perception data of the corresponding area are gathered within the same zoning. Based on the preset timestamp (such as 1-hour / 5-minute interval timestamp), the building equipment operation data and environmental perception data within the same spatial zoning are timestamped to ensure that the two types of time-series data are synchronously matched in the time dimension and to eliminate the association deviation caused by the time difference of data collection. Based on the obtained spatial overlap, only the multimodal data with a spatial overlap of not less than the reference spatial overlap in the timestamped multimodal data within the same spatial zoning is retained to form a spatial association dataset. The reference spatial overlap is preset to 85%, which is usually set based on the historical functional zoning characteristics of the historical urban renewal area, the positioning accuracy of the data collection equipment, and the business requirements of multimodal data association.

[0027] If the field corresponding to the coordinate parameters of the target building equipment in the spatial association dataset is empty or has no collection timestamp record, it is determined that there is data missing in the spatial association dataset. The statistical average of the same type of data within the same spatial partition is then extracted to fill in the missing data. This approach not only aligns with the stability characteristics of similar equipment operation and spatial parameters within the urban renewal area but also quickly completes key information, avoiding data gaps that could affect the fusion effect. Labeling the source of the missing data ensures data traceability; the source of the missing data must be labeled as the statistical average of similar data within the same partition. If the field corresponding to the coordinate parameters of the target building equipment in the spatial association dataset is empty or has no collection timestamp record, it is determined that there is data missing in the spatial association dataset. If the recorded values ​​are verified, but the data collection frequency within the same time period (e.g., 1 hour, 1 collection batch) is greater than the data collection frequency pre-marked by the preset personnel in the CAD drawing, then it is determined that there is a data conflict in the spatial association dataset. Correction is made based on the spatial reference information marked in the CAD drawing. For example, if the CAD drawing marks the collection frequency of coordinate parameters of a certain type of elevator as 2 times per hour, and 3 times are actually collected, the valid data of the first 2 times are retained, and the redundant data of the 1st time is discarded to ensure consistency with the design standard of the drawing. After the conflicting data is corrected, the values ​​before and after the correction and the reason for the correction must be recorded.

[0028] After completing the data missing filling and data conflict correction in the spatial association dataset, the spatial association datasets of each spatial partition are spliced ​​region by region according to the spatial topology hierarchy marked by the CAD drawings of the specified urban renewal area (such as area-building-floor-functional zone). Using the reference coordinate system of the drawings as a unified reference, the spatial connection accuracy of the boundary data of adjacent partitions is first checked (such as coordinate coincidence and parameter consistency). The logical conflicts at the partition connection are eliminated by the redundancy data deduplication and boundary data completion algorithms. Then, the data of each partition are integrated according to the spatial topology order to generate unified data of the whole domain with three core dimensions: spatial geometry, equipment operation and environmental status and uniform data format.

[0029] In this embodiment, by verifying spatial, temporal, and semantic dimensions and deeply binding with CAD drawing benchmarks, invalid data is removed, missing and conflicting data are precisely handled, and the entire domain is topologically stitched together to construct a unified data system with a consistent format, logical coherence, and both engineering constraints and dynamic adaptability. This effectively resolves the conflict between the static engineering benchmarks of CAD drawings and the dynamic management needs of urban renewal areas, ensuring the integrity, consistency, and compliance of the data. It also achieves the synergy of spatial geometry, equipment operation, and environmental status data, providing highly reliable data support for subsequent three-dimensional coupled analysis and AI decision generation. This improves the refined management level and engineering implementation efficiency of urban renewal projects, thereby effectively reducing operation and maintenance risks.

[0030] like Figure 2The flowchart shown illustrates the compliance screening and requirement suitability verification process. The specific steps of the compliance screening are as follows: First, obtain the divided verification units. Then, retrieve the CAD drawing design parameters from the unified data across the entire domain within each verification unit, along with the currently collected building equipment operation data and environmental perception data, to form a two-dimensional verification dataset. Next, call the constructed data consistency verification model. Using the CAD drawing design parameters as a benchmark, calculate the consistency deviation value between the corresponding data in each verification unit within the two-dimensional verification dataset. This quantifies the degree of matching between the CAD drawing design parameters and the currently collected building equipment operation data and environmental perception data. The consistency deviation value is obtained by calculating the Euclidean distance between the currently collected building equipment operation data and environmental perception data in the two-dimensional verification dataset and the corresponding design parameters in the CAD drawings, and then averaging the results.

[0031] like Figure 3 The diagram shows the architecture of the data consistency verification model. This model is a multimodal data verification architecture built on a deep learning framework (such as TensorFlow). The inputs to the model are: CAD drawing design parameters for each verification unit, currently collected building equipment operation data, and environmental perception data. The output is: the consistency deviation value corresponding to each verification unit. The model first captures the correlation features of different data dimensions through cross-modal multi-head attention, then performs residual connection and normalization processing, and finally completes the deep integration of features through a multimodal feature fusion network. The right branch uses multimodal target reference embedding as a benchmark, further strengthening the comparison features between the benchmark and the data to be verified through multimodal mask multi-head attention and cross-modal multi-head attention. Finally, the multi-label decision fusion thread layer outputs the consistency deviation value corresponding to each verification unit.

