A multi-channel error correction and visualization method based on real estate data

Abstract

The application provides a multi-channel error correction and visualization method based on real estate data, and belongs to the technical field of error correction and visualization, and comprises the following steps: real-time integration and coordinate unification of multi-channel heterogeneous real estate data are performed to determine standardized channel data; abnormal features of current real estate data are extracted, current real estate data are subjected to abnormality determination to determine current abnormal real estate data, multi-channel technical error correction is performed on the current abnormal real estate data within an error correction time limit, and three-dimensional visualization display of the current real estate data and error correction results is realized on a spatial dimension platform. Abnormality is accurately identified by using deep learning, efficient error correction is realized by combining spatial interpolation and real-time scheduling, and finally, the data and error correction results are intuitively presented through three-dimensional visualization, so that the accuracy, processing efficiency and visual interaction experience of the real estate data are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of error correction and visualization technology, and in particular to a multi-channel error correction and visualization method based on building data. Background Technology

[0002] With the advancement of digitalization in the real estate industry, the scale of data is growing exponentially. Property data is dynamic, exhibiting characteristics such as price fluctuations, status updates, and spatial correlations, such as the relationship between building location and surrounding amenities. Existing anomaly detection methods are mostly based on single-dimensional thresholds, such as classifying price increases exceeding 30% as anomalies, lacking in-depth analysis of spatiotemporal characteristics. Current error correction methods rely primarily on manual verification, depending on operator experience, resulting in long correction cycles (averaging over 72 hours) and a lack of clear time-limit management mechanisms, leading to the inability to promptly correct critical data anomalies. Furthermore, single-channel error correction, such as correcting price anomalies solely based on transaction data, is susceptible to channel data bias, resulting in insufficient reliability of correction results and difficulty in supporting accurate decision-making.

[0003] Therefore, this invention proposes a multi-channel error correction and visualization method based on building data. Summary of the Invention

[0004] This invention provides a multi-channel error correction and visualization method based on building data to solve the aforementioned technical problems.

[0005] This invention provides a multi-channel error correction and visualization method based on building data, including:

[0006] Step 1: Based on the distributed heterogeneous data fusion algorithm, integrate heterogeneous property data from multiple channels in real time, achieve coordinate unification through a spatial benchmark normalization model, and determine standardized channel data;

[0007] Step 2: Use a spatiotemporal feature autoencoder to automatically compare standardized channel data with the current number of properties and extract abnormal features of the current property data. Then, use a multi-dimensional anomaly confidence calculation model to determine the current property data as abnormal and identify the current abnormal property data.

[0008] Step 3: Based on the spatial interpolation error correction algorithm and real-time error correction scheduling mechanism, perform multi-channel technical error correction on the current abnormal building data within the error correction time limit;

[0009] Step 4: Using data rendering technology and anomaly heat dynamic overlay algorithm, a three-dimensional visualization of the current building data and error correction results is realized on the spatial dimension platform.

[0010] Preferably, step 1 includes:

[0011] Acquire heterogeneous property data from multiple channels and preprocess the heterogeneous property data to remove data noise;

[0012] Based on a distributed computing framework, a heterogeneous data fusion algorithm is deployed to perform real-time fusion processing on pre-processed heterogeneous building data. Through data format conversion, field mapping, and redundant data removal, a fused dataset is obtained.

[0013] A spatial reference normalization model is constructed, and coordinate transformation is performed on the spatial coordinate data in the fused dataset to unify spatial coordinates from different sources to the reference coordinate system.

[0014] The fused dataset, after coordinate normalization, is standardized according to preset data standards to generate standardized channel data.

[0015] Preferably, step 2 includes:

[0016] A spatiotemporal feature autoencoder is constructed, which is composed of an LSTM network and a CNN network connected in series. By inputting the current building data into the autoencoder and comparing it with standardized channel data, high-dimensional spatiotemporal features are extracted and reconstruction error is calculated. Feature segments with reconstruction error exceeding a preset threshold are marked as abnormal features. The extraction dimension of high-dimensional spatiotemporal features is not less than 128 dimensions, and the reconstruction error threshold is adaptively generated through training with historical normal data.

[0017] A multi-dimensional anomaly confidence calculation model is established, and anomaly confidence in time dimension, spatial dimension, and attribute dimension is calculated respectively using anomaly features as input. The confidence weights of each dimension are optimized by gradient descent algorithm, and the comprehensive anomaly confidence value is output.

[0018] When the overall anomaly confidence value exceeds the judgment threshold, the corresponding current property data will be identified as the current abnormal property data.

[0019] Preferably, step 3 includes:

[0020] Analyze the anomaly type and corresponding anomaly correction time limit of the current abnormal property data;

[0021] Based on the anomaly type and the spatial location of the current abnormal property data, multi-channel credible reference data within a preset spatial neighborhood of the current abnormal property data is selected from standardized channel data. The multi-channel credible reference data includes at least externally sourced surveying and mapping benchmark data, authoritative transaction record data, third-party evaluation data, and historical error correction and verification data.

[0022] An improved spatial interpolation error correction algorithm is adopted, using the multi-channel reliable reference data as interpolation samples. The interpolation weight factor is dynamically adjusted according to the reliability score of each reference data and the spatial distance with the current abnormal building data. The correction value of spatial coordinate anomaly and the correction candidate value of attribute information anomaly are calculated respectively.

[0023] A real-time error correction scheduling mechanism is constructed. Based on the urgency of the error correction time limit and the impact weight of the current abnormal building data, the error correction task priority is generated, and the error correction task is allocated to edge computing nodes for parallel processing to ensure that the error correction calculation is completed within the error correction time limit.

[0024] Based on the spatial coordinate correction value, attribute information correction candidate value, and consistency verification results of multi-channel reliable reference data, the current abnormal building data is corrected, and the final error correction data and corresponding error correction reliability are output.

[0025] Preferably, step 4 includes:

[0026] Based on data rendering technology, a three-dimensional spatial model of the current building data is constructed in the spatial dimension platform. The three-dimensional spatial model includes the building form, attribute information and corresponding error correction result markers. The attribute information is dynamically displayed through model surface texture mapping.

[0027] Based on the abnormal heat dynamic overlay algorithm, a dynamic heat layer is generated according to the spatial distribution of the current abnormal building data and the difference before and after the error correction. The dynamic heat layer reflects the spatiotemporal evolution of the abnormal density and the error correction effect through changes in color gradient and transparency.

[0028] The three-dimensional spatial model and dynamic heat map are aligned and layered on the spatial dimension platform to achieve smooth rendering and interactive viewing of large-scale building data. At the same time, clicking on the building model triggers a pop-up window displaying anomaly details and the error correction process, realizing a three-dimensional presentation of the current building data and error correction results.

[0029] Preferably, the time limit for correcting anomalies in the parsing of the current abnormal property data includes:

[0030] Extract the anomaly impact factor, data decay coefficient, error correction complexity coefficient, and data correlation coefficient of the current abnormal property data, and determine the initial time limit;

[0031] The multi-channel data is mapped to a baseline dimension, while the assisting data is mapped to a collaborative dimension.

[0032] Determine the initial correlation between all baseline dimensions and each collaborative dimension;

[0033]

[0034] Where Ri1 represents the initial correlation between all baseline dimensions and the i1th collaborative dimension; ws, wa, and wt are the weights of the spatial, attribute, and temporal dimensions, respectively, and ws + wa + wt = 1; d is the average spatial coordinate distance between the baseline spatial dimension and the i1th collaborative dimension; k is the spatial decay coefficient, with a value of 0.001; V 基 The attribute value of the baseline dimension of the attribute; V 协 t represents the attribute value of the i1th collaborative dimension. 基 The timestamp is the baseline dimension of time; t 协 T is the timestamp of the i1th collaborative dimension; 周期 For data update cycle;

[0035] Determine the coverage of the multi-channel data for each anomaly type, and obtain the set of correction factors by matching from the type combination-factor lookup table, and correct the initial time limit to obtain a first correction time limit;

[0036]

[0037] Among them, T 初 is the initial time limit; wi is the weight of the data from the i-th channel; ci is the coverage of the anomaly corresponding to the i-th channel; Softmax9M 类型 ) is for dynamically allocating exception type M 类型 The weights of T; 一 The correction period is defined as one correction cycle; wici is the correction factor for matching data from the i-th channel.

[0038] Based on the initial correlation degree, combined with the historical average time μ and historical time standard deviation δ determined based on historical error correction data, and also combined with the error correction delay factor determined based on the data transmission link length and the judgment time of different anomaly types, a multi-maintenance positive factor is determined.

[0039] The first correction time limit is further corrected based on the aforementioned multiple maintenance positive factors to obtain the second correction time limit T. 二 ;

[0040]

[0041] Among them, S 协同 This represents the degree of synergy between the baseline dimension and the synergistic dimension; Sigmoid() is the synergy correction function; F 多维 For multiple maintenance positive factors;

[0042] Based on the delay retrieval deviation of the baseline dimension caused by the aforementioned collaborative dimension, the time limit confidence level Pz is determined;

[0043] Based on the aforementioned time limit confidence level and combined with the secondary correction time limit, the anomaly correction time limit is obtained.

[0044] Preferably, the positive maintenance factor F is determined. 多维 ,include:

[0045]

[0046] Where ReLU() is the correction function, if but =0 otherwise = L 链路 k0 represents the length of the multi-channel data transmission link; k0 represents the latency-sensitive time per unit length; T represents the multi-channel data transmission link length. 链路阈值 This represents the maximum tolerable link delay time; (Ri1) ave The mean of all Ri1; σ Ri1 Let be the standard deviation of all Ri1.

[0047] Preferably, determining the time-limit confidence level Pz includes:

[0048]

[0049] Where ei represents the processing efficiency of the i-th channel data; L 延迟偏差 The collaborative data slows down the latency of the baseline data; T 偏差阈值 represents the maximum tolerance time for delay deviation; max(ei) is the maximum value among all ei.

[0050] Compared with the prior art, the beneficial effects of this application are as follows:

[0051] By generating standardized benchmarks through distributed integration of heterogeneous data from multiple channels, accurately identifying anomalies using deep learning, and achieving efficient error correction through spatial interpolation and real-time scheduling, the data and error correction results are presented intuitively through 3D visualization. This significantly improves the accuracy, processing efficiency, and visual interactive experience of property data, providing reliable technical support for real estate data management, market analysis, and decision-making.

[0052] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0053] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0054] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0055] Figure 1 This is a flowchart of a multi-channel error correction and visualization method based on building data in an embodiment of the present invention. Detailed Implementation

[0056] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0057] This invention provides a multi-channel error correction and visualization method based on building data, such as... Figure 1 As shown, it includes:

[0058] Step 1: Based on the distributed heterogeneous data fusion algorithm, integrate heterogeneous property data from multiple channels in real time, achieve coordinate unification through a spatial benchmark normalization model, and determine standardized channel data;

[0059] Step 2: Use a spatiotemporal feature autoencoder to automatically compare standardized channel data with the current number of properties and extract abnormal features of the current property data. Then, use a multi-dimensional anomaly confidence calculation model to determine the current property data as abnormal and identify the current abnormal property data.

[0060] Step 3: Based on the spatial interpolation error correction algorithm and real-time error correction scheduling mechanism, perform multi-channel technical error correction on the current abnormal building data within the error correction time limit;

[0061] Step 4: Using data rendering technology and anomaly heat dynamic overlay algorithm, a three-dimensional visualization of the current building data and error correction results is realized on the spatial dimension platform.

[0062] Preferably, step 1 includes:

[0063] Acquire heterogeneous property data from multiple channels and preprocess the heterogeneous property data to remove data noise;

[0064] Based on a distributed computing framework, a heterogeneous data fusion algorithm is deployed to perform real-time fusion processing on pre-processed heterogeneous building data. Through data format conversion, field mapping, and redundant data removal, a fused dataset is obtained.

[0065] A spatial reference normalization model is constructed, and coordinate transformation is performed on the spatial coordinate data in the fused dataset to unify spatial coordinates from different sources to the reference coordinate system.

[0066] The fused dataset, after coordinate normalization, is standardized according to preset data standards to generate standardized channel data.

[0067] Preferably, step 2 includes:

[0068] A spatiotemporal feature autoencoder is constructed, which is composed of an LSTM network and a CNN network connected in series. By inputting the current building data into the autoencoder and comparing it with standardized channel data, high-dimensional spatiotemporal features are extracted and reconstruction error is calculated. Feature segments with reconstruction error exceeding a preset threshold are marked as abnormal features. The extraction dimension of high-dimensional spatiotemporal features is not less than 128 dimensions, and the reconstruction error threshold is adaptively generated through training with historical normal data.

[0069] A multi-dimensional anomaly confidence calculation model is established, and anomaly confidence in time dimension, spatial dimension, and attribute dimension is calculated respectively using anomaly features as input. The confidence weights of each dimension are optimized by gradient descent algorithm, and the comprehensive anomaly confidence value is output.

[0070] When the overall anomaly confidence value exceeds the judgment threshold, the corresponding current property data will be identified as the current abnormal property data.

[0071] Preferably, step 3 includes:

[0072] Analyze the anomaly type and corresponding anomaly correction time limit of the current abnormal property data;

[0073] Based on the anomaly type and the spatial location of the current abnormal property data, multi-channel credible reference data within a preset spatial neighborhood of the current abnormal property data is selected from standardized channel data. The multi-channel credible reference data includes at least externally sourced surveying and mapping benchmark data, authoritative transaction record data, third-party evaluation data, and historical error correction and verification data.

[0074] An improved spatial interpolation error correction algorithm is adopted, using the multi-channel reliable reference data as interpolation samples. The interpolation weight factor is dynamically adjusted according to the reliability score of each reference data and the spatial distance with the current abnormal building data. The correction value of spatial coordinate anomaly and the correction candidate value of attribute information anomaly are calculated respectively.

[0075] A real-time error correction scheduling mechanism is constructed. Based on the urgency of the error correction time limit and the impact weight of the current abnormal building data, the error correction task priority is generated, and the error correction task is allocated to edge computing nodes for parallel processing to ensure that the error correction calculation is completed within the error correction time limit.

[0076] Based on the spatial coordinate correction value, attribute information correction candidate value, and consistency verification results of multi-channel reliable reference data, the current abnormal building data is corrected, and the final error correction data and corresponding error correction reliability are output.

[0077] Preferably, step 4 includes:

[0078] Based on data rendering technology, a three-dimensional spatial model of the current building data is constructed in the spatial dimension platform. The three-dimensional spatial model includes the building form, attribute information and corresponding error correction result markers. The attribute information is dynamically displayed through model surface texture mapping.

[0079] Based on the abnormal heat dynamic overlay algorithm, a dynamic heat layer is generated according to the spatial distribution of the current abnormal building data and the difference before and after the error correction. The dynamic heat layer reflects the spatiotemporal evolution of the abnormal density and the error correction effect through changes in color gradient and transparency.

[0080] The three-dimensional spatial model and dynamic heat map are aligned and layered on the spatial dimension platform to achieve smooth rendering and interactive viewing of large-scale building data. At the same time, clicking on the building model triggers a pop-up window displaying anomaly details and the error correction process, realizing a three-dimensional presentation of the current building data and error correction results.

[0081] In this embodiment, the distributed heterogeneous data fusion algorithm is based on a distributed computing architecture and is used to integrate multi-channel data with different sources and different formats / structures. It achieves data fusion through operations such as format conversion, field matching, and redundancy removal. For example, when integrating JSON format property descriptions from Internet platforms, CAD format coordinate data from external surveying, and price data from database tables in transaction systems, the algorithm can map the area field in the JSON to a unified field of building area (square meters), convert CAD coordinates to numerical coordinates, and remove duplicate listing records.

[0082] Multi-channel heterogeneous property data refers to a collection of property data from different channels (such as internet platforms, external procurement teams, transaction systems, user feedback, etc.) that differ in format, structure, and standards. The raw dataset is formed by crawling internet data, connecting to transaction systems via API, and manually entering externally procured data.

[0083] Real-time integration involves low-latency (minute-level) fusion processing of data from multiple channels to ensure data timeliness.

[0084] The spatial reference normalization model is a mathematical model used to convert spatial coordinates from different channels into the same reference coordinate system (such as the National Geodetic Coordinate System 2000). This solves the problem of coordinate system differences. For example, it can unify the X=350000, Y=4500000 coordinates of local coordinate systems (such as the Beijing 54 coordinate system) and the fuzzy positioning coordinates of Internet platforms into precise coordinates under the National Geodetic Coordinate System 2000 through a seven-parameter transformation method (translation, rotation, and scaling parameters). This eliminates the reference differences of spatial coordinates from different channels and makes all data comparable under the same coordinate system.

[0085] Standardized channel data is a collection of multi-channel data that has been integrated, with unified coordinates and standardized formats. It has unified fields, formats and standards. For example, it is a CSV format dataset containing fields such as building ID, coordinates in the national 2000 coordinate system, building area (square meters), average transaction price (yuan / square meter), and data update timestamp. The field definitions conform to industry data standards.

[0086] Preprocessing is the process of cleaning the original heterogeneous data and removing noise (such as erroneous values, duplicate values, and missing values). Mean filtering and the isolated forest algorithm are used to identify and remove noise, and missing values ​​are filled by the K-nearest neighbor algorithm.

[0087] Data noise refers to data in the original data that does not conform to logic or deviates from the normal range, such as erroneous values, duplicate values, and extreme outliers.

[0088] Distributed computing frameworks are software frameworks that support parallel computing across multiple nodes, used for efficient processing of large-scale data, such as Apache Spark (an in-memory computing framework suitable for batch processing) and Apache Flink (a stream processing framework suitable for real-time data).

[0089] Data format conversion is the process of converting data in different formats (such as JSON, CAD, Excel) into a unified format (such as CSV, Parquet).

[0090] Field mapping is the operation of mapping synonymous fields (such as area, building area, house size) from different channels to a unified field.

[0091] Redundant data removal involves deleting duplicate or unnecessary data records to reduce the amount of data.

[0092] A fused dataset is a unified dataset formed after preprocessing, format conversion, field mapping, and redundancy removal.

[0093] A reference coordinate system is a coordinate system that serves as a unified reference for coordinate systems. It is usually an authoritative coordinate system (such as the National Geodetic Coordinate System 2000).

[0094] Preset data standards are pre-defined data classifications, structures, interfaces, and logical rules used to standardize data formats and content.

[0095] The spatiotemporal feature autoencoder is a deep learning model composed of LSTM (for processing time series features) and CNN (for processing spatial features) cascaded together. It is used to learn the normal spatiotemporal features of standardized data and extract anomalies by comparing with the current data. For example, LSTM learns the monthly price increase pattern of a certain property (normal fluctuation ±5%), and CNN learns the spatial distribution features of apartment types in the same community (area difference ≤10%). When new data shows a monthly price increase of 30% and an area difference of 20%, the model extracts it as an anomalous feature. The model is built based on TensorFlow, using 100,000 historical normal data as training samples, and a 128-dimensional hidden layer is set to extract high-dimensional features.

[0096] High-dimensional spatiotemporal features are abstract features extracted from time and space dimensions, with no fewer than 128 dimensions. They are used to characterize complex patterns in data. For example, they include feature vectors with 128 dimensions such as price time-series fluctuation coefficient, coordinate space clustering distance, and area-house type matching degree.

[0097] Reconstruction error is the difference between the reconstruction result of the input data by the autoencoder and the original data. The larger the difference, the more likely the data is to be abnormal. For example, the average reconstruction error of normal data is 0.05, while the reconstruction error of abnormal data is 0.4 (far exceeding the normal range).

[0098] The preset threshold is a critical value used to determine whether the reconstruction error is abnormal. It is generated through training on historical normal data. For example, based on the reconstruction error of 95% of historical normal data, the threshold is set to 0.3 (if it exceeds, it is marked as abnormal).

[0099] Anomalies are significant deviations between current property data and standardized data in terms of spatiotemporal characteristics, such as anomalous time series fluctuations, anomalous spatial distribution, and anomalous attribute associations. For example, a property's price may have increased by more than 20% for three consecutive months (anomalous time), its coordinates may have deviated from other buildings in the same community by 100 meters (anomalous space), and its three-bedroom units may have an area of ​​only 50 square meters (anomalous attributes).

[0100] The multi-dimensional anomaly confidence calculation model calculates anomaly confidence from three dimensions: time, space, and attribute. It outputs a comprehensive confidence model through weight optimization. For example, the time dimension calculates the price time-series consistency entropy value (the higher the entropy value, the more anomaly); the space dimension calculates the deviation of the coordinates from the surrounding buildings (the higher the deviation, the more anomaly); and the attribute dimension calculates the matching degree between the area and the apartment type (the lower the matching degree, the more anomaly). The weights are optimized through the gradient descent algorithm (time 0.3, space 0.4, attribute 0.3) to output the comprehensive confidence score.

[0101] The confidence level for time-dimensional anomalies measures the likelihood of data exhibiting anomalies over time (0-1, with higher values ​​indicating greater likelihood of anomalies).

[0102] Spatial dimension anomaly confidence measures the credibility of data anomalies in spatial dimensions (such as coordinate location). For example, if the coordinates of a certain building deviate from the baseline coordinates of the same community by 50 meters, the spatial dimension confidence is 0.85.

[0103] Attribute dimension anomaly confidence measures the credibility of data when there are anomalies in attribute dimensions (such as area or apartment type).

[0104] The overall anomaly confidence score is a weighted average of the confidence scores for time, space, and attribute dimensions, used for the final anomaly determination. For example, time 0.9×0.3 + space 0.85×0.4 + attribute 0.8×0.3 = 0.86 (overall confidence score).

[0105] The judgment threshold is a critical value used to determine whether the overall confidence level is abnormal. It is dynamically adjusted based on the distribution characteristics of historical abnormal data. For example, based on historical data, the judgment threshold is set to 0.8 (if the overall confidence level is >0.8, it is judged as abnormal).

[0106] The current abnormal property data refers to the current property data that has been determined to be abnormal after anomaly detection.

[0107] Spatial interpolation error correction algorithm is based on spatial correlation. It uses reliable reference data around abnormal data to correct spatial coordinate or attribute anomalies through interpolation calculation. The algorithm filters reference data within 500 meters of the abnormal point and dynamically adjusts the interpolation weight according to distance (the closer the distance, the higher the weight) and reliability (0.6 for surveying data and 0.4 for transaction data).

[0108] The real-time error correction scheduling mechanism prioritizes and allocates error correction tasks to computing nodes based on the urgency of the error correction deadline and the weight of the data's impact, ensuring timely completion. For example, errors in the coordinates of buildings in core business districts (time limit 24 hours, impact weight 0.9) are prioritized and allocated to edge node A, while errors in the area of ​​buildings in remote areas (time limit 72 hours, impact weight 0.3) are allocated to node B, enabling parallel processing. Specifically, the priority is calculated based on the task's urgency (1-10 points, higher scores indicate greater urgency) and impact weight (0-1) (priority = urgency × 0.6 + weight × 0.4), and then scheduled to edge nodes via Kubernetes.

[0109] The error correction time limit is the time limit for completing the error correction of abnormal data, which is determined by the initial time limit after multiple corrections.

[0110] The current abnormal property data refers to the abnormal data identified in step 2, which is the data that needs to be corrected.

[0111] Multi-channel error correction is a process of cross-validating and correcting anomalous data using trusted data from multiple channels.

[0112] Anomaly type refers to the specific type of abnormal data, such as spatial coordinate anomalies or attribute information anomalies (price, area, etc.).

[0113] The preset spatial neighborhood is a specific spatial range (such as a 500-meter radius) around the abnormal data, used to filter reference data. For example, the area within a 500-meter radius centered on the abnormal building can be used to filter surrounding reference data.

[0114] The multi-channel credible reference data comes from multiple authoritative sources and is highly reliable, and is used for error correction and verification.

[0115] External surveying benchmark data: Coordinate and area data measured on-site by the external survey team (error < 0.5 meters);

[0116] Authoritative transaction record data: Transaction data filed with the housing and construction department (e.g., the average transaction price of XX property in 2023 was 30,000 yuan / square meter);

[0117] Third-party assessment data: Property attribute report issued by a professional assessment agency (e.g., area of ​​120 square meters, assessed value of 3.6 million yuan);

[0118] Historical error correction verification data: Historical error correction results that have been verified multiple times (such as the correction in 2022 that the area of ​​XX building is 120 square meters, and there have been no subsequent objections).

[0119] The improved spatial interpolation error correction algorithm is an algorithm that dynamically adjusts the weights based on the reliability of reference data, on the basis of the traditional interpolation algorithm. Specifically, the traditional IDW algorithm only considers distance, while the improved algorithm adds a reliability weight (e.g., 0.6 for surveying data and 0.4 for transaction data).

[0120] Interpolation samples are used as reference data points for interpolation calculations. For example, there are 3 external survey points and 2 authoritative transaction points around the abnormal building, totaling 5 interpolation samples.

[0121] Credibility scores measure the credibility of reference data (0-1, the higher the value, the more credible). For example, externally sourced surveying data has a credibility score of 0.9, transaction data 0.7, and user feedback data 0.3.

[0122] Spatial distance is the straight-line distance between abnormal data and reference data points. For example, the abnormal building is 100 meters away from reference point A and 200 meters away from reference point B.

[0123] The interpolation weight factor is the weight of each reference data in the interpolation calculation (positively correlated with confidence and negatively correlated with distance). For example, reference point A (distance 100 meters, confidence 0.9) has a weight of 0.4, and reference point B (distance 200 meters, confidence 0.7) has a weight of 0.2.

[0124] The correction value for spatial coordinate anomalies is calculated using an interpolation algorithm and is the value after correcting for spatial coordinate anomalies. For example, the anomaly coordinates (116.5°E, 39.8°N) are corrected to (116.49°E, 39.81°N).

[0125] The correction candidate value for abnormal attribute information is a candidate value for correcting abnormal attributes (such as area and price), and the final value needs to be determined through consistency verification.

[0126] Impact weight measures the degree of impact of abnormal data on business (0-1, the higher the value, the greater the impact). For example, the impact weight of abnormal data on properties in core business districts is 0.9, while that of abnormal data on remote areas is 0.3.

[0127] Error correction task priority refers to the order in which error correction tasks are processed; those with higher priority are processed first.

[0128] Edge computing nodes are computing nodes located close to the data source, reducing transmission latency and improving processing efficiency. For example, edge servers deployed in various districts of a city are responsible for handling error correction tasks in their respective areas.

[0129] Parallel processing involves multiple error correction tasks being processed simultaneously on different nodes to improve efficiency. For example, node A processes coordinate error correction while node B processes area error correction, and so on.

[0130] The consistency check result is the degree of consistency between multiple reference data and the correction candidate value (the higher the consistency rate, the more reliable the candidate value). For example, if two out of three reference data support an area of ​​120 square meters, the consistency rate is 67%, and it is determined as the final correction value.

[0131] The final error-corrected data is the corrected data determined after consistency verification.

[0132] Error correction reliability is an indicator that measures the reliability of the final error correction data (0-1, the higher the value, the more reliable).

[0133] Data rendering technology is a technique that converts data into a three-dimensional visualization model, supporting the intuitive display of building form and attribute information. For example, using WebGL technology, building coordinates and height data can be rendered into a three-dimensional building model (height corresponds to the number of floors).

[0134] The anomaly thermal dynamic overlay algorithm is an algorithm that dynamically overlays the spatial distribution density of anomaly data and the difference before and after error correction onto a 3D model through color gradients and transparency. For example, a region with high anomaly density (50 per square kilometer) is represented by red (80% transparency), and after error correction, it is reduced to 5 and represented by yellow (30% transparency). The algorithm is dynamically updated by calculating the anomaly density based on kernel density estimation (KDE) and rendering the thermal layer in real time through GPU shaders.

[0135] The spatial dimension platform is a software platform that supports the display and interaction of three-dimensional spatial data. It integrates maps and rendering engines. For example, a platform developed based on Cesium supports the rotation, scaling, translation of building models, as well as the selection and query of abnormal areas.

[0136] 3D visualization is a way of presenting data and error correction results in a visual way using 3D models, heat maps, etc. For example, on the platform, you can see the 3D buildings of XX community. Red buildings indicate anomalies. After clicking, it will show the original area of ​​150 square meters - the corrected area of ​​120 square meters.

[0137] A 3D spatial model is a scaled-down 3D model of a building, containing information such as building form and attributes. For example, the 3D model of XX community contains 3 buildings of 18 floors each, with each building labeled with unit number and apartment type.

[0138] The architectural form of a building refers to its physical characteristics, such as the number of buildings, the number of floors, and the exterior structure.

[0139] Attribute information refers to the non-spatial characteristics of a property, such as price, area, and unit type.

[0140] Error correction result marking involves marking the differences before and after error correction on a 3D model, such as using color or symbol markings.

[0141] Model surface texture mapping is a technique that maps attribute information (such as color and pattern) onto the surface of a 3D model.

[0142] For example: Blue represents buildings with an average price of 20,000-30,000 yuan per square meter, and red represents buildings with an average price of 30,000-40,000 yuan per square meter.

[0143] The dynamic heatmap layer is a layer that dynamically displays the density of anomalies and the effect of error correction. It reflects the evolution of time and space through changes in color and transparency. For example, at 8 a.m., the core business district has an abnormally dense heatmap (red), which becomes sparse (yellow) after 2 hours of error correction.

[0144] Color gradients use different colors to represent differences in data values ​​(e.g., red → yellow → green represents anomaly density from high to low). For example, red is used for anomaly density >30 per square kilometer, yellow for 10-30, and green for <10.

[0145] Transparency is used to represent data reliability through layer transparency (0-100%) (lower transparency means higher reliability). High reliability error correction results have 30% transparency, while low reliability results have 70% transparency.

[0146] Anomaly density is the number of outliers per unit area (e.g., individuals per square kilometer).

[0147] The spatiotemporal evolution of the error correction effect is the change in the spatial distribution of abnormal data before and after error correction at different time points. For example, displaying an abnormal heatmap from 0:00 to 12:00 to 24:00 shows the process of the abnormal area gradually shrinking after error correction.

[0148] Coordinate alignment ensures that the coordinates of the 3D model and heatmap are perfectly matched on the spatial dimension platform.

[0149] Layered overlay fusion is the process of stacking 3D models and heat maps in layers to form a complete visualization effect. For example, the bottom layer is a 3D building model, the upper layer is an anomaly heat map, and the top layer is interactive controls.

[0150] Large-scale housing data refers to a massive amount of housing data (such as millions of records), for example, a dataset of 5,000 residential communities and 300,000 buildings in a certain city.

[0151] Smooth rendering means that the display of 3D models and heatmaps is lag-free and the frame rate is stable (≥30fps).

[0152] Interactive viewing allows users to interact with visual content through actions such as rotating, zooming, and clicking.

[0153] The pop-up window displaying the exception details and error correction process is a window that pops up when clicked, showing information such as the reason for the exception and a comparison before and after the error correction.

[0154] In this embodiment, the process of format conversion and field mapping specifically includes:

[0155] Semantic parsing is performed on each piece of unstructured data to determine a fuzzy set, wherein the fuzzy set contains fuzzy parameters and part-of-speech tags for the fuzzy parameters, and the fuzzy parameters are numerical parameters;

[0156] Based on each unstructured data point, each structured data point is traversed sequentially according to the parameter combination of the corresponding fuzzy set to obtain the structure set for each fuzzy parameter;

[0157] The structure set is expanded by performing noun expansion and unit expansion processing based on the limiting part-of-speech of the fuzzy parameters to obtain an expanded set, and the ambiguity of the expanded set is eliminated to obtain the clear limiting of the corresponding fuzzy parameters.

[0158] Based on the clear constraint results, the corresponding extended set is uniquely constrained to obtain the final quantization result of the corresponding fuzzy parameter, wherein the final quantization result includes: quantization value and quantization unit.

[0159] In this embodiment, unstructured data refers to property data that lacks a fixed format and is difficult to store directly using database fields. It is usually in the form of text and descriptive language. For example, the property description on the Internet platform is: XX community has 3 bedrooms and 2 living rooms, with an area of ​​about 120 square meters. It is close to the subway and there are supermarkets nearby. The user feedback is that the apartment layout is good, but the area feels a bit smaller than described.

[0160] Semantic parsing is the process of analyzing unstructured text using Natural Language Processing (NLP) techniques to extract key information (such as attributes and descriptive words) and understand their semantic relationships. For example, semantic parsing of an area of ​​approximately 120 square meters can identify that area is the core attribute, approximately is a modifier, and 120 square meters is the attribute value. Specifically, a BERT pre-trained model combined with a real estate domain dictionary is used to extract semantic elements through Named Entity Recognition (NER) and dependency parsing.

[0161] A fuzzy set is a collection of fuzzy information contained in unstructured data. It consists of fuzzy parameters and qualifying parts of speech and is used to characterize the uncertainty in the data. For example, for the text: "The area is about 120 square meters and it is quite close to the subway", the fuzzy set is {fuzzy parameters: [area, distance from the subway]; qualifying parts of speech: [about, quite close]}.

[0162] Fuzzy parameters are property attributes in unstructured data that are not precise in value or description and require further quantification.

[0163] The qualifying part of speech is a word that modifies a vague parameter and is used to describe the degree or range of fuzziness of the parameter (such as approximation, fuzzy range, uncertainty).

[0164] Structured data is property data with a fixed format and defined fields. It is usually stored in a relational database and can be directly recognized and processed by computers. For example, in a transaction system, the property ID is 1001; the area is 122 square meters; the distance from the subway station is 800 meters; and the building area is 120 square meters.

[0165] A parameter combination is a combination of multiple fuzzy parameters within a fuzzy set, which serves as a search keyword for traversing structured data.

[0166] A structure set is a collection of structured data fragments corresponding to fuzzy parameters, obtained after traversal. It contains specific values ​​related to the fuzzy parameters in the structured data. For example, for the fuzzy parameter area, the structure set obtained after traversal is {120 square meters, 122 square meters, 118 square meters}, which comes from different structured data records; for the distance from the subway, the structure set is {800 meters, 750 meters, 900 meters}.

[0167] Noun expansion processing converts colloquial and abbreviated nouns in unstructured data into standardized nouns, ensuring consistency with structured data fields. For example, "flat" is expanded to "square meters". This makes the nouns in unstructured data match the fields in structured data. Specifically, a noun mapping table for the real estate sector (e.g., flat → square meters) is established, and the expansion is automatically replaced through string matching.

[0168] Unit expansion processing supplements fuzzy parameters missing units in unstructured data with standardized units, or unifies different unit formats to ensure consistency with the units of structured data. Based on the fuzzy parameter type (such as area corresponding to square meters, distance corresponding to meters), the rule engine automatically completes or converts the units.

[0169] An extended set is a structured set that has been expanded in terms of nouns and units. It contains structured data values ​​with standardized nouns and uniform units. For example, the structure set {120 square meters, 122 square meters} becomes {120 square meters, 122 square meters} after expansion; the structure set {800 meters, 750 meters} becomes {800 meters, 750 meters} after expansion.

[0170] Ambiguity elimination is the process of determining a reasonable range for conflicting or ambiguous values ​​(such as differences in area values ​​from different sources) in an extended set, combined with limiting parts of speech and business logic. For example, if the area in the extended set is {120 square meters, 118 square meters, 150 square meters}, and the limiting part of speech is “approximately”, then 150 square meters (with a deviation exceeding 20%) is eliminated, and the range of 118-122 square meters is retained. Specifically, this involves introducing business thresholds for building data (such as allowing ±5% for area error and ±10% for distance error), eliminating outliers through standard deviation analysis, and retaining a reasonable range.

[0171] A clear definition is a definite range or constraint for a fuzzy parameter obtained after ambiguity elimination, used for subsequent quantification. For example, the area is clearly defined as 118-122 square meters.

[0172] A unique constraint is a constraint rule that determines a unique and precise value from an extended set based on a clearly defined range, eliminating uncertainty. For example, if the area is clearly defined as 118-122 square meters, and 120 square meters appears most frequently in the extended set (2 out of 3 structured data entries are 120 square meters), then the unique constraint is 120 square meters.

[0173] The final quantification result is the precise result obtained after processing the fuzzy variables. It contains specific values ​​and units and can be directly integrated with structured data.

[0174] The quantified value is the specific numerical value in the final quantification result, and it is a precise measurement of the fuzzy parameter. For example, 120 (corresponding to area), 800 (corresponding to distance), and 500 (corresponding to price, in ten thousand yuan). The quantification unit is a standardized unit that matches the quantified value to ensure that the physical meaning of the data is clear. For example, square meters (area), meters (distance), and ten thousand yuan (price).

[0175] In this embodiment, since there may be one, two, or more limiting terms, it is necessary to determine the fuzzy interval of the fuzzy variables involved in each unstructured data. Among them, X i2i2 The initial values ​​of the corresponding fuzzy parameters (e.g., 120 square meters) are used for the i2th unstructured data and the j2nd structured data; π ko The basic fuzzy coefficient for the ko-th qualifying part of speech is found in the part-of-speech-coefficient table, ε1 i2j2ko The credibility decay factor for the ko-th qualifying part of speech is pre-defined and is associated with the data type, with a value ranging from 0 to 1; m1 is the number of qualifying parts of speech.

[0176] If the qualifying word does not exist, then it is X. i2i2 .

[0177] Among them, when the data source type is related to external surveying and mapping, ε1 i2j2ko The value is 0.8, and ε1 is used when it is related to user feedback. i2j2ko The value is 0.3, and it applies to the part-of-speech tag-coefficient table, for example, Table 1 contains a portion of the content:

[0178] Table 1 Partial Part-of-Speech Coefficient Table

[0179] Defining Parts of Speech Fuzzy type Fuzzy coefficient Physical meaning (taking area as an example) none Precise description 0 The area value is unambiguous, such as 120 square meters. Approximately Numerical fuzziness 0.05 The area is allowed to have an error of ±5% (120±6 square meters). Close, relatively close Spatial blur 0.1 Distance tolerance ±10% (800±80 meters) Possible, suspected Logically ambiguous 0.2 Additional verification is required (e.g., for misspellings in the property name).

[0180] In this embodiment, the fuzzy interval provides a buffer space for multi-source data conflicts (e.g., {114, 126} square meters can cover multi-source data such as 118 and 122). The intensity of the conflict is judged by the degree of interval overlap, avoiding the direct discarding of conflicting data and retaining potentially correct information.

[0181] By extracting fuzzy information from unstructured data through semantic parsing, and combining it with structured data for expansion, ambiguity elimination, and unique constraints, accurate quantification results are obtained. This solves the problems of format differences and field mapping for heterogeneous data (unstructured and structured) from multiple channels, improves the accuracy of data fusion, and provides a standardized and high-quality data foundation for subsequent anomaly identification, error correction, and other processes.

[0182] The beneficial effects of the above technical solution are as follows: by generating standardized benchmarks through distributed integration of heterogeneous data from multiple channels, using deep learning to accurately identify anomalies, combining spatial interpolation and real-time scheduling to achieve efficient error correction, and finally presenting the data and error correction results intuitively through three-dimensional visualization, the accuracy, processing efficiency and visualization interaction experience of property data are significantly improved, providing reliable technical support for real estate data management, market analysis and decision-making.

[0183] This invention provides a multi-channel error correction and visualization method based on property data, which analyzes the error correction time limit of the current abnormal property data, including:

[0184] Extract the anomaly impact factor, data decay coefficient, error correction complexity coefficient, and data correlation coefficient of the current abnormal property data, and determine the initial time limit;

[0185] The multi-channel data is mapped to a baseline dimension, while the assisting data is mapped to a collaborative dimension.

[0186] Determine the initial correlation between all baseline dimensions and each collaborative dimension;

[0187]

[0188] Where Ri1 represents the initial correlation between all baseline dimensions and the i1th collaborative dimension; ws, wa, and wt are the weights of the spatial, attribute, and temporal dimensions, respectively, and ws + wa + wt = 1; d is the average spatial coordinate distance between the baseline spatial dimension and the i1th collaborative dimension; k is the spatial decay coefficient, with a value of 0.001; V 基 The attribute value of the baseline dimension of the attribute; V 协 t represents the attribute value of the i1th collaborative dimension. 基 The timestamp is the baseline dimension of time; t 协 T is the timestamp of the i1th collaborative dimension; 周期 For data update cycle;

[0189] Determine the coverage of the multi-channel data for each anomaly type, and obtain the set of correction factors by matching from the type combination-factor lookup table, and correct the initial time limit to obtain a first correction time limit;

[0190]

[0191] Among them, T 初 is the initial time limit; wi is the weight of the data from the i-th channel; ci is the coverage of the anomaly corresponding to the i-th channel; Softmax(M 类型 ) is for dynamically allocating exception type M 类型 The weights of T; 一The correction period is defined as one correction cycle; wici is the correction factor for matching data from the i-th channel.

[0192] Based on the initial correlation degree, combined with the historical average time μ and historical time standard deviation δ determined based on historical error correction data, and also combined with the error correction delay factor determined based on the data transmission link length and the judgment time of different anomaly types, a multi-maintenance positive factor is determined.

[0193] The first correction time limit is further corrected based on the aforementioned multiple maintenance positive factors to obtain the second correction time limit T. 二 ;

[0194]

[0195] Among them, S 协同 The degree of synergy between the baseline dimension and the synergistic dimension is represented; Sigmoid() is the synergy correction function; F 多维 For multiple maintenance positive factors;

[0196] Based on the delay retrieval deviation of the baseline dimension caused by the aforementioned collaborative dimension, the time limit confidence level Pz is determined;

[0197] Based on the aforementioned time limit confidence level and combined with the secondary correction time limit, the anomaly correction time limit is obtained.

[0198] Preferably, the positive maintenance factor F is determined. 多维 ,include:

[0199]

[0200] Where ReLU() is the correction function, if but =0 otherwise = L 链路 k0 represents the length of the multi-channel data transmission link; k0 represents the latency-sensitive time per unit length; T represents the multi-channel data transmission link length. 链路阈值 This represents the maximum tolerable link delay time; (Ri1) ave The mean of all Ri1; σ Ri1 Let be the standard deviation of all Ri1.

[0201] Preferably, determining the time-limit confidence level Pz includes:

[0202]

[0203] Where ei represents the processing efficiency of the i-th channel data; L 延迟偏差 The collaborative data slows down the latency of the baseline data; T 偏差阈值 represents the maximum tolerance time for delay deviation; max(ei) is the maximum value among all ei.

[0204] In this embodiment, the anomaly impact factor reflects the quantitative value of the impact of abnormal data on business. It is associated with transaction records, user queries, etc. Specifically, it is the ratio of the number of occurrences of the corresponding abnormal data to the total number of occurrences of all data under the corresponding association type, multiplied by the weight of the corresponding association type, and then summed. For example, if there are 100 abnormal transaction records, 1000 total transaction records, 500 abnormal user queries, and 1000 total queries, and the weight of the transaction record is 0.4 and the weight of the query is 0.6, then the calculation result is: (100 / 1000)×0.4+(500 / 1000)×0.6.

[0205] In this embodiment, the data decay coefficient γ1 is the degree to which data decays over time. Based on an exponential model, it is related to the time interval for detecting anomalies, according to the formula γ1 = e -k2·Δt Where k2 is determined by the business type, it is 0.02 for residential and 0.03 for commercial.

[0206] In this embodiment, the error correction complexity coefficient measures the difficulty of error correction and is determined by the type of abnormal field, the degree of data conflict, etc. It is divided into 5 levels and determined by a decision model. When the abnormal type, the degree of conflict, and the historical time consumption are input, the output can be mapped to 1-5 levels and the values ​​are standardized to between 0 and 1. Table 2 shows the abnormal type lookup table.

[0207] Table 2: Lookup Table by Abnormal Type

[0208] Exception types Complexity coefficient Spatial coordinate anomaly 1 Attribute information is incorrect (price). 0.6 Spatiotemporal correlation anomaly 0.8

[0209] In this embodiment, the data correlation coefficient is the deviation rate between the current outlier and the standard data benchmark, reflecting the degree of data anomaly. The formula is:

[0210] In this embodiment, the baseline dimension and the collaborative dimension are multi-channel data divided into baseline (authoritative, such as surveying and mapping) and collaborative (auxiliary, such as transaction) dimensions, used for correlation analysis. The baseline and collaborative dimensions are manually labeled to determine the baseline and collaborative dimensions. For example, the baseline dimension coordinates are (x1, y1), the attribute value is 100, and the timestamp is t1; the collaborative dimension coordinates are (x2, y2), the attribute value is 95, and the timestamp is t2, d = 100 meters, k = 0.001, ws = 0.5, wa = 0.3, wt = 0.2. At this time, Ri1 takes the value of 0.47.

[0211] In this embodiment, at least 10,000 historical abnormal data (such as 5,000 spatial abnormalities, 3,000 attribute abnormalities, and 2,000 spatiotemporal correlation abnormalities) are collected, and the average repair time Tx, resource consumption (such as the number of API calls), and number of user complaints (scope of impact) for each type of abnormality are calculated.

[0212] By using a linear regression model, the relationship between anomaly type and complexity is fitted, and the output M type (e.g., M space = 0.6 × Tx / Tmax + 0.3 × complaint rate + 0.1 × API call rate) is obtained, where Tmax is the maximum time spent in repairing each type of anomaly.

[0213] In this embodiment, an initial time limit consistent with the anomaly impact factor, data decay coefficient, error correction complexity coefficient, and data correlation coefficient is matched from a four-dimensional lookup table. This table contains initial time limits under different four-dimensional combinations, which are preset. For example, with anomaly impact factor of 1, data decay coefficient of 0.2, error correction complexity coefficient of 0.8, and data correlation coefficient of 0.7, the mapped initial time limit is 1.66 hours.

[0214] In this embodiment, Among them, w 基 w 协 Let w be the weights, and w be the weights. 基 >w 协 Generally w 基 The value is 0.6, w 协 The value is 0.4. E1 and E2 are the historical processing efficiencies based on the baseline dimension and the collaboration dimension, respectively, and the historical processing efficiency = the number of successful repairs in the corresponding dimension / the total number of tasks.

[0215] In this embodiment, parameters such as k and ws are currently set manually or based on experience. After the introduction of reinforcement learning, the model can automatically optimize the parameters based on historical error correction results (such as timeout rate and resource waste rate) to adapt to the building data of different regions and business types.

[0216] The beneficial effects of the above technical solution are: through multi-factor quantitative modeling and dynamic correction, the time limit for error correction of abnormal building data can be accurately calculated; from the experience-based estimation of the initial time limit, it is upgraded to intelligent control that integrates data correlation, processing complexity, and system status, which can reduce the error correction timeout rate and reduce invalid time consumption, providing core support for real estate data governance.

[0217] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A multi-channel error correction and visualization method based on property data, characterized in that, include: Step 1: Based on the distributed heterogeneous data fusion algorithm, integrate heterogeneous property data from multiple channels in real time, achieve coordinate unification through a spatial benchmark normalization model, and determine standardized channel data; Step 2: Use a spatiotemporal feature autoencoder to automatically compare standardized channel data with current property data and extract abnormal features of the current property data. Then, use a multi-dimensional anomaly confidence calculation model to determine the current property data as abnormal and identify the current abnormal property data. Step 3: Based on the spatial interpolation error correction algorithm and real-time error correction scheduling mechanism, perform multi-channel technical error correction on the current abnormal building data within the error correction time limit; Step 4: Using data rendering technology and anomaly heat dynamic overlay algorithm, realize the three-dimensional visualization display of the current building data and error correction results on the spatial dimension platform; Step 3 includes: Analyze the anomaly type and corresponding anomaly correction time limit of the current abnormal property data; Based on the anomaly type and the spatial location of the current abnormal property data, multi-channel credible reference data within a preset spatial neighborhood of the current abnormal property data is selected from standardized channel data. The multi-channel credible reference data includes at least externally sourced surveying and mapping benchmark data, authoritative transaction record data, third-party evaluation data, and historical error correction and verification data. An improved spatial interpolation error correction algorithm is adopted, using the multi-channel reliable reference data as interpolation samples. The interpolation weight factor is dynamically adjusted according to the reliability score of each reference data and the spatial distance with the current abnormal building data. The correction value of spatial coordinate anomaly and the correction candidate value of attribute information anomaly are calculated respectively. A real-time error correction scheduling mechanism is constructed. Based on the urgency of the error correction time limit and the impact weight of the current abnormal building data, the error correction task priority is generated, and the error correction task is allocated to edge computing nodes for parallel processing to ensure that the error correction calculation is completed within the error correction time limit. Based on the correction values ​​of the spatial coordinate anomalies, the correction candidate values ​​of the attribute information anomalies, and the consistency verification results of the multi-channel reliable reference data, the current abnormal building data is corrected, and the final error correction data and the corresponding error correction reliability are output.

2. The multi-channel error correction and visualization method based on building data according to claim 1, characterized in that, Step 1 includes: Acquire heterogeneous property data from multiple channels and preprocess the heterogeneous property data to remove data noise; Based on a distributed computing framework, a heterogeneous data fusion algorithm is deployed to perform real-time fusion processing on pre-processed heterogeneous building data. Through data format conversion, field mapping, and redundant data removal, a fused dataset is obtained. A spatial reference normalization model is constructed, and coordinate transformation is performed on the spatial coordinate data in the fused dataset to unify spatial coordinates from different sources to the reference coordinate system. The fused dataset, after coordinate normalization, is standardized according to preset data standards to generate standardized channel data.

3. The multi-channel error correction and visualization method based on building data according to claim 1, characterized in that, Step 2 includes: A spatiotemporal feature autoencoder is constructed, which is composed of an LSTM network and a CNN network connected in series. By inputting the current building data into the autoencoder and comparing it with standardized channel data, high-dimensional spatiotemporal features are extracted and reconstruction error is calculated. Feature segments with reconstruction error exceeding a preset threshold are marked as abnormal features. The extraction dimension of high-dimensional spatiotemporal features is not less than 128 dimensions, and the reconstruction error threshold is adaptively generated through training with historical normal data. A multi-dimensional anomaly confidence calculation model is established, and anomaly confidence in time dimension, spatial dimension, and attribute dimension is calculated respectively using anomaly features as input. The confidence weights of each dimension are optimized by gradient descent algorithm, and the comprehensive anomaly confidence value is output. When the overall anomaly confidence value exceeds the judgment threshold, the corresponding current property data will be identified as the current abnormal property data.

4. The multi-channel error correction and visualization method based on building data according to claim 1, characterized in that, Step 4 includes: Based on data rendering technology, a three-dimensional spatial model of the current building data is constructed in the spatial dimension platform. The three-dimensional spatial model includes the building form, attribute information and corresponding error correction result markers. The attribute information is dynamically displayed through model surface texture mapping. Based on the abnormal heat dynamic overlay algorithm, a dynamic heat layer is generated according to the spatial distribution of the current abnormal building data and the difference before and after the error correction. The dynamic heat layer reflects the spatiotemporal evolution of the abnormal density and the error correction effect through changes in color gradient and transparency. The three-dimensional spatial model and dynamic heat map are aligned and layered on the spatial dimension platform to achieve smooth rendering and interactive viewing of large-scale building data. At the same time, clicking on the building model triggers a pop-up window displaying anomaly details and the error correction process, realizing a three-dimensional presentation of the current building data and error correction results.

5. The multi-channel error correction and visualization method based on building data according to claim 1, characterized in that, The time limit for error correction of the current abnormal property data includes: Extract the anomaly impact factor, data decay coefficient, error correction complexity coefficient, and data correlation coefficient of the current abnormal property data, and determine the initial time limit; The heterogeneous property data from multiple channels is mapped to a baseline dimension, while the assisting data is mapped to a collaborative dimension. Determine the initial correlation between all baseline dimensions and each collaborative dimension; ;in, This represents the initial correlation between all baseline dimensions and the i1th collaborative dimension; The weights are respectively for the spatial, attribute, and temporal dimensions, and , Let be the average spatial coordinate distance between the base dimension and the i1th cooperating dimension; This is the spatial attenuation coefficient, with a value of 0.001; The attribute value for the base dimension of the attribute; This represents the attribute value of the i1th collaborative dimension; The timestamp is the baseline dimension for time. The timestamp of the i1th collaborative dimension; For data update cycle; Determine the coverage of the heterogeneous building data from the multi-channel sources for each anomaly type, and obtain the set of correction factors by matching them from the type combination-factor lookup table. Then, correct the initial time limit to obtain a first correction time limit. ;in, This is the initial time limit; The weight of the data from the i-th channel; Let i be the coverage of the anomaly corresponding to the i-th channel; For dynamically assigned exception types The weights; This is a one-time correction period; The correction factor for matching the data from the i-th channel; based on the initial correlation and combined with the historical average time consumption determined by historical error correction data. and historical time standard deviation Furthermore, it combines the error correction delay factor determined based on the data transmission link length and the judgment time of different anomaly types to determine the multi-maintenance positive factor; The first correction time limit is revised again based on the aforementioned multiple maintenance positive factors to obtain the second correction time limit. ; ;in, This indicates the degree of synergy between the baseline dimension and the collaborative dimension; For the degree of synergy correction function; For multiple maintenance positive factors; Based on the delay in retrieving the baseline dimension caused by the aforementioned collaborative dimension, the time limit confidence level is determined. Based on the time limit confidence level and combined with the second correction time limit, the anomaly correction time limit is obtained.

6. The multi-channel error correction and visualization method based on building data according to claim 5, characterized in that, Determine multiple maintenance positive factors ,include: ;in, For the correction function, if =0; otherwise = ; For the data transmission link length of heterogeneous buildings across multiple channels; The delay-sensitive time per unit length; This is the maximum tolerance time for link delay; For all The mean; For all The standard deviation.

7. The multi-channel error correction and visualization method based on building data according to claim 6, characterized in that, Determine the confidence level for the time limit ,include: in, The processing efficiency of the data from the i-th channel; The delay caused by collaborative data slows down the latency of baseline data; This is the maximum tolerance time for delay deviation; It is the maximum value among all ei.

Images