Land space use review method and system based on deep learning and big data
By using deep learning and big data technologies, an integrated above-ground and underground land use review system has been built, which solves the problem of difficulty in identifying underground violations under the traditional regulatory model, and achieves efficient identification of underground space violations and full-space coverage of regulatory effects.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING TECH UNIV
- Filing Date
- 2026-06-12
- Publication Date
- 2026-07-14
AI Technical Summary
Traditional regulatory models struggle to penetrate the physical boundaries of buildings to achieve integrated identification of above-ground and underground spaces, making it difficult to effectively identify violations in hidden underground spaces and resulting in insufficient urban and rural planning control.
By using deep learning and big data methods, an integrated above-ground and underground land use review system is constructed. It utilizes building physical registration data and municipal metabolic characteristic data to extract above-ground and underground behavioral characteristics, conduct violation scoring and judgment, and generate land use review reports.
It has enabled efficient identification of violations in underground hidden spaces, improved the full spatial coverage and regulatory effectiveness of urban and rural planning, and enhanced the efficiency and accuracy of detecting violations.
Smart Images
Figure CN122390508A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of urban and rural planning information technology, specifically to a method and system for reviewing land use based on deep learning and big data. Background Technology
[0002] Territorial spatial planning is a fundamental system for sustainable urban and rural development, with land use control as its core means, requiring full life-cycle supervision of construction activities on urban and rural construction land. However, with the acceleration of urban renewal and stock development, some construction entities circumvent planning control through layered operations of "compliant above-ground, non-compliant underground," arbitrarily converting underground spaces registered as parking garages, equipment rooms, or warehouses into commercial establishments such as restaurants, entertainment venues, and accommodations. Such behavior not only violates the land use and development intensity requirements determined by the detailed control plan and disrupts the overall coordination of urban and rural spatial functions, but also poses a potential threat to urban and rural public safety and livability because it is carried out without the supporting reviews of fire protection, environmental protection, and sanitation required by planning permits. The traditional regulatory model, which relies on on-site inspections and document verification, is insufficient to penetrate the physical boundaries of buildings and achieve integrated identification of above-ground and underground spaces, urgently requiring intelligent technological support adapted to the urban and rural planning system.
[0003] The rapid development of big data and artificial intelligence technologies has provided new tools for refined governance in urban and rural planning. Metabolic data such as time-of-use electricity and drainage generated during the operation of urban infrastructure objectively reflect the intensity of actual activities and business patterns within buildings. Deep learning models can automatically extract behavioral features from multi-source heterogeneous data, enabling intelligent identification of usage patterns. However, existing research largely focuses on single-dimensional analysis of visible features above ground, and has not yet established a technical framework consistent with the hierarchical management logic of urban and rural planning—that is, how to separate the metabolic contributions of compliant above-ground activities based on the registered use and area parameters of planning permits, and thus identify the independent activity fingerprints of hidden underground spaces, remains a key bottleneck restricting the effectiveness of off-site supervision.
[0004] Therefore, this invention proposes a land use review method and system based on deep learning and big data, which enables intelligent identification of integrated above-ground and underground violations and provides a new data-driven technical tool for the supervision of urban and rural planning implementation. Summary of the Invention
[0005] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a method and system for reviewing land use based on deep learning and big data, which can effectively solve the problems mentioned in the existing technology.
[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a method for reviewing land use based on deep learning and big data, comprising the following steps: The system acquires above-ground behavioral characteristic data of the target building, municipal metabolic characteristic data of the municipal access node corresponding to the target building, and building physical registration data. The above-ground behavioral characteristic data is used to reflect the actual activity status of the above-ground floors of the target building. The municipal metabolic characteristic data includes time-of-use drainage flow data and time-of-use electricity consumption data. The building physical registration data includes the registered area, registered use, and basement area data of each floor. Based on the building physics registration data and the preset use metabolic parameter library, the time series of the theoretical metabolic baseline amount of the above-ground layer is calculated; the preset use metabolic parameter library stores the empirical values and time series distribution patterns of the standard metabolic amount per unit area and per unit time for each use type. The metabolic characteristic data of each time unit is subtracted from the benchmark quantity of the corresponding time unit in the theoretical metabolic benchmark quantity time series of the above-ground layer to obtain the metabolic residual quantity time series; the metabolic residual quantity time series represents the load contribution of the actual activities in the underground space to the municipal access node after deducting the metabolic contribution of the compliant activities in the above-ground layer. The peak time period, peak-to-valley ratio, day-night ratio, and weekend-to-weekday ratio are extracted from the metabolic residual time series to form a residual metabolic feature vector; the residual metabolic feature vector is matched with a preset commercial activity metabolic template library to obtain the underground activity type matching result and the underground violation score. The above-ground behavior feature data is input into a pre-trained above-ground activity intensity classification model to obtain above-ground layer violation scores; the above-ground layer violation scores characterize the weighted deviation of the above-ground behavior feature data from the benchmark value of the registered use of the target building. The above-ground violation score is compared with a preset above-ground violation threshold, and the underground violation score is compared with a preset underground violation threshold. When both exceed their respective preset thresholds, a composite violation judgment result is generated. When only the underground violation score exceeds the preset underground violation threshold, a hidden underground violation judgment result is generated. When only the above-ground violation score exceeds the preset above-ground violation threshold, an above-ground violation judgment result is generated. Based on the judgment results, the above-ground behavioral characteristic data, and the matching results of the underground activity types, a usage review report is generated, which includes violation nature classification labels, violation spatial layer attribution, inference of underground activity business type, and multi-dimensional evidence chain data.