[0032] The model is trained using a two-dimensional compliant / non-compliant dataset of historical urban renewal projects as samples. Data features are extracted through convolutional neural networks, parameters are optimized by gradient descent, and the number of network layers and activation function are adjusted through cross-validation. Finally, a highly adaptable model with both binary classification (discriminating compliance) and deviation quantification capabilities is formed. It can accurately output the consistency deviation value of each verification unit, providing a core basis for subsequent initial screening of data compliance.

[0033] Within a preset monitoring period, the data transmission time of building equipment operation and the environmental perception data acquisition time of each verification unit are integrated to obtain the data real-time index of each verification unit. This index quantifies the real-time performance of multimodal data in the acquisition and transmission environment, as detailed below: Data Real-Time Index In the formula, i represents the number of verification units, and R i This represents the data real-time performance index of the i-th verification unit, with a value range of... The closer to 1, the better the real-time performance. T represents the preset monitoring period (unit: seconds), t represents the moment within the preset monitoring period, and D... max C represents the maximum allowable transmission time for building equipment operation data. max This indicates the maximum allowable collection time for environmental sensing data. This represents the data transmission duration of the building equipment at time t for the i-th verification unit. This represents the duration of environmental perception data acquisition for the i-th verification unit at time t.

[0034] The preset monitoring period T represents the time span, (D max +C max The maximum allowable time for data transmission and acquisition is multiplied by the total time, which is then used as the normalized denominator to obtain the theoretically acceptable maximum total time within the monitoring period. This design maps the integral result of the actual time consumption (total actual time consumption within the monitoring period) to the [0,1] interval, avoiding interference from the dimensional / numerical differences in the time consumption of the period and a single moment. Simultaneously, through... This yields a data real-time index, which accurately reflects the real-time level of data acquisition and transmission, ensuring the comparability of real-time indices for different verification units.

[0035] Specifically, the data transmission time for building equipment operation represents the time between the completion of data acquisition by the building equipment and the moment the data is transmitted to the corresponding server and received. The environmental perception data acquisition time represents the time between the trigger moment of the acquisition command corresponding to the environmental perception data and the moment the corresponding sensor completes data acquisition and encapsulates it into the expected format. The maximum allowable acquisition time and the maximum allowable transmission time are both comprehensively set based on the monitoring accuracy requirements of the specified urban renewal area, equipment response sensitivity standards, real-time requirements of data application scenarios, and industry standards. By monitoring the transmission delay and acquisition delay of each verification unit at each moment within the preset monitoring period, and integrating the two types of delays over time, the total delay within the period is obtained. Combined with the preset monitoring period, the maximum allowable transmission and acquisition time, normalization calibration is performed to finally obtain the data real-time index. This index comprehensively reflects the timeliness of data from acquisition to transmission, providing a quantitative basis for subsequent data validity judgment and transmission strategy optimization.

[0036] If the acquired data real-time index is not lower than the preset data real-time index, and the acquired consistency deviation value is not higher than the preset consistency deviation value, then the corresponding verification unit is marked as a qualified data unit, and the unified data of the entire domain after the compliance initial screening is obtained simultaneously and the requirement adaptability verification is performed; otherwise, the corresponding verification unit is marked as an abnormal data unit, and a deviation correction plan is generated in combination with the current CAD drawing reference information of the corresponding verification unit, an abnormal handling list is output and an abnormal handling prompt is sent, and the preset personnel take corresponding deviation correction measures based on the received abnormal handling list, such as optimizing the transmission protocol of the corresponding sensor; the preset data real-time index and the preset consistency deviation value are respectively represented by the sum and average of the historical data real-time index and historical consistency deviation value in the historical data compliance initial screening stage.

[0037] The requirement adaptation verification specifically involves: retrieving a list of data items reflecting the unified data across the entire domain; calculating the ratio of the actual number of data items already included in the unified data to the expected total number of data items to obtain the dimensional coverage completeness of the unified data for various application scenarios in the designated urban renewal area; if the dimensional coverage completeness is lower than the preset dimensional coverage completeness, the requirement adaptation verification result is defined as the first adaptation anomaly label; if the dimensional coverage completeness is not lower than the preset dimensional coverage completeness, the effective data volume of each application scenario covered in the unified data across the entire domain is counted, and the ratio is calculated with the total data volume of the corresponding application scenario requirements to obtain the application requirement satisfaction index to quantify the degree of support of the unified data across the entire domain for actual application requirements. The effective data volume of data items represents the number of actual data items in each application scenario that meet the actual usage requirements of each application scenario among the data items included in the unified data across the entire domain; the preset dimensional coverage completeness is set based on the industry data coverage standards of the application scenarios in the designated urban renewal area and historical adaptation verification experience, and is usually set to a value of not less than 80% to ensure that no core requirement dimensions are omitted.

