Leasant housing abnormity early warning device and method based on Internet of Things

Abstract

The invention provides a rental housing abnormity early warning device and method based on the Internet of Things, and relates to the technical field of the Internet of Things. Comprising the following steps: carrying out data acquisition on the public rental housing through a plurality of different types of sensors pre-arranged in the public rental housing to obtain a monitoring data set; preprocessing the monitoring data set, and performing feature conversion on the preprocessed monitoring data set to obtain a house state data set; performing feature extraction on the house state data set to obtain a plurality of use state features, performing anomaly analysis on the plurality of use state features to obtain a plurality of abnormal use state data, and performing early warning category division according to the plurality of abnormal use state data to obtain a house early warning result; and sending the house early warning result to an administrator according to the pre-configured communication interface. According to the method, the overall use state of the house is identified by analyzing the characteristics of the multiple different types of Internet of Things data, and the abnormal early warning analysis is performed, so that the early warning only depending on a single index is avoided, and the false alarm and missing alarm conditions are reduced.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of the Internet of Things, and particularly relates to a rental housing anomaly warning device and method based on the Internet of Things. Background Art

[0002] In the implementation process of the public rental housing policy, there are problems such as non - centralized housing areas, limited human resources of the regulatory department, single supervision methods, and high supervision costs, which may lead to violations such as housing vacancy, sub - leasing, rent arrears, and overdue non - return of the lease. With the increasing demand for public rental housing and the expanding coverage year by year, the resource investment and management cost investment in housing source management, household management, use supervision, statistics, and analysis involved in the management of public rental housing are huge. To further standardize the management of public rental housing, make full use of scientific and technological information technology means to strengthen post - rental management, improve the lease - return mechanism, ensure fair distribution, standardize operation and use, strengthen the subsequent supervision of public rental housing, ensure the fair and good use of housing security resources, and truly play the role of public rental housing in benefiting the people. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a rental housing anomaly warning device and method based on the Internet of Things in view of the deficiencies of the prior art.

[0004] The technical solution of the present invention for solving the above - mentioned technical problem is as follows:

[0005] A rental housing anomaly warning device based on the Internet of Things, comprising:

[0006] A collection unit, configured to collect data from public rental housing through a variety of different types of sensors pre - arranged in public rental housing to obtain a monitoring data set;

[0007] A pre - processing unit, configured to pre - process the monitoring data set and perform feature transformation on the pre - processed monitoring data set to obtain a housing status data set;

[0008] An analysis unit, configured to extract features from the housing status data set to obtain multiple usage status features, perform anomaly analysis on the multiple usage status features respectively to obtain multiple abnormal usage status data, and perform warning category division according to the multiple abnormal usage status data to obtain a housing warning result;

[0009] A warning unit, configured to send the housing warning result to the administrator according to a pre - configured communication interface.

[0010] Another technical solution of the present invention for solving the above - mentioned technical problem is as follows:

[0011] A rental housing anomaly warning method based on the Internet of Things, comprising the following steps:

[0012] Collect data on public rental housing through a variety of different types of sensors pre - arranged in public rental housing to obtain a monitoring data set;

[0013] Pre - process the monitoring data set, perform feature transformation on the pre - processed monitoring data set to obtain a housing status data set;

[0014] Extract features from the housing status data set to obtain multiple usage status features, perform anomaly analysis on multiple usage status features respectively to obtain multiple abnormal usage status data, and divide the warning categories according to multiple abnormal usage status data to obtain a housing warning result;

[0015] Send the housing warning result to the administrator according to a pre - configured communication interface.