[0007] Furthermore, in the step of calculating the time series of the theoretical metabolic baseline quantity of the aboveground layer, the time series of the theoretical metabolic baseline quantity of the aboveground layer is calculated according to the following formula: ; in, The theoretical metabolic baseline amount of the aboveground layer at time t; The total number of registered floors above ground in the target building; Let be the registered area of the i-th floor; The standard metabolic rate per unit area at time t is the registered use of the i-th floor retrieved from the preset use metabolic parameter library. The temporal distribution is determined based on the weekday, weekend, and holiday patterns corresponding to the registered purpose of the floor. The time-sharing drainage baseline and time-sharing electricity consumption baseline in the above-ground theoretical metabolic baseline time series are calculated independently according to the above formula, and are respectively compared with the time-sharing drainage flow data and time-sharing electricity consumption data in the municipal metabolic characteristic data.
[0008] Furthermore, after obtaining the time series of metabolic residuals, the method further includes: Abnormal interval detection is performed on the time series of the metabolic residual: if the time series of the metabolic residual is continuously negative within a preset duration, it is determined that the measured metabolic amount of the aboveground layer is lower than the theoretical metabolic baseline amount of the aboveground layer, an early warning mark of insufficient activity of the aboveground layer is generated, and the early warning mark is included in the use review report to trigger the review process of false registration of the aboveground layer. If the metabolic residual time series is close to zero, it is determined that the overall metabolism of the target building is consistent with the registered use, there is no obvious substantial activity in the underground layer, and the underground layer violation scoring calculation process is not triggered.
[0009] Furthermore, in the step of obtaining the above-ground layer violation score, the above-ground layer violation score is calculated according to the following formula: ; in, The aboveground layer is used to score violations; M is the total number of feature dimensions of the aboveground behavioral feature data; The preset weight coefficient is the j-th feature dimension, and the sum of all weight coefficients is 1. The preset weight coefficient is determined based on the historical statistical correlation strength between the data of each feature dimension and the illegal commercial use. Let be the measured value of the j-th feature dimension; The benchmark value is the j-th feature dimension corresponding to the registered use of the target building, and the benchmark value is obtained from a preset registered use behavior benchmark library; The After normalization, it is then included in the weighted summation. The normalization method is as follows: Standardize the difference in the denominator; when When the preset ground-level violation threshold is exceeded, the ground-level violation judgment process is triggered.
[0010] Furthermore, in the step of performing similarity matching between the residual metabolic feature vector and the preset commercial activity metabolic template library, the similarity matching is calculated using the cosine similarity formula: ; in, The residual metabolic feature vector V and the k-th business activity template vector are... Cosine similarity between them; It is the dot product of two vectors; and These are the Euclidean norms of the two vectors; the similarity range is [0,1], and the larger the value, the more closely the underground activities match the business format represented by the corresponding template; The maximum cosine similarity value among all templates is taken as the violation score of the underground layer, and the template type corresponding to the maximum value is taken as the matching result of the activity type of the underground layer; the preset commercial activity metabolism template library contains at least catering activity template vectors, entertainment activity template vectors and accommodation activity template vectors.
[0011] Furthermore, after obtaining the municipal metabolic characteristic data of the municipal access node corresponding to the target building, before calculating the time series of the theoretical metabolic baseline quantity of the above-ground layer, the method further includes: The municipal metabolic characteristic data is preprocessed by time-series decomposition, which decomposes the time-sharing drainage flow data and the time-sharing electricity consumption data into trend components, periodic components and random noise components, respectively. The trend components and periodic components are retained, and the random noise components are removed to eliminate the disturbance of metabolic volume caused by holidays, emergencies and metering fluctuations. The sum of the trend component and the periodic component after time-series decomposition preprocessing is used as the effective municipal metabolic feature data for the difference operation in subsequent steps.
[0012] Furthermore, the construction method of the preset business activity metabolism template library includes: Collect a preset number of time-series samples of metabolic residuals from benchmark buildings that are confirmed to be engaged in catering, entertainment, and accommodation activities. Extract the peak time period, peak-to-valley ratio, day-night ratio, and weekend-to-weekday ratio from each sample. Calculate the statistical mean of each dimension of the feature vector of all samples under the same business type to form a metabolic template vector for the corresponding business type. The preset number is not less than the preset minimum sample size threshold to ensure the statistical representativeness of the template vector; when the sample size is lower than the preset minimum sample size threshold, it is supplemented by interpolation of template vectors of adjacent business types until the preset minimum sample size threshold is met.
[0013] Furthermore, after generating the composite violation determination result or the hidden underground violation determination result, the process also includes: Based on the business type in the matching results of the underground activity type, the empirical value of the standard metabolic rate per unit area for that business type is obtained from the preset metabolic parameter library. The estimated area of illegal activities in underground space is obtained by dividing the peak metabolic rate in the time series of the metabolic residual by the empirical value of the standard metabolic rate per unit area. The estimated area is compared with the basement area data in the building physical registration data to obtain the rate of illegal use of underground space; The estimated area of illegal activities in the underground space and the rate of illegal use of the underground space will be included in the use review report.
[0014] A land use review system based on deep learning and big data includes: The data acquisition module is used to acquire above-ground behavioral characteristic data of the target building, municipal metabolic characteristic data of the municipal access node corresponding to the target building, and building physical registration data; the municipal metabolic characteristic data includes time-of-use drainage flow data and time-of-use electricity consumption data; the building physical registration data includes the registered area of each floor, registered use, and basement area data; The benchmark calculation module is used to calculate the theoretical metabolic benchmark amount time series of the above-ground layer based on the building physical registration data and the preset use metabolic parameter library; the preset use metabolic parameter library stores the standard metabolic amount empirical values and time series distribution patterns of each use type per unit area and per unit time. The residual extraction module is used to subtract the municipal metabolic characteristic data of each time unit from the benchmark quantity of the corresponding time unit in the time series of the theoretical metabolic benchmark quantity of the above-ground layer to obtain the metabolic residual quantity time series; the metabolic residual quantity time series represents the load contribution of the substantive activities in the underground space to the municipal access node after deducting the metabolic contribution of the compliant activities in the above-ground layer. The activity identification module is used to extract the intraday peak time period, peak-to-valley ratio, day-night ratio, and weekend-to-weekday ratio from the metabolic residual time series to form a residual metabolic feature vector; the residual metabolic feature vector is matched with a preset commercial activity metabolic template library to obtain the underground activity type matching result and the underground violation score. The scoring calculation module is used to input the above-ground behavior feature data into a pre-trained above-ground activity intensity classification model to obtain the above-ground layer violation score; the above-ground layer violation score represents the weighted deviation of the above-ground behavior feature data from the benchmark value of the registered use of the target building. The violation determination module is used to compare the violation score of the above-ground layer with a preset above-ground violation threshold and the violation score of the underground layer with a preset underground violation threshold. When both exceed their respective preset thresholds, a composite violation determination result is generated. When only the violation score of the underground layer exceeds the preset underground violation threshold, a hidden underground violation determination result is generated. When only the violation score of the above-ground layer exceeds the preset above-ground violation threshold, an above-ground violation determination result is generated. The report generation module is used to generate a usage review report based on the judgment results, the above-ground behavior characteristic data, and the underground activity type matching results. The report includes a violation nature classification label, the violation spatial layer attribution, the inference of the underground activity business type, and multi-dimensional evidence chain data.
[0015] A computer-readable storage medium having stored thereon computer program instructions, which, when executed by a processor, implement any of the land use review methods based on deep learning and big data.
[0016] The technical solution provided by this invention has the following advantages compared with the known prior art: This invention addresses the challenge of lacking direct observation methods for underground hidden spaces by constructing a residual extraction mechanism. It utilizes building physical registration data to isolate the municipal metabolic contributions of compliant above-ground activities, extracts feature vectors with business type differentiation from the residual time series, and performs similarity matching with a preset commercial activity metabolic template. This method can infer the type of underground activities without entering the building, providing a technical capability to penetrate the physical spatial layers for the supervision of land and space planning implementation, significantly improving the efficiency and accuracy of detecting hidden violations.
[0017] This invention innovatively establishes a dual-dimensional scoring mechanism based on above-ground behavioral characteristics and underground metabolic characteristics. Through independent calculation of above-ground and underground violation scores and comparison of dual thresholds, it generates spatial layer attribution results encompassing three categories: compound violations, concealed underground violations, and above-ground violations. This system overcomes the limitations of existing technologies that focus only on a single spatial layer, achieving holistic identification of complex circumvention behaviors and strengthening the full spatial coverage capability of urban and rural planning land use control. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.
[0019] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system structure of the present invention. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0021] The present invention will be further described below with reference to embodiments.
[0022] Example 1:
[0023] This embodiment provides a land use review method based on deep learning and big data. It is applicable to the intelligent identification and review of whether existing buildings within urban built-up areas violate land use planning regulations, particularly in cases where the registered above-ground use does not match the actual activity, and where underground space is concealed for illegal commercial operations. See also Figure 1 The method includes the following steps: First, obtain three types of basic data for the target building.
[0024] The above-ground behavioral characteristic data of the target building is acquired, reflecting the actual activity status of the above-ground floors. In this embodiment, the above-ground behavioral characteristic data can be obtained through multi-source sensing methods, including but not limited to: building facade signage text information obtained by OCR recognition based on street view images or drone aerial images; entrance and exit pedestrian flow time-series data obtained by video analysis or WiFi probes; vehicle entry and exit frequency and dwell time obtained by license plate recognition or geomagnetic sensors; nighttime light intensity index extracted from satellite remote sensing nighttime light data; and order popularity distribution obtained by aggregating POI data from food delivery platforms. The above characteristic data is acquired by accessing the city's big data platform or deploying IoT sensing devices, and aligned according to a unified time.
[0025] The municipal metabolic characteristic data of the municipal access node corresponding to the target building is obtained. This data includes time-of-use drainage flow data and time-of-use electricity consumption data. In this embodiment, the municipal access node refers to the pipeline network node or distribution transformer node that provides drainage and electricity services to the building. The correspondence between the building and the municipal access node is established through GIS spatial topology analysis: the building boundary vector data is spatially overlaid with the municipal drainage network topology and the power grid distribution topology. The pipeline access point or the nearest neighbor node is used as the corresponding municipal access node for the building, thereby assigning the node metering data to the target building.
[0026] Obtain the building physical registration data of the target building, including the registered area, registered use, and basement area of each floor. This data is obtained by accessing the real estate registration database, construction project planning permit database, and completion acceptance filing database. The registered use is standardized and coded according to the secondary category in the national standard "Guidelines for Classification of Land and Sea Use in Territorial Spatial Survey, Planning, and Use Control".
[0027] After acquiring municipal metabolic characteristic data, before calculating the time series of the theoretical metabolic baseline quantity of the above-ground layer, the municipal metabolic characteristic data is subjected to time series decomposition preprocessing. The time-sharing drainage flow data and time-sharing electricity consumption data are decomposed into trend components, periodic components and random noise components, respectively. The trend components and periodic components are retained, and the random noise components are removed to eliminate the disturbance of metabolic quantity caused by holidays, emergencies and metering fluctuations.
[0028] Specifically, in this embodiment, the STL decomposition method is used to achieve the above decomposition. The sum of the trend component and the periodic component after time-series decomposition preprocessing is used as the effective municipal metabolic feature data for the difference operation in subsequent steps. This can effectively eliminate abnormal disturbances without losing long-term trends and periodic patterns.
[0029] Based on building physics registration data and a pre-defined metabolic parameter library, the time series of theoretical metabolic baseline quantities for above-ground floors is calculated. The pre-defined metabolic parameter library stores empirical values and time series distribution patterns of standard metabolic quantities per unit area and per unit time for various building types. This parameter library is constructed based on: collecting historical sub-item metering data from compliant buildings of the same type in the local area, combining this data with electricity and water consumption intensity data by industry from the "Water Conservation Design Standard for Civil Buildings," the "Energy Conservation Design Standard for Public Buildings," and local statistical yearbooks; normalizing the data by unit area; and statistically obtaining the time series distribution curves of standard metabolic quantities per unit area by building type and by weekday / weekend / holiday pattern.
[0030] The theoretical metabolic baseline for the aboveground layer is calculated using the following formula: ; in, The theoretical metabolic baseline amount of the aboveground layer at time t; The total number of registered floors above ground in the target building; Let be the registered area of the i-th floor; This represents the standard metabolic rate per unit area at time t for the registered purpose of the i-th floor, retrieved from the pre-defined purpose metabolic parameter library. The temporal distribution is determined based on the weekday, weekend, and holiday patterns corresponding to the registered purpose of the floor.
[0031] It should be noted that the time-of-use drainage baseline and time-of-use electricity baseline in the theoretical metabolic baseline time series of the above-ground layer are calculated independently according to the above formula, each forming a time series curve, and are respectively compared with the time-of-use drainage flow data and time-of-use electricity consumption data in the municipal metabolic characteristic data.
[0032] The metabolic characteristic data of each time unit are subtracted from the baseline values of the corresponding time units in the theoretical metabolic baseline time series of the above-ground layer to obtain the metabolic residual time series. The metabolic residual time series represents the load contribution of substantive activities in the underground space to the municipal access nodes after deducting the metabolic contribution of compliant activities in the above-ground layer.
[0033] Let the measured municipal metabolic rate after time-series decomposition pretreatment be: Then the metabolic residual for: ; The drainage residual and the electricity residual are calculated independently. If there are unregistered commercial activities in the underground space, the actual water and electricity consumption will be reflected in the municipal metering data, resulting in a positive fluctuation of the residual with a specific pattern; conversely, if there are no actual activities in the underground space, the residual should fluctuate randomly and slightly around zero.
[0034] After obtaining the time series of metabolic residuals, anomaly detection is performed on the time series: if the time series of metabolic residuals remains negative for a continuously preset duration (e.g., 7 consecutive days), it is determined that the measured metabolic rate of the aboveground layer is lower than the theoretical metabolic baseline, generating an insufficient activity warning flag for the aboveground layer. This warning flag is then included in the land use review report to trigger the review process for false registration of aboveground layers. This situation may indicate that the registered use of the aboveground layer is for high-intensity commercial activities, while the actual activity intensity is significantly insufficient, raising suspicion of falsely reporting the use to cover up illegal underground activities.
[0035] If the metabolic residual time series is close to zero (e.g., the absolute value of the mean of the residual sequence is less than the preset threshold and the variance is less than the preset fluctuation threshold), then it is determined that the overall metabolism of the target building is consistent with the registered use, there is no obvious substantial activity in the underground layer, and the subsequent underground layer violation scoring calculation process is not triggered, thereby avoiding unnecessary computing power consumption.
[0036] The peak time period, peak-to-valley ratio, day-to-night ratio, and weekend-to-weekday ratio are extracted from the time series of metabolic residuals to construct the residual metabolic feature vector. The specific extraction methods for each feature are as follows: Intraday peak periods: Divide the residual time series into daily segments, calculate the time when the maximum residual occurs each day, and then take the mode or mean of the results over multiple days to obtain typical peak periods; Peak-to-valley ratio: The ratio of the maximum to the minimum value of the intraday residual, reflecting the intraday fluctuation range of activity intensity; Day-night ratio: The ratio of the mean residual during the daytime period (e.g., 08:00-20:00) to the mean residual during the nighttime period (e.g., 20:00-08:00 the next day). The intensity of daytime and nighttime activities varies significantly among different business types. Weekend to Weekday Ratio: This ratio is calculated to the average daily residual on weekends and the average daily residual on weekdays. It is used to distinguish between entertainment and accommodation activities that are less affected by weekdays and typical office activities.
[0037] The residual metabolic feature vectors are matched with a pre-defined commercial activity metabolic template library to obtain the matching results for underground activity types and underground violation scores. The similarity matching is calculated using the cosine similarity formula: ; in, Let V be the residual metabolic feature vector and the k-th business activity template vector. Cosine similarity between them; It is the dot product of two vectors; and These are the Euclidean norms of the two vectors; the similarity range is [0,1], and the larger the value, the more closely the underground activities match the business format represented by the corresponding template; The maximum cosine similarity value among all templates is used as the score for violations in the underground layer. The template type corresponding to the maximum value is taken as the matching result for the underground activity type. The preset commercial activity metabolism template library contains at least catering activity template vectors, entertainment activity template vectors, and accommodation activity template vectors.
[0038] The construction of the pre-set commercial activity metabolism template library involves the following steps: collecting a pre-set number (no less than the pre-set minimum sample size threshold, such as no less than 30 buildings per type of business) of time-series samples of metabolic residuals from confirmed benchmark buildings engaged in catering, entertainment, and accommodation activities; extracting the above-mentioned intraday peak hours, peak-to-valley ratio, day-to-night ratio, and weekend-to-weekday ratio values from each sample; calculating the statistical mean of each dimension for all sample feature vectors under the same business type to form the corresponding business type's metabolism template vector. The pre-set number is no less than the pre-set minimum sample size threshold to ensure the statistical representativeness of the template vectors; when the sample size is lower than the pre-set minimum sample size threshold, it is supplemented by interpolation using template vectors of adjacent business types until the pre-set minimum sample size threshold is met.
[0039] The above-ground behavior feature data is input into a pre-trained above-ground activity intensity classification model to obtain the above-ground layer violation score. The above-ground layer violation score represents the weighted deviation of the above-ground behavior feature data from the baseline value of the registered use of the target building. In this embodiment, the pre-trained above-ground activity intensity classification model is constructed using a gradient boosting decision tree (such as XGBoost or LightGBM) or a deep neural network. The training samples of the model are historically labeled compliant and non-compliant buildings, the input features are the above-mentioned above-ground behavior feature data, and the output is the violation probability or deviation. During the model inference phase, to enhance interpretability, the above-ground layer violation score is explicitly calculated according to the following formula: ; in, The above-ground layer is used to score violations; M represents the total number of feature dimensions for above-ground behavioral characteristic data. The preset weight coefficient is the j-th feature dimension, and the sum of all weight coefficients is 1. The preset weight coefficient is determined based on the historical statistical correlation strength between the data of each feature dimension and the illegal commercial use. Let be the measured value of the j-th feature dimension; This is the benchmark value corresponding to the j-th feature dimension under the registered use of the target building. The benchmark value is obtained from the preset registered use behavior benchmark library.
[0040] It should be noted that, After normalization, it is then included in the weighted summation. The normalization method is as follows: Standardizing the difference in the denominator makes the deviations of characteristics across different dimensions comparable. When the preset ground-level violation threshold is exceeded, the ground-level violation judgment process is triggered.
[0041] The system compares the violation scores of the above-ground level with preset above-ground violation thresholds, and the violation scores of the underground level with preset underground violation thresholds. A composite violation determination result is generated when both exceed their respective preset thresholds. A hidden underground violation determination result is generated when only the underground level violation score exceeds the preset underground violation threshold, and a above-ground violation determination result is generated when only the above-ground level violation score exceeds the preset above-ground violation threshold. These thresholds can be determined through ROC curve analysis of historical samples based on the regional regulatory accuracy requirements, for example, by taking the score cutoff point that maximizes the F1 score.
[0042] After generating composite violation determination results or hidden underground violation determination results, the process also includes: based on the business type in the underground activity type matching results, obtaining the empirical value of the standard metabolic rate per unit area for that business type from the preset use metabolic parameter library. The peak metabolic rate in the metabolic residual time series. By quoting the empirical value of standard metabolic rate per unit area, the estimated area of illegal activities in underground space is obtained. : ; Compare the estimated area with the basement area from the building physics registration data. By comparison, the rate of illegal use of underground space was obtained. : ; The estimated area of illegal activities in underground space and the rate of illegal use of underground space will be included in the land use review report.
[0043] Based on the judgment results, surface behavior characteristic data, and underground activity type matching results, a usage review report is generated, including violation nature classification labels, violation spatial layer attribution, underground activity type inference, and multi-dimensional evidence chain data. The multi-dimensional evidence chain data includes, but is not limited to: original municipal metabolic time-series curves, effective metabolic curves after decomposition and preprocessing, theoretical metabolic baseline time-series curves for surface layers, metabolic residual time-series curves, radar charts comparing residual feature vectors with matching templates, deviation comparison tables of measured surface behavior characteristics and baseline values, and calculation process data for estimated area and illegal utilization rate of underground space violations. The report is output in structured data format, providing natural resources law enforcement and supervision departments with clues for off-site supervision and on-site verification.
[0044] Example 2:
[0045] Reference Figure 2 This embodiment provides a land use review system corresponding to the method in Embodiment 1, including the following functional modules: The data acquisition module is used to acquire above-ground behavioral characteristic data of the target building, municipal metabolic characteristic data of the municipal access node corresponding to the target building, and building physical registration data. The municipal metabolic characteristic data includes time-of-use drainage flow data and time-of-use electricity consumption data. The building physical registration data includes the registered area, registered use, and basement area data of each floor.
[0046] The benchmark calculation module is used to calculate the time series of theoretical metabolic benchmarks for the above-ground floors based on building physical registration data and a preset metabolic parameter library. The preset metabolic parameter library stores the empirical values and time series distribution patterns of standard metabolic amounts per unit area and per unit time for each use type.
[0047] The residual extraction module is used to subtract the municipal metabolic characteristic data of each time unit from the benchmark quantity of the corresponding time unit in the theoretical metabolic benchmark quantity time series of the above-ground layer to obtain the metabolic residual quantity time series; the metabolic residual quantity time series represents the load contribution of the actual activities in the underground space to the municipal access node after deducting the metabolic contribution of the compliant activities in the above-ground layer.
[0048] The activity identification module is used to extract the intraday peak time, peak-to-valley ratio, day-night ratio, and weekend-to-weekday ratio from the metabolic residual time series to form a residual metabolic feature vector. The residual metabolic feature vector is then matched with a preset commercial activity metabolic template library to obtain the matching results of underground activity types and underground violation scores.
[0049] The scoring calculation module is used to input the above-ground behavior feature data into the pre-trained above-ground activity intensity classification model to obtain the above-ground layer violation score; the above-ground layer violation score represents the weighted deviation of the above-ground behavior feature data from the benchmark value of the registered use of the target building.
[0050] The violation determination module is used to compare the violation score of the above-ground layer with the preset above-ground violation threshold and the violation score of the underground layer with the preset underground violation threshold. When both exceed their respective preset thresholds, a composite violation determination result is generated. When only the underground layer violation score exceeds the preset underground violation threshold, a hidden underground violation determination result is generated. When only the above-ground layer violation score exceeds the preset above-ground violation threshold, an above-ground layer violation determination result is generated.
[0051] The report generation module is used to generate a usage review report based on the judgment results, surface behavior characteristic data, and underground activity type matching results. The report includes a classification label for the nature of the violation, the spatial layer attribution of the violation, the inference of the underground activity type, and multi-dimensional evidence chain data.
[0052] The specific implementation of each of the above modules refers to the corresponding steps in Example 1. Each module communicates with the other through a standard data interface and can be deployed on a government cloud or a private server cluster.
[0053] Example 3:
[0054] This embodiment provides a computer-readable storage medium storing computer program instructions. When the computer program instructions are executed by a processor, they implement the land use review method based on deep learning and big data as described in any of Embodiment 1.
[0055] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.
Claims
1. A land use review method based on deep learning and big data, characterized in that, Includes the following steps: The system acquires above-ground behavioral characteristic data of the target building, municipal metabolic characteristic data of the municipal access node corresponding to the target building, and building physical registration data. The above-ground behavioral characteristic data is used to reflect the actual activity status of the above-ground floors of the target building. The municipal metabolic characteristic data includes time-of-use drainage flow data and time-of-use electricity consumption data. The building physical registration data includes the registered area, registered use, and basement area data of each floor. Based on the building physics registration data and the preset use metabolic parameter library, the time series of the theoretical metabolic baseline amount of the above-ground layer is calculated; the preset use metabolic parameter library stores the empirical values and time series distribution patterns of the standard metabolic amount per unit area and per unit time for each use type. The metabolic residual time series is obtained by subtracting the municipal metabolic characteristic data of each time unit from the baseline data of the corresponding time unit in the theoretical metabolic baseline time series of the aboveground layer. The metabolic residual time series characterizes the load contribution of substantive activities in the underground space to the municipal access node after deducting the metabolic contribution of compliant activities in the above-ground layer. The peak time period, peak-to-valley ratio, day-night ratio, and weekend-to-weekday ratio are extracted from the metabolic residual time series to form a residual metabolic feature vector; the residual metabolic feature vector is matched with a preset commercial activity metabolic template library to obtain the underground activity type matching result and the underground violation score. The above-ground behavior feature data is input into a pre-trained above-ground activity intensity classification model to obtain above-ground layer violation scores; The above-ground violation score characterizes the weighted deviation of the above-ground behavioral feature data from the benchmark value of the registered use of the target building; The above-ground violation score is compared with a preset above-ground violation threshold, and the underground violation score is compared with a preset underground violation threshold. When both exceed their respective preset thresholds, a composite violation judgment result is generated. When only the underground violation score exceeds the preset underground violation threshold, a hidden underground violation judgment result is generated. When only the above-ground violation score exceeds the preset above-ground violation threshold, an above-ground violation judgment result is generated. Based on the judgment results, the above-ground behavioral characteristic data, and the matching results of the underground activity types, a usage review report is generated, which includes violation nature classification labels, violation spatial layer attribution, inference of underground activity business type, and multi-dimensional evidence chain data.
2. The land use review method based on deep learning and big data according to claim 1, characterized in that, In the step of calculating the time series of theoretical metabolic baseline quantities in the aboveground layer, the time series of theoretical metabolic baseline quantities in the aboveground layer is calculated according to the following formula: ； in, The theoretical metabolic baseline amount of the aboveground layer at time t; The total number of registered floors above ground in the target building; Let be the registered area of the i-th floor; The standard metabolic rate per unit area at time t is the registered use of the i-th floor retrieved from the preset use metabolic parameter library. The temporal distribution is determined based on the weekday, weekend, and holiday patterns corresponding to the registered purpose of the floor. The time-sharing drainage baseline and time-sharing electricity consumption baseline in the above-ground theoretical metabolic baseline time series are calculated independently according to the above formula, and are respectively compared with the time-sharing drainage flow data and time-sharing electricity consumption data in the municipal metabolic characteristic data.
3. The land use review method based on deep learning and big data according to claim 1, characterized in that, After obtaining the time series of metabolic residuals, the method further includes: Abnormal interval detection is performed on the time series of the metabolic residual: if the time series of the metabolic residual is continuously negative within a preset duration, it is determined that the measured metabolic amount of the aboveground layer is lower than the theoretical metabolic baseline amount of the aboveground layer, an early warning mark of insufficient activity of the aboveground layer is generated, and the early warning mark is included in the use review report to trigger the review process of false registration of the aboveground layer. If the metabolic residual time series is close to zero, it is determined that the overall metabolism of the target building is consistent with the registered use, there is no obvious substantial activity in the underground layer, and the underground layer violation scoring calculation process is not triggered.
4. The land use review method based on deep learning and big data according to claim 1, characterized in that, In the step of obtaining the above-ground layer violation score, the above-ground layer violation score is calculated according to the following formula: ； in, The aboveground layer is used to score violations; M is the total number of feature dimensions of the aboveground behavioral feature data; The preset weight coefficient is the j-th feature dimension, and the sum of all weight coefficients is 1. The preset weight coefficient is determined based on the historical statistical correlation strength between the data of each feature dimension and the illegal commercial use. Let be the measured value of the j-th feature dimension; The benchmark value is the j-th feature dimension corresponding to the registered use of the target building, and the benchmark value is obtained from a preset registered use behavior benchmark library; The After normalization, it is then included in the weighted summation. The normalization method is as follows: Standardize the difference in the denominator; when When the preset ground-level violation threshold is exceeded, the ground-level violation judgment process is triggered.
5. The land use review method based on deep learning and big data according to claim 1, characterized in that, In the step of matching the residual metabolic feature vector with a preset commercial activity metabolic template library, the similarity matching is calculated using the cosine similarity formula: ； in, The residual metabolic feature vector V and the k-th business activity template vector are... Cosine similarity between them; It is the dot product of two vectors; and These are the Euclidean norms of the two vectors; the similarity range is [0,1], and the larger the value, the more closely the underground activities match the business format represented by the corresponding template; The maximum cosine similarity value among all templates is taken as the violation score of the underground layer, and the template type corresponding to the maximum value is taken as the matching result of the activity type of the underground layer; the preset commercial activity metabolism template library contains at least catering activity template vectors, entertainment activity template vectors and accommodation activity template vectors.
6. The land use review method based on deep learning and big data according to claim 1, characterized in that, After obtaining the municipal metabolic characteristic data of the municipal access node corresponding to the target building, and before calculating the time series of the theoretical metabolic baseline quantity of the above-ground layer, the method further includes: The municipal metabolic characteristic data is preprocessed by time-series decomposition, which decomposes the time-sharing drainage flow data and the time-sharing electricity consumption data into trend components, periodic components and random noise components, respectively. The trend components and periodic components are retained, and the random noise components are removed to eliminate the disturbance of metabolic volume caused by holidays, emergencies and metering fluctuations. The sum of the trend component and the periodic component after time-series decomposition preprocessing is used as the effective municipal metabolic feature data for the difference operation in subsequent steps.
7. The land use review method based on deep learning and big data according to claim 5, characterized in that, The construction methods of the preset business activity metabolism template library include: Collect a preset number of time-series samples of metabolic residuals from benchmark buildings that are confirmed to be engaged in catering, entertainment, and accommodation activities. Extract the peak time period, peak-to-valley ratio, day-night ratio, and weekend-to-weekday ratio from each sample. Calculate the statistical mean of each dimension of the feature vector of all samples under the same business type to form a metabolic template vector for the corresponding business type. The preset number is not less than the preset minimum sample size threshold to ensure the statistical representativeness of the template vector; when the sample size is lower than the preset minimum sample size threshold, it is supplemented by interpolation of template vectors of adjacent business types until the preset minimum sample size threshold is met.
8. The land use review method based on deep learning and big data according to claim 1, characterized in that, After generating the composite violation determination result or the hidden underground violation determination result, the method further includes: Based on the business type in the matching results of the underground activity type, the empirical value of the standard metabolic rate per unit area for that business type is obtained from the preset metabolic parameter library. The estimated area of illegal activities in underground space is obtained by dividing the peak metabolic rate in the time series of the metabolic residual by the empirical value of the standard metabolic rate per unit area. The estimated area is compared with the basement area data in the building physical registration data to obtain the rate of illegal use of underground space; The estimated area of illegal activities in the underground space and the rate of illegal use of the underground space will be included in the use review report.
9. A land use review system based on deep learning and big data, characterized in that: include: The data acquisition module is used to acquire above-ground behavioral characteristic data of the target building, municipal metabolic characteristic data of the municipal access node corresponding to the target building, and building physical registration data; the municipal metabolic characteristic data includes time-of-use drainage flow data and time-of-use electricity consumption data; the building physical registration data includes the registered area of each floor, registered use, and basement area data; The benchmark calculation module is used to calculate the theoretical metabolic benchmark amount time series of the above-ground layer based on the building physical registration data and the preset use metabolic parameter library; the preset use metabolic parameter library stores the standard metabolic amount empirical values and time series distribution patterns of each use type per unit area and per unit time. The residual extraction module is used to subtract the municipal metabolic characteristic data of each time unit from the benchmark quantity of the corresponding time unit in the time series of the aboveground theoretical metabolic benchmark quantity to obtain the time series of metabolic residual quantities. The metabolic residual time series characterizes the load contribution of substantive activities in the underground space to the municipal access node after deducting the metabolic contribution of compliant activities in the above-ground layer. The activity identification module is used to extract the intraday peak time period, peak-to-valley ratio, day-night ratio, and weekend-to-weekday ratio from the metabolic residual time series to form a residual metabolic feature vector; the residual metabolic feature vector is matched with a preset commercial activity metabolic template library to obtain the underground activity type matching result and the underground violation score. The scoring calculation module is used to input the above-ground behavior feature data into a pre-trained above-ground activity intensity classification model to obtain the above-ground layer violation score; The above-ground violation score characterizes the weighted deviation of the above-ground behavioral feature data from the benchmark value of the registered use of the target building; The violation determination module is used to compare the violation score of the above-ground layer with a preset above-ground violation threshold and the violation score of the underground layer with a preset underground violation threshold. When both exceed their respective preset thresholds, a composite violation determination result is generated. When only the violation score of the underground layer exceeds the preset underground violation threshold, a hidden underground violation determination result is generated. When only the violation score of the above-ground layer exceeds the preset above-ground violation threshold, an above-ground violation determination result is generated. The report generation module is used to generate a usage review report based on the judgment results, the above-ground behavior characteristic data, and the underground activity type matching results. The report includes a violation nature classification label, the violation spatial layer attribution, the inference of the underground activity business type, and multi-dimensional evidence chain data.
10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the land use review method based on deep learning and big data as described in any one of claims 1 to 8.