[0038] If the application requirements of the unified data across the entire domain do not meet the performance tolerance indicators, the requirement adaptation verification result is defined as the second adaptation anomaly label. Otherwise, the requirement adaptation verification result is defined as feasible, and three-dimensional coupled data is generated using a specified AI algorithm. The performance tolerance range is represented by the closed interval corresponding to the maximum and minimum values ​​of the historical requirement adaptation performance benchmark values ​​for each application scenario in the historical urban renewal area. The first and second adaptation anomaly labels are parallel and independent adaptation anomaly types, corresponding to different core dimensions of requirement adaptation verification. Specifically, the first adaptation anomaly label focuses on the completeness of data dimension coverage, while the second adaptation anomaly label focuses on the rationality of data support capabilities. Both types of labels require the initiation of anomaly handling procedures (such as supplementing missing data and optimizing data filtering logic), and verification is carried out again after correction to ensure that the unified data across the entire domain meets the adaptation requirements.

[0039] In this embodiment, data consistency verification and compliance screening are used as dual checks. The data matching degree is quantified based on CAD drawings, and the transmission and collection timeliness is evaluated by combining the real-time index. Qualified data units are accurately screened to effectively ensure the consistency and timeliness of unified data across the entire domain. Demand adaptability verification is carried out from two dimensions: dimensional coverage completeness and application demand satisfaction. Abnormal adaptation scenarios and feasible scope are identified to ensure that the data fits the actual needs of urban renewal scenarios. This provides high-quality and highly adaptable data support for the subsequent generation of 3D coupled data, significantly improving the reliability of data application and the efficiency of scenario implementation.

[0040] like Figure 4 The flowchart shown illustrates the generation of 3D coupled data and scene compliance screening. The specific generation process for the 3D coupled data is as follows: Dimensional correlation verification is performed on adaptable, unified data across the entire domain. This verification includes checking the matching of spatial geometric data with the installation location of equipment operating parameters, and the logical consistency of the influence of equipment operating parameters and environmental perception indicators. Dimensional correlation verification is then performed, specifically using building equipment operating efficiency (e.g., energy consumption value m1, operating failure rate m2) and environmental indicator changes (e.g., temperature and humidity fluctuation values ​​m3, air quality index change rate m4) as the dependent variable set Y. The set of independent variables X is defined by the changes in spatial coordinates (including offsets x1 in the X-axis direction, y2 in the Y-axis direction, and z3 in the Z-axis direction). Construct a multivariate nonlinear mapping function, the specific expression of which is: Here, f1 represents the nonlinear influence of spatial coordinate offset on energy consumption. For example, the correlation between X-axis offset and equipment pipeline length will affect energy consumption through pipeline loss coefficient. f2 represents the nonlinear mapping of spatial coordinate offset on operational failure rate. For example, changes in equipment component assembly gaps caused by Y-axis offset will affect the frequency of failures through changes in mechanical wear rate. f3 and f4 respectively reflect the interference of spatial position changes on temperature and humidity distribution and air circulation path. For example, changes in ventilation opening height caused by Z-axis offset will affect the regional fluctuation range of temperature and humidity through airflow circulation efficiency. If the joint offset of X / Y axes changes the relative position of fresh air inlet, it will affect the rate of change of air quality index through airflow exchange rate.

[0041] Specifically, the energy or medium consumption of the corresponding building equipment is collected within a preset period (such as a complete operating cycle of the building equipment) by the electricity meter or flow meter built into the building equipment, and recorded as the energy consumption value; the operation failure rate is the ratio of the number of times the building equipment is triggered by failure to the running time of the corresponding building equipment within the preset period; the temperature and humidity of the corresponding building equipment area are collected by temperature and humidity sensors deployed in the building within the preset period, and the difference between the maximum and minimum values ​​is calculated to obtain the temperature and humidity fluctuation value; the air quality monitoring equipment deployed in the building area monitors the initial and final air quality index within the preset period, and the ratio of the difference between the final air quality index and the initial air quality index to the initial air quality index is recorded as the air quality index change rate.

[0042] The constructed multivariate nonlinear mapping function is solved using the gradient descent method. During the solution process, the goal is to minimize the prediction errors of equipment operating efficiency and environmental indicator changes. The optimal solution is gradually approximated through iterative optimization: First, the changes in the Y and Z axes are fixed as baseline constants, and only the X-axis is perturbed. The resulting changes in equipment operating efficiency (e.g., energy consumption, failure rate) and environmental indicator changes (e.g., temperature and humidity fluctuations, air quality index change rate) are calculated, yielding the independent influence coefficients of the X-axis on the two dependent variables. Similarly, the other two coordinate axes are fixed, and the independent influence coefficients of the Y and Z axes are solved sequentially. Based on this, the synergistic effect of simultaneous changes in the two spatial coordinate dimensions is further analyzed. By calculating the comprehensive change magnitude of the dependent variable under the linked perturbation, the cross-influence coefficients of different spatial dimension combinations on equipment and the environment are obtained. The specific calculation method is as follows: For each combination of spatial dimensions (e.g., X+Y, X+Z, Y+Z), within a preset disturbance range (e.g., a design deviation range of ±5% for each coordinate axis; in practical applications, the values ​​corresponding to the design deviation range can be fine-tuned according to the installation accuracy requirements of building equipment and the actual adjustment margin of the spatial layout), the changes in the corresponding coordinate axes are adjusted according to a pre-set equal step gradient (e.g., a deviation increment of 0.5% per step; if the data acquisition accuracy of building equipment for spatial changes is limited, the magnitude of the corresponding deviation increment can also be increased or decreased). Real-time data of equipment operating efficiency (energy consumption, failure rate) and environmental index changes (temperature and humidity fluctuations, air quality index change rate) are recorded simultaneously. After collecting the actual change values ​​of the dependent variable, the comprehensive change magnitude of the dependent variable under the combined dimension linkage disturbance is calculated, and then the difference is calculated with the sum of the change magnitudes when a single dimension is disturbed independently. The deviation value between the comprehensive change magnitude and the sum of the change magnitudes of the single dimension disturbance is calculated, and this deviation value is used as the core quantitative basis for the cross-influence coefficient.

[0043] Finally, all independent and cross-influence coefficients are normalized, outlier removed, and data smoothed. A single quantitative parameter, the three-dimensional linkage coupling coefficient, is synthesized through vector normalization. The specific steps of vector normalization are as follows: First, based on the weight of equipment operating efficiency (60%, determined by the priority of energy consumption and failure in building management) and the weight of environmental indicator change (40%, referring to the weight of environmental indicators in the green building evaluation standard), the independent and cross-influence coefficients are weighted and assigned. Then, all the weighted coefficients are included in the same vector space, and the final three-dimensional linkage coupling coefficient is obtained through vector magnitude normalization. The coefficient ranges from [0,1]. The higher the value, the stronger the linkage influence of the corresponding spatial layout change on equipment operation and environmental status, thereby comprehensively quantifying the complex two-way linkage relationship between spatial layout change and equipment operation and environmental status.

[0044] like Figure 5 The trend chart of independent influence coefficients clearly shows the dynamic changes of the independent influence coefficients of the X, Y, and Z axes during the iteration process: During the iteration count from 0 to 50, the X-axis coefficient (grey curve) rapidly decreased from approximately 0.5 initially, dropping below 0.1 after 10 iterations, and eventually approaching 0.05; the Y-axis coefficient (red curve) fluctuated and decreased from 0.5, stabilizing at around 0.1 after 20 iterations; and the Z-axis coefficient (blue curve) slowly declined from 0.5, maintaining around 0.25 after 50 iterations. All three entered a stable range after 40 iterations, indicating that the gradient descent method has driven the influence coefficients of each dimension to converge to a stable state, providing a reliable quantitative basis for the calculation of the three-dimensional linkage coupling coefficient.

[0045] For unified data across the entire domain with a 3D linkage coupling coefficient greater than a preset 3D linkage coupling coefficient, a linkage coupling anomaly alert is sent. The preset 3D linkage coupling coefficient is represented by the sum and average of historical 3D linkage coupling coefficients during the historical correlation verification process. For unified data across the entire domain with a 3D linkage coupling coefficient not greater than the preset 3D linkage coupling coefficient, a temporal interpolation completion and deviation correction algorithm is first used as a benchmark, based on a preset timestamp (e.g., 5-minute interval), to complete the temporal alignment of equipment operation, environmental perception, and spatial coordinate data, and synchronously associate the corresponding 3D linkage coupling coefficients, forming a spatiotemporal-coupling coefficient associated data unit containing spatiotemporal dimension information and a quantitative index of coupling strength. Subsequently, according to the X / Y / Z axis hierarchical order of the urban renewal area's spatial coordinates (e.g., area-building-floor-functional zone), all associated data units are arranged in a 3D matrix. The matrix rows, columns, and depth correspond to the X / Y / Z axis spatial coordinates, respectively, and the matrix elements are the core parameters of the associated data units, ultimately generating structured 3D coupling data.

[0046] The process of scenario compliance screening involves: retrieving correlation factors from the three categories of space, equipment, and environment in the 3D coupled data; normalizing all correlation factors; using the normalized correlation factors as the row and column dimensions of the matrix; and constructing an n×n dimensional feature correlation matrix with a total number of correlation factors of n. The quantified data of each correlation factor, such as the normalized value of the X-axis coordinate offset (0.62), the normalized value of the energy consumption fluctuation coefficient (0.48), and the normalized value of the temperature and humidity change correlation value (0.73), are then filled into the matrix according to their corresponding row and column positions to ensure that each element accurately corresponds to the sample data set of two correlation factors, forming a structured feature correlation matrix. The linear correlation between any two correlation factors in the feature correlation matrix is ​​calculated using the Pearson correlation coefficient formula, and the Pearson correlation coefficient between each correlation factor is output. The Pearson correlation coefficient ranges from [-1, 1], with positive and negative values ​​representing the direction of correlation and the absolute value reflecting the strength of correlation.

[0047] If the Pearson correlation coefficients of all associated factors are within the preset Pearson correlation coefficient range (i.e., a closed interval between -0.6 and 0.6), which is usually determined by a combination of correlation analysis results of historical coupling data of urban renewal areas and industry data correlation strength standards, then the verification is considered successful, and the three-dimensional coupling data after verification is obtained. Otherwise, the verification is considered unsuccessful, and the corresponding associated factors are marked as abnormal associations. At the same time, based on the obtained Pearson correlation coefficient deviation, that is, the absolute value of the difference between the obtained Pearson correlation coefficient and the preset Pearson correlation coefficient range boundary value (0.6 if it exceeds the upper limit, and -0.6 if it is below the lower limit), the abnormal associations are gradient corrected in combination with the constructed feature correction function. That is, the Pearson correlation coefficient deviation is used as the input of the feature correction function, and the feature correction function reduces the weight of the corresponding associated factor according to the corresponding proportion of the Pearson correlation coefficient deviation, and the weight is improved through gradient iterative optimization.

[0048] After gradient correction, if the Pearson correlation coefficients of all reacquired correlation factors are within the preset Pearson correlation coefficient range, the verification is considered successful, and the verified 3D coupling data is obtained. If the verification fails after a preset number of corrections (usually set to 3), an invalid correction prompt for abnormal correlation items is sent. The verified 3D coupling data is matched item by item according to the management requirements list of the specified urban renewal area, including constraints such as spatial layout specifications and equipment operation and maintenance standards. If all 3D coupling data meet the corresponding constraints in the management requirements list, the scenario compliance screening is considered successful, and decision dimension fusion processing is performed. Otherwise, the corresponding 3D coupling data is marked as a compliance conflict item, and a scenario conflict prompt is sent.

[0049] In this embodiment, consistency verification and real-time index evaluation based on CAD drawings ensure the accuracy and timeliness of unified data across the entire domain. The three-dimensional linkage coupling coefficient is solved using a multivariate nonlinear mapping function and gradient descent method, accurately quantifying the bidirectional linkage relationship between spatial layout and equipment / environment. Structured three-dimensional coupling data is generated through time-series alignment and three-dimensional matrix processing, coupled with Pearson correlation coefficient verification and gradient correction mechanisms, effectively ensuring the rationality of data association and scenario adaptability. Compliance screening is performed using the urban renewal area management needs list, accurately identifying conflict items and triggering prompts, providing decision support for spatial optimization, equipment operation and maintenance, and environmental control, achieving standardized processing of multimodal data across the entire chain from collection and verification to coupled application.

[0050] like Figure 6The flowchart shown illustrates the data acquisition and compliance review process for decision-making dimension fusion. The data acquisition method for decision-making dimension fusion is as follows: Using the three-dimensional coupled data after scenario compliance screening as the sample set, the dynamic change intervals of each dimension's data (such as spatial coordinate offset, equipment energy consumption, and ambient temperature and humidity) in the sample set are first divided. The median point is calculated based on the corresponding interval boundary values, and the derivative of the median point is calculated using a numerical differentiation algorithm to characterize the local rate of data change. Then, using this derivative as a weight, the min-max standardization method is combined to eliminate differences in dimensions and distributions, linearly mapping each dimension's data to the [0,1] interval. Finally, through spatial coordinate alignment and feature dimension normalization, the standardized data is mapped to a unified decision space.

[0051] By limiting the convergence range of each dimension of data through boundary constraint method, the data of each dimension approaches the optimal solution for the dynamic management needs of the specified urban renewal area. This allows for the selection of the optimal mapping results of each dimension of data in the unified decision space that meet the requirements of spatial correlation constraints, equipment operation logic, and environmental adaptation, and these results are recorded as preliminary fusion data. The iterative optimization process of the preliminary fusion data is monitored, and the absolute difference between the optimization results of two adjacent iterations is calculated. That is, the absolute value of the difference between the corresponding dimension data of the preliminary fusion data output by the k-th iteration and the preliminary fusion data output by the (k-1)-th iteration, to reflect the change in the step size of the iterative optimization.

[0052] The iterative optimization process uses initially fused data as the starting point and aims to achieve the optimal solution based on the dynamic management needs of urban renewal areas. It continuously corrects the data's mapping position in the unified decision space using a gradient descent algorithm. This example monitors the convergence of the iterative optimization by calculating the absolute difference between each dimension of data in two adjacent iterations to obtain the trend of the iteration step size, thereby determining whether the data is approaching the optimal solution. When the absolute difference is less than a preset absolute difference, the iterative optimization stops, and fused decision dimension data that satisfies decision consistency and adaptability to dynamic needs is output. When the absolute difference is not less than the preset absolute difference, iterative optimization continues, and monitoring continues. The preset absolute difference is represented by the sum and average of historical absolute differences during the iterative optimization process.

[0053] The specific steps for compliance review are as follows: First, obtain decision dimension fusion data that satisfies both decision consistency and dynamic demand adaptability. Second, perform bidirectional benchmarking with pre-set CAD drawing standard parameters. Specifically, calculate the Euclidean distance between the spatial coordinate mapping values ​​in the decision dimension fusion data and the pre-set CAD drawing standard parameters to obtain a deviation quantification value used to quantify the degree of deviation between the spatial coordinate mapping values ​​and the CAD drawing standard parameters. If the deviation quantification value is less than the preset deviation quantification value, the compliance review is deemed passed, and the final decision dimension fusion data is output, providing accurate data support for CAD drawing optimization and adjustment and the implementation of dynamic management in designated urban renewal areas. If the deviation quantification value is not less than the preset deviation quantification value, the corresponding spatial coordinate mapping value is marked as a review conflict item, and a review anomaly prompt is sent to remind designated personnel to check the review conflict item.

[0054] Specifically, the spatial coordinate mapping values ​​include the spatial coordinate X-axis mapping value, spatial coordinate Y-axis mapping value, and spatial coordinate Z-axis mapping value. The pre-set CAD drawing standard parameters include the spatial coordinate X-axis reference value, spatial coordinate Y-axis reference value, and spatial coordinate Z-axis reference value. The pre-set CAD drawing standard parameters are usually based on the planning and design specifications of the historical urban renewal area, the building structure safety standards, the equipment installation engineering technical requirements, and the verification results of historical project acceptance data. The preset deviation quantification value is represented by the sum and average of the historical deviation quantification values ​​in the historical decision-making dimension fusion process.

[0055] In this embodiment, the dynamic interval median mapping method combined with min-max standardization effectively eliminates the differences in the dimensions and distribution of multi-dimensional data, achieving accurate mapping of data to a unified decision space and ensuring data consistency. By employing the boundary constraint method and iterative convergence monitoring mechanism, and using the average of historical iteration absolute differences as a preset threshold, the data is driven to converge toward the optimal solution for dynamic management of the urban renewal area. Preliminary fusion data that meets the constraints of space, equipment, and environment across all dimensions is selected, improving data adaptability. Through comparison and verification with the Euclidean distance of standard parameters in CAD drawings, and using the average of historical deviations as the judgment benchmark, spatial coordinate mapping conflicts are accurately identified and prompts are triggered, ensuring the final data is compliant and reliable. This provides high-precision data support for CAD drawing optimization and adjustment, equipment installation calibration, intelligent operation and maintenance of the area, and environmental control.

[0056] This invention provides a multimodal fusion data processing system based on AI modeling, such as... Figure 7The diagram shows the structure of a multimodal fusion data processing system based on AI modeling. This system may include: a multimodal data processing and compliance screening module, used to collaboratively fuse collected multimodal data to obtain unified data across the entire domain, and perform data consistency verification and compliance screening based on an AI-driven data consistency verification model; an adaptability verification and scenario compliance screening module, used to perform requirement adaptability verification on the unified data across the entire domain after compliance screening, and generate three-dimensional coupled data using a specified AI algorithm for scenario compliance screening; and a decision dimension fusion processing and compliance review module, used to perform decision dimension fusion processing based on the three-dimensional coupled data after scenario compliance screening, combined with a decision fusion algorithm to obtain decision dimension fusion data, and perform compliance review.

[0057] In this embodiment, multimodal data collaborative fusion and AI-driven consistency verification and compliance screening quickly eliminate abnormal data, ensuring the integrity and accuracy of unified data across the entire domain. Demand adaptability verification and the generation of 3D coupled data using a specified AI algorithm, combined with a scenario compliance screening mechanism, achieve precise matching between data and the needs of urban renewal area scenarios, improving data scenario adaptability. Dimensional fusion is completed based on a decision fusion algorithm, coupled with a compliance review process, ensuring the compliance and reliability of decision-making dimension fusion data. These three modules form a standardized end-to-end processing system encompassing data preprocessing, scenario adaptation, and decision output, providing scientific data support for CAD drawing optimization and dynamic area management, improving the efficiency of data processing and the accuracy of decision-making in urban renewal scenarios.

[0058] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0059] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0060] In various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0061] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0062] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multimodal fusion data processing method based on AI modeling, characterized in that, The method includes: Step 1: Based on the CAD drawings of the designated urban renewal area, the collected multimodal data is collaboratively fused to obtain unified data across the entire area, and data consistency verification and initial compliance screening are performed based on an AI-driven data consistency verification model. Step 2: After the initial compliance screening is passed, the unified data across the entire domain is verified for demand adaptability, and three-dimensional coupled data is generated by combining the specified AI algorithm to screen for scenario compliance. Step 3: Based on the three-dimensional coupled data that has passed the scenario compliance screening, the decision dimension fusion processing is carried out in combination with the decision fusion algorithm to obtain the decision dimension fusion data, and then compliance review is performed.

2. The multimodal fusion data processing method based on AI modeling as described in claim 1, characterized in that, The specific process for performing collaborative fusion processing is as follows: After data cleaning, the multimodal data is classified and grouped to obtain groups to be collaboratively fused and processed. The multimodal data includes CAD drawing data, building equipment operation data, and environmental perception data. The spatial overlap, temporal synchronization rate, and semantic fit of each group to be collaboratively integrated are obtained respectively. The spatial overlap is calculated by the ratio of the total amount of building equipment operation data and environmental perception data within the spatial coordinate range corresponding to the CAD drawing data to the maximum allowable proportion within the spatial coordinate range. The temporal synchronization rate is obtained by the ratio of the number of timestamps of equipment operation sequence data and environmental perception data collection are aligned. The semantic fit is obtained by extracting the engineering semantics of CAD drawings and the business semantics of measured data to construct feature vectors and calculating the cosine similarity. For a certain group to be collaboratively integrated, data that does not meet the preset conditions is recorded as invalid data, and data that meets the preset conditions is recorded as valid data. Invalid data in each group to be collaboratively integrated is removed, and valid data is retained. The preset conditions mean that the spatial overlap, temporal synchronization rate and semantic fit are all greater than the corresponding reference values. The process of full-domain data collaboration and fusion is executed based on the retained valid data.

3. The multimodal fusion data processing method based on AI modeling as described in claim 2, characterized in that, The process of executing the full-domain data collaborative fusion is as follows: Based on the spatial zoning boundaries of the designated urban renewal areas, the valid data retained in each group to be collaboratively integrated and processed are spatially classified. Based on preset timestamps, the timestamps of building equipment operation data and environmental perception data within the same spatial partition are aligned. Based on the obtained spatial overlap, only multimodal data with spatial overlap not less than the reference spatial overlap in the timestamp-aligned multimodal data within the same spatial partition are retained to form a spatial association dataset; If the field corresponding to the coordinate parameter of the target building equipment in the spatial association dataset is empty, it is determined that there is missing data in the spatial association dataset, and the statistical mean of the same type of data in the same spatial partition is extracted to fill it; If the field corresponding to the coordinate parameters of the target building equipment in the spatial association dataset is a record value that has been formatted and validated, but the data acquisition frequency within the same time period is greater than the data acquisition frequency pre-marked in the CAD drawing, then it is determined that there is a data conflict in the spatial association dataset, and correction is made based on the spatial reference information marked in the CAD drawing. After completing the data missing filling and data conflict correction, the spatial association datasets of each spatial partition are spliced ​​together region by region to generate unified data across the entire domain that includes three dimensions: spatial geometry, equipment operation, and environmental status, and has a unified data format.

4. The multimodal fusion data processing method based on AI modeling as described in claim 3, characterized in that, The specific process for performing data consistency verification and initial compliance screening is as follows: Obtain the divided verification units, retrieve the CAD drawing design parameters from the unified data of the whole domain in each verification unit, and the currently collected building equipment operation data and environmental perception data to form a two-dimensional verification dataset; Call the existing data consistency verification model, and use the CAD drawing design parameters as a benchmark to calculate the consistency deviation value between the corresponding data of each verification unit in the dual-dimensional verification dataset. Within a preset monitoring period, the data real-time index of each verification unit is obtained by integrating the data transmission time of building equipment operation and the data collection time of environmental perception in each verification unit. If the real-time index of the acquired data is not lower than the preset real-time index and the consistency deviation value is not higher than the preset consistency deviation value, then the corresponding verification unit will be marked as a qualified data unit, and the unified data of the whole domain after the compliance screening is qualified will be acquired and the requirement adaptability verification will be performed. Otherwise, the corresponding verification unit will be marked as an abnormal data unit, and an exception handling prompt will be sent.

5. The multimodal fusion data processing method based on AI modeling as described in claim 4, characterized in that, The process of obtaining unified data across the entire domain after passing the initial compliance screening and performing requirement adaptability verification specifically involves: Retrieve the list of data items reflecting unified data across the entire region, and calculate the ratio of the actual number of data items already included in the unified data across the entire region to the total number of expected data items in the list to obtain the dimensional coverage completeness of various application scenarios in the designated urban renewal area. If the dimension coverage completeness is lower than the preset dimension coverage completeness, the requirement adaptability verification result is defined as the first adaptability anomaly label. If the dimension coverage completeness is not lower than the preset dimension coverage completeness, the effective data volume of each application scenario covered in the unified data of the whole domain is counted, and the ratio is calculated with the total data volume of the data items required by the corresponding application scenario to obtain the application requirement satisfaction index. The effective data volume of the data item represents the number of actual data items that meet the actual usage requirements of each application scenario among the data items of each application scenario that have been included in the unified data of the whole domain. If the application requirements of the unified data across the entire domain do not meet the performance requirements within the acceptable range, the requirement adaptability verification result is defined as the second adaptation anomaly label; otherwise, the requirement adaptability verification result is defined as adaptability feasible, and three-dimensional coupled data is generated in conjunction with the specified AI algorithm.

6. The multimodal fusion data processing method based on AI modeling as described in claim 5, characterized in that, The specific process for generating the three-dimensional coupled data is as follows: Perform dimensional correlation verification on the adaptable and feasible unified data across the entire domain. The correlation verification includes checking the matching of the installation location between spatial geometric data and equipment operating parameters, and the logical consistency of the influence of equipment operating parameters and environmental perception indicators. The dimensional correlation verification is performed as follows: A multivariate nonlinear mapping function is constructed with building equipment operating efficiency and environmental index changes as dependent variables and spatial coordinate changes as independent variables. The first-order partial derivatives of the constructed multivariate nonlinear mapping function with respect to the spatial coordinate changes are obtained by using the gradient descent method. The independent influence coefficients and cross-influence coefficients of each spatial coordinate dimension on the changes in equipment operating efficiency and environmental indicators are obtained, and the three-dimensional linkage coupling coefficient is obtained by summing them up. For unified data across the entire domain with a 3D linkage coupling coefficient not greater than a preset 3D linkage coupling coefficient, time-series alignment and fusion processing is performed to form spatiotemporal-coupling coefficient associated data units. The formed spatiotemporal-coupling coefficient associated data units are then processed into 3D matrices according to the spatial coordinate order of the entire urban renewal area to obtain 3D coupling data containing the 3D linkage relationship of space, equipment, and environment, as well as quantified coupling parameters. For unified data across the entire domain where the 3D linkage coupling coefficient is greater than the preset 3D linkage coupling coefficient, a linkage coupling abnormality prompt is sent.

7. The multimodal fusion data processing method based on AI modeling as described in claim 6, characterized in that, The scenario compliance screening is specifically performed as follows: Retrieve the correlation factors of space, equipment and environment from the three-dimensional coupled data, construct the feature correlation matrix, and calculate the Pearson correlation coefficient between each correlation factor; If the Pearson correlation coefficients of all correlation factors are within the preset Pearson correlation coefficient range, the verification is deemed successful, and the three-dimensional coupling data after verification is obtained. Otherwise, the verification is deemed unsuccessful, and the corresponding correlation factor is marked as an abnormal correlation item. Based on the obtained Pearson correlation coefficient deviation, the abnormal correlation item is graded and corrected using the constructed feature correction function. After gradient correction, if the Pearson correlation coefficients of all reacquired correlation factors are within the preset Pearson correlation coefficient range, the verification is considered successful and three-dimensional coupling data is acquired. If the verification is still considered unsuccessful after the preset number of corrections, an invalid correction prompt for abnormal correlation items is sent. The verified 3D coupled data is matched item by item with the management requirements list of the designated urban renewal area. If all 3D coupled data meet the corresponding constraints in the management requirements list, the scenario compliance screening is deemed to have passed and decision dimension fusion processing is performed. Otherwise, the corresponding 3D coupled data is marked as a compliance conflict item and a scenario conflict prompt is sent.

8. The multimodal fusion data processing method based on AI modeling as described in claim 7, characterized in that, The data fusion method for the decision-making dimensions is as follows: Using the three-dimensional coupled data that has passed the scenario compliance screening as a sample set, the median derivative of the change intervals corresponding to the data of each dimension in the sample set is solved. After standardizing and transforming the data of each dimension with different scales and distributions, it is mapped to a unified decision space. By limiting the convergence range of each dimension of data through the boundary constraint method, the data of each dimension approaches the optimal solution of the dynamic management needs of the specified urban renewal area, so as to select the optimal mapping result of each dimension of data in the unified decision space that meets the requirements of spatial correlation constraints, equipment operation logic and environmental adaptation, and record it as the preliminary fusion data. The iterative optimization process of the initial fused data is monitored, and the absolute difference between two adjacent iterations is calculated. When the absolute difference is less than the preset absolute difference, stop iterative optimization and output decision dimension fusion data that satisfies decision consistency and dynamic demand adaptability; When the absolute difference is not less than the preset absolute difference, continue iterative optimization and monitoring.

9. The multimodal fusion data processing method based on AI modeling as described in claim 8, characterized in that, The specific steps for conducting the compliance review are as follows: Acquire decision-dimensional fusion data that satisfies both decision consistency and dynamic demand adaptability, and perform bidirectional benchmarking with pre-defined CAD drawing standard parameters, specifically: Euclidean distance is calculated between the spatial coordinate mapping values ​​in the decision dimension fusion data and the pre-set standard parameters of CAD drawings to obtain the deviation quantification value; The spatial coordinate mapping values ​​include spatial coordinate X-axis mapping values, spatial coordinate Y-axis mapping values, and spatial coordinate Z-axis mapping values. The pre-set CAD drawing standard parameters include spatial coordinate X-axis reference values, spatial coordinate Y-axis reference values, and spatial coordinate Z-axis reference values. If the deviation quantification value is less than the preset deviation quantification value, the compliance review is deemed to have passed, and the final decision dimension fusion data is output. If the deviation quantization value is not less than the preset deviation quantization value, the corresponding spatial coordinate mapping value will be marked as a review conflict item, and a review exception prompt will be sent.

10. A multimodal fusion data processing system based on AI modeling, employing the multimodal fusion data processing method based on AI modeling as described in any one of claims 1-9, characterized in that, include: The multimodal data processing and compliance screening module is used to collaboratively integrate and process the collected multimodal data to obtain unified data across the entire domain, and to perform data consistency verification and compliance screening based on an AI-driven data consistency verification model. The adaptability verification and scenario compliance screening module is used to verify the adaptability of the unified data across the entire domain after the initial compliance screening, and to generate three-dimensional coupled data in combination with a specified AI algorithm for scenario compliance screening. The Decision Dimension Fusion Processing and Compliance Review Module is used to perform decision dimension fusion processing based on the three-dimensional coupled data after the scenario compliance screening is passed, combined with the decision fusion algorithm to obtain decision dimension fusion data, and to perform compliance review.

Images