[0016] The beneficial effects of the present invention are: the fusion of multi - dimensional data can provide more comprehensive and accurate housing usage status information, reducing the possibility of information loss and misjudgment. Based on a variety of different types of Internet of Things data, intelligent identification and abnormal warning analysis of the overall housing usage status can quickly respond to abnormal situations, avoid false alarms and missed alarms that are prone to occur in single - indicator warnings, analyze the abnormal usage status of tenants, and ensure the timeliness and accuracy of warnings. Brief Description of the Drawings

[0017] Figure 1 It is a unit block diagram of an abnormal warning device for rental housing based on the Internet of Things provided by an embodiment of the present invention;

[0018] Figure 2 It is a data collection schematic diagram provided by an embodiment of the present invention;

[0019] Figure 3 It is a data analysis schematic diagram provided by an embodiment of the present invention;

[0020] Figure 4 It is an abnormal warning processing schematic diagram provided by an embodiment of the present invention;

[0021] Figure 5 It is a system integration architecture diagram provided by an embodiment of the present invention;

[0022] Figure 6 It is a circuit connection schematic diagram provided by an embodiment of the present invention;

[0023] Figure 7 It is a timing diagram of abnormal warning for rental housing provided by an embodiment of the present invention;

[0024] Figure 8 It is an architecture diagram of an abnormal warning device for rental housing based on the Internet of Things provided by an embodiment of the present invention;

[0025] Figure 9Flowchart of the rental housing anomaly warning method based on the Internet of Things provided by the embodiments of the present invention. Detailed implementation manners

[0026] The principles and features of the present invention will be described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not intended to limit the scope of the present invention.

[0027] With the acceleration of the urbanization process, the demand for rental housing is increasing continuously. However, the management of rental housing faces many challenges, such as the illegal use status of tenants (such as subletting or group renting, etc.), damage to housing facilities and potential safety hazards, etc. These problems not only affect the rational allocation of housing resources, but also bring great management difficulties to the management department. The traditional rental housing anomaly warning management mainly relies on manual inspections and the conscious cooperation of tenants, with low management efficiency and difficulty in timely discovering and handling problems.

[0028] The rapid development of Internet of Things (IOT) technology provides new ideas and means for the management of rental housing. By deploying various sensor devices (such as smart door locks, cameras, and environmental sensors, etc.) in rental housing, the usage status data of the house is collected in real time, and the data includes but is not limited to the switch records of the door lock, the entry and exit situations of people, and the indoor environmental parameters. Using big data analysis and artificial intelligence technology to process and analyze these data, the intelligent identification and anomaly warning of the usage status of rental housing are realized. For example, by analyzing the door lock switch records, it is judged whether the house is being used normally and whether there are abnormal entry and exit usage statuses; by monitoring the indoor environmental parameters, potential safety hazards in the house can be discovered in time.

[0029] Although the existing rental housing anomaly warning management technologies have achieved intelligent management to a certain extent, there are still some technical drawbacks, including: relying only on a single data source (such as face recognition) for monitoring, lacking the integrated analysis of multi-dimensional IOT data. For example, it is difficult to comprehensively reflect the actual usage status of the house only through the smart door lock data. There are problems such as a high false alarm rate and untimely warning in the detection of abnormal usage statuses. For example, it cannot adapt to complex rental scenarios, resulting in inaccurate warnings. The degree of functional unitization is low, and it is difficult to effectively integrate with other management systems (such as rent management systems and maintenance management systems, etc.), resulting in a serious data island phenomenon. It is weak in the dynamic monitoring and warning of special groups such as the elderly and the elderly living alone, and it is impossible to discover abnormal situations in time.

[0030] As Figures 1-8 shown, an anomaly warning device for rental housing based on the Internet of Things provided by the embodiments of the present invention includes:

[0031] The acquisition unit is used to collect data of the public rental housing through a variety of different types of sensors pre - arranged in the public rental housing, so as to obtain a monitoring data set;

[0032] The pre - processing unit is used to pre - process the monitoring data set, perform feature transformation on the pre - processed monitoring data set, and obtain a housing status data set;

[0033] The analysis unit is used to extract features from the housing status data set to obtain multiple usage status features, perform anomaly analysis on multiple said usage status features respectively to obtain multiple abnormal usage status data, and perform early - warning category division according to multiple said abnormal usage status data to obtain a housing early - warning result;

[0034] The early - warning unit is used to send the housing early - warning result to the administrator according to a pre - configured communication interface.

[0035] Specifically, as Figure 6 and Figure 7 shown, a variety of different types of sensors include the intelligent door lock of the entrance door of the public rental housing, the camera outside the public rental housing, the camera inside the public rental housing, the environmental sensors inside the public rental housing (such as temperature sensors, smoke sensors, humidity sensors, light sensors), other Internet of Things devices inside the public rental housing (such as intelligent water meters, intelligent electricity meters, WIFI gas meters, and intelligent trash cans, etc.); the intelligent door lock circuit includes a door lock module and a communication module, the camera circuit includes a camera module and a communication module, and the environmental sensor circuit includes a sensor module and a communication module. The processing sequence of the rental housing anomaly early - warning device for early - warning is to first start the device for data acquisition, then transmit the data for data pre - processing, and finally analyze the data for early - warning judgment. Analyze the monitoring data set in time series, select a suitable time window (such as every hour, every day, or every week) to summarize the data. For example, taking every day as a time window, count the number of door openings per day, the longest door - opening duration, etc.

[0036] Furthermore, as Figure 8As shown in the figure, the architecture of the rental housing anomaly warning device includes an Internet of Things (IoT) device layer, a data collection layer, a data processing layer, a data application layer, and a data product operation layer; the IoT device layer is used to deploy various IoT devices (such as smart door locks, cameras, environmental sensors, and other IoT devices) in public rental housing; the data collection layer is used to store the monitoring data of various IoT devices (such as smart door lock data, camera monitoring data, environmental sensing data, etc.), preprocess the monitoring data, fuse the preprocessed monitoring data, and store it in the housing usage status database (i.e., store the housing status data set) of the data processing layer; the data processing layer is used to call the housing usage status database to the data application layer according to the topics of data mining (such as abnormal subletting); the data application layer generates user portraits, user behavior trajectory analysis, intelligent recognition, anomaly warning, and decision-making and disposal based on the housing status data; the data product operation layer is used to perform housing source management, special group care, property services, community services, and value-added services according to the data generated by the data application layer.

[0037] In the embodiment of the present invention, by integrating the data of multiple IoT devices such as smart door locks, cameras, and environmental sensors, the data analysis application scenarios are enriched. In the field of smart door locks, computer vision technology is combined to simulate human visual analysis. Through preprocessing, feature extraction, feature selection, and deep learning algorithm analysis, the detection, recognition, and classification of objects are realized. Using big data analysis and artificial intelligence technology, the collected IoT data is deeply mined and analyzed to realize the intelligent recognition of the usage status of rental housing and the accurate warning of abnormal usage status.

[0038] Preferably, as Figure 2 shown, in the data collection unit, the public rental housing is data-collected through a variety of different types of sensors arranged in the public rental housing to obtain a monitoring data set, including: user identification, door opening records, access frequency records, and vacant duration records through the smart door lock; video streams are collected through the camera to realize behavior recognition; data values of temperature, humidity, light, and dangerous gases are respectively collected through different environmental sensors; water usage, electricity consumption, WIFI gas, and garbage disposal are monitored through other devices.

[0039] Preferably, in the preprocessing unit, the monitoring data set is preprocessed, including:

[0040] Select monitoring data from the monitoring data set according to the set data categories through an ETL tool, clean the selected monitoring data, convert the format of the cleaned monitoring data, and load and store the converted monitoring data, and so on, until all categories of monitoring data in the monitoring data set are processed.

[0041] Specifically, the monitoring data includes initial door lock data, initial monitoring data and initial environment data; the set data categories are door lock category, monitoring category and environment category, and the initial door lock data is selected from the monitoring data set according to the door lock category through the ETL tool, and cleaned, converted and loaded; among them, the ETL (Extract-Transform-Load) tool is used to extract (extract), transform (transform) and load (load) data from the source end to the destination end.

[0042] Data cleaning includes removing duplicate, erroneous, null or irrelevant information to find identical or partially identical records in a data set; checking whether the data conforms to a predefined format, such as whether a date field is in a valid date format and whether a numeric field is a number; checking whether the data conforms to business logic, such as whether electricity consumption is a positive number; converting or correcting formatted data, and manually modifying logically incorrect data or using a rule engine to correct it; eliminating irrelevant information and regularly checking data quality to ensure data accuracy, completeness and consistency.

[0043] Furthermore, the data collected by different devices are converted into a unified format to facilitate subsequent processing.

[0044] The fingerprint data of the initial door lock data is formatted as follows: Fingerprint data is usually stored in the form of images or feature vectors. During the data formatting stage, if the fingerprint data is stored in the form of images, the resolution and format of all images must be consistent, so the image processing library is used to crop, scale and normalize the images. The fingerprint feature vector is extracted through the fingerprint recognition algorithm (Minutiae-based algorithm) and stored in a unified format (such as CSV or JSON).

[0045] The face data of the initial monitoring data is formatted as follows: Face data is usually stored in the form of images or videos. During the data formatting stage, the image processing library is used to crop, scale, and normalize the face images. If the face data collected by the camera contains annotation information (such as face location), the face data is converted to the format required by the detection model. Use a deep learning model (such as FaceNet) to extract the face feature vector and store it in a unified format (such as CSV or JSON).

[0046] Preferably, the monitoring data includes initial environmental data;

[0047] In the preprocessing unit, preprocessing the monitoring data set includes:

[0048] Select the initial environmental data from the monitoring data set according to the set data categories through an ETL tool, clean the selected initial environmental data, unify the units of the initial environmental data after cleaning, unify the time format of the initial environmental data after unit unification according to the set time benchmark, and load and store the initial environmental data in the unified format.

[0049] Specifically, environmental monitoring data usually includes sensor data such as temperature, humidity, and smoke. Format the initial environmental data: In the data formatting stage, standardize the data to ensure that the units and formats of all sensor data are consistent. For example, unify the temperature data to degrees Celsius and the humidity data to percentages. If the sensor data contains timestamps, it is necessary to ensure that the timestamp formats of all data are consistent and aligned to a unified time benchmark (such as YYYY-MM-DD HH:MM:SS) to achieve timestamp alignment. Among them, for sensor data with high-frequency collection, perform time window aggregation (such as taking the average value per minute) to reduce the data volume and improve processing efficiency. Among them, select the timestamp of a main sensor as the benchmark and perform interpolation processing on the data of other sensors to achieve a unified time benchmark.

[0050] Integrate multi-source data (i.e., formatted door lock data, monitoring data, and environmental data), select Hive technology to store massive data, support multiple storage models such as relational, document-based, and key-value, and form a complete database of the house usage status.

[0051] In the embodiments of the present invention, for the collected real-time data, such as door opening time, frequency, duration, etc. In the preprocessing stage, remove noise and missing values, and perform standardization or normalization processing on the data to eliminate the scale influence between different features, achieve the management and monitoring of data quality, and ensure that the data meets the predefined standards. The unified format of the data helps to directly interact with other management systems, achieve rapid connection to various data sources, and efficiently integrate data.

[0052] Preferably, the preprocessed monitoring data set includes door lock data, monitoring data, and environmental data;

[0053] In the preprocessing unit, perform feature transformation on the preprocessed monitoring data set to obtain a house status data set, including:

[0054] Perform feature transformation on the door lock data according to the time series to obtain door lock sequence features, screen out the key frames of the monitoring data according to the spatial sequence to obtain image features, and perform feature transformation on the environmental data according to the time series to obtain environmental sequence features.

[0055] Specifically, the monitoring data is a monitoring video. The optical flow algorithm (Lucas-Kanade optical flow method) is used to calculate the motion vectors between consecutive frames of the monitoring video (i.e., the spatial sequence of objects in the video frames). Motion features (such as the intensity and direction of the motion vectors) are extracted from the motion vectors. The difference in the change of the motion features in each frame image spatially from the motion features of the previous frame image is calculated. The frames with a difference greater than the threshold are selected as key frames, and multiple key frames (i.e., multiple image features) are obtained. The opening time, frequency, and duration of the intelligent door lock are converted into time series data (i.e., door lock sequence features); image features are extracted from the video frames of the camera; the readings of the water and electricity meters and the alarm data of the smoke sensor are converted into numerical sequences (i.e., environmental sequence features). The environmental data such as temperature, humidity, and smoke sensors and the visual data of the camera (i.e., monitoring data) are combined using a data fusion algorithm (such as Kalman filtering) to form more comprehensive environmental perception data.

[0056] In the embodiments of the present invention, the respective sensing data is characterized to facilitate analysis based on the data characteristics during anomaly analysis and improve the response efficiency.

[0057] Preferably, in the analysis unit, after the step of obtaining multiple abnormal usage status data, it further includes:

[0058] The image features are recognized through a face recognition model to obtain face features, and the image features are recognized through an age recognition model to obtain age features. The person type is divided according to the face features and the age features. Based on the person type, multiple abnormal usage statuses are matched with the set warning rules to obtain a person warning result.

[0059] It should be understood that the face recognition model is based on computer vision and deep learning technologies to identify or verify personal identity by analyzing face images or video sequences. The age recognition model uses deep learning technologies to predict age by analyzing face images. The face features are compared with the face features of the lease declaration to determine whether the lessee is correct, and the person type (such as a child, a young person, or an elderly person) is determined according to the age features.

[0060] Specifically, when an abnormal usage status is analyzed, identity verification is performed through the smart door lock and camera to determine whether it is a set special group. Real-time data is collected through the smart door lock and camera, and after preprocessing operations such as denoising, stabilizing the image, and correcting the illumination, key features are extracted, such as facial flatness, hair color, unlocking time, unlocking frequency, crowd density, movement direction, etc. The face recognition model (FaceNet) and age recognition model are applied to the video data (i.e., image features) to identify the age of the people in the camera in real time. For example, a solitary elderly person may frequently unlock the door during specific time periods (such as in the morning or evening), children may unlock the door after school, and the coming and going times of a solitary female are basically consistent with the work commuting time periods.

[0061] Warning rules are set according to the housing usage status patterns and living habits of the special group. Warnings are issued in a timely manner for abnormal situations and the management staff is notified for handling. For example, under normal circumstances, a solitary elderly person will open the door at least 1-2 times a day (such as going out to buy groceries, take out the trash, etc.). If there is no record of opening the door for several consecutive days, it may indicate that the elderly person is unwell or has encountered other problems. If a solitary female fails to close the door for a long time after opening it, it may be due to reasons such as receiving takeout or picking up a courier and not locking the door in time, triggering a reminder to improve residential safety. If an unauthorized door opening usage status is detected (such as multiple incorrect password attempts or fingerprints of non-registered users), there is a risk of burglary, triggering an alarm and notifying the management staff.

[0062] Furthermore, basic information of solitary elderly people, including age, gender, marital status, living environment, etc., is obtained by combining the public rental housing application data. The community registration system is accessed to supplement the detailed information of solitary elderly people, such as health status, family member situation, economic source, etc. The opening time frequency of solitary elderly people is recorded to judge whether their daily activities are normal. For example, monitoring data collected by cameras in public areas is obtained, and the activity trajectories of solitary elderly people are analyzed through computer vision technology to determine whether there is an abnormal usage status. Health monitoring devices, such as smart bracelets and health monitors, are accessed to monitor the vital signs (such as heart rate, blood pressure, etc.) of the elderly in real time and upload the data. All data is aggregated to establish a portrait label system for the special group, associated with care indicators, create data analysis rules and scheduled tasks for care themes, and analyze the daily activity patterns of solitary elderly people. For example, if the number of times the elderly person opens the door significantly decreases during a certain period, or if they do not appear in the public area for a long time, attention is required; computer vision technology is used to identify abnormal usage status of solitary elderly people, such as falling, not moving for a long time, etc.; the safety of the living environment is evaluated through environmental sensor data. When smoke or abnormal temperature is detected, an alarm is issued in a timely manner. Health and usage status reports of solitary elderly people are generated regularly for reference by community staff or family members. According to the warning information, community staff or volunteers are arranged to visit the door to provide necessary assistance, and spiritual comfort and chatting services are provided to solitary elderly people regularly.

[0063] In the embodiment of the present invention, a care mechanism for special groups is realized. For special groups such as the elderly and those living alone, abnormal situations are timely detected through data analysis and early warnings are issued, improving the humanized level of abnormal early warning management of rental housing.

[0064] Preferably, after the step of obtaining the housing status data set in the preprocessing unit, the following steps are further included:

[0065] Calculate the Pearson correlation coefficients between the housing status data in the housing status data set to obtain a plurality of statistical indicators;

[0066] Predict the contribution degrees of the housing status data set through the Shapley additive explanation model to obtain a plurality of contribution degrees;

[0067] Calculate the weights of the corresponding housing status data according to the plurality of statistical indicators and the plurality of contribution degrees to obtain a plurality of analysis weight values, and respectively assign each of the analysis weight values to the corresponding housing status data.

[0068] Specifically, in the initial stage of data analysis, a statistical indicator that measures the linear correlation degree between two features through Pearson correlation coefficient analysis. The features include multi-dimensional data such as face capture, access records, warning events, etc. The potential relationships between the data are mined, and a set of highly correlated indicators (i.e., variable set) is identified and applied to the design of the abnormal indicator analysis weight, thereby optimizing the management process of rental housing and improving the operation efficiency.

[0069] Introduce the SHAP model to assist in understanding the contribution degree of each feature to the model prediction result, and analyze multi-dimensional data such as face capture, access records, warning events, etc. First, evaluate the contribution degree of multi-dimensional data such as face capture, access records, warning events, etc. through the SHAP model; for example, analyze the features that are most critical for predicting the usage status mode of tenants (such as frequent access times, activity frequencies in specific areas, types of warning events, etc.), thereby optimizing the data collection and processing process and reducing unnecessary data storage and calculation costs. Second, analyze the access records and housing vacancy rate data through the SHAP model to optimize the allocation of housing resources; for example, if the access records in a certain area show a low utilization rate of public facilities and a high housing vacancy rate, then adjust the housing allocation strategy. Third, the SHAP model displays the complex data analysis results in an intuitive way; for example, display the contribution degree of each feature to the model prediction result through a SHAP value graph, and display it in a visual way, which helps the staff in the management center to better understand the data and thus make more reasonable decisions.

[0070] It should be understood that the Pearson correlation coefficient is a statistic used to measure the degree of linear correlation between two variables, with a value ranging from -1 to 1, where 1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 indicates no linear correlation.

[0071] The SHAP (SHapley Additive exPlanations) model, namely the Shapley Additive Explanation model, is a machine learning model interpretation tool based on game theory, used to quantify the contribution of each feature to the model prediction result. By calculating the Shapley value of each feature, it explains the decision-making process of the model.

[0072] The model prediction result refers to the association between multi-dimensional data such as frequent in-and-out times, activity frequencies in specific areas, warning events, etc. and the judgment result. For example, when judging whether there are abnormal visitors in a certain house, data with a high degree of association needs to be selected for calculation. Therefore, the ratio of the number of internal door openings to the number of external door openings and the time period when door openings are concentrated will become key elements, while other data may not be concerned. That is, after analyzing the key features, the calculation logic of the pre-built visualization report will be adjusted (such as including data with a high association contribution rate in the calculation logic or assigning a higher weight). The database still stores the full amount of Internet of Things data, and non-key features are no longer called in specific analysis scenarios.

[0073] In the data mining stage, an analysis model is established according to the business of the regulatory unit. The analysis model is constructed based on the abnormal index analysis weight obtained by analyzing the Pearson correlation coefficient and the calculation logic of the visualization report adjusted by the SHAP model. For example, if the focus is on rectifying subleases, a sublease analysis model is constructed to achieve a special analysis of abnormal subleases. After analyzing the abnormal usage status data, a comprehensive analysis is carried out according to the weights of the data to obtain the housing warning result.

[0074] In the embodiments of the present invention, by analyzing the correlation between various types of data, different analysis weights are assigned to each data, so as to perform abnormal analysis according to the weights, generate abnormal results, and then give a warning reminder for the housing situation.

[0075] Preferably, in the analysis unit, feature extraction is performed on the housing status data set to obtain multiple usage status features, including:

[0076] Feature extraction is performed on the housing status data set through a convolutional neural network to obtain multiple local features, and feature extraction is performed on the housing status data set through a deep neural network to obtain multiple hidden features. Multiple local features and corresponding hidden features are feature-stitched to obtain multiple usage status features.

[0077] It should be understood that a Convolutional Neural Network (CNN) is a deep learning architecture widely used in fields such as image recognition, video analysis, and natural language processing. It can automatically extract the features of input data to achieve efficient classification, detection, and regression tasks. A Deep Neural Network (DNN) is a neural network architecture with multiple hidden layers that can automatically learn complex feature representations of data. The convolutional neural network and the deep neural network are respectively trained through a pre-constructed training data set. The local features and hidden features are combined through methods such as feature concatenation, end-to-end training, or multi-task learning to form a more comprehensive behavior pattern.

[0078] Specifically, machine learning algorithms are used to learn the door-opening features, behavior patterns, etc. in the normal living mode. Supervised feature learning is used with labeled data, and the feature representations related to the task are directly optimized through the training model. The local features of the image are mainly extracted through the convolutional layer and pooling layer of the convolutional neural network, such as edges, textures, and shapes; the mapping relationship from the input data to the output data is automatically learned mainly through the multi-layer non-linear transformation of the deep neural network, and hidden features are extracted from complex door-opening features (such as time, frequency, duration, etc.). The extracted features can effectively distinguish the key features of normal and abnormal usage state patterns. For example, for the door-opening usage state, the extracted local features include: door-opening frequency (the number of times the door is opened per day), door-opening time (specific time periods, such as night, day), door-opening duration (the length of time for each door opening), and door-opening interval time (the time interval between two door openings); hidden features such as the internal door-opening frequency between 3 am and 7 am. Regularization is used to select the optimal features from the extracted features for subsequent abnormal data analysis.

[0079] In the embodiments of the present invention, various usage state features are generated according to the actual business requirements, which is convenient for analyzing the user's behavior pattern based on feature anomalies and timely supervising abnormal behaviors. In the behavior recognition task of the usage state, the CNN extracts the local spatial features in the video frame, while the DNN processes the time series data to capture the time dynamics of the behavior. By combining the features extracted by the CNN and the DNN, the behavior pattern of the usage state can be more comprehensively understood, thereby improving the recognition accuracy.

[0080] Preferably, as Figure 3 shown, in the analysis unit, abnormal analysis is respectively performed on multiple said usage state features to obtain multiple abnormal usage state data, including:

[0081] The usage status features are compressed through an autoencoder to obtain low-dimensional features, and the low-dimensional features are reconstructed to obtain approximate usage status features. The reconstruction error is calculated for the usage status features and the approximate usage status features. If the reconstruction error is greater than the set reconstruction threshold, it is determined that the usage status features are abnormal usage status data. Through this process, multiple usage status features are analyzed for anomalies to obtain multiple abnormal usage status data.

[0082] If the usage status features are time series data, the time series features are extracted from the usage status features through a recurrent neural network to obtain status features. The status error between the status features and the set normal status features is calculated. If the status error is greater than the set status threshold, it is determined that the usage status features are abnormal usage status data. Through this process, multiple usage status features are analyzed for anomalies to obtain multiple abnormal usage status data.

[0083] It should be understood that warning rules and thresholds are set according to business requirements and historical data. For example: Set the frequency threshold as: if the number of door openings exceeds the set threshold (such as more than 10 times a day), it is abnormal usage status data and a warning is triggered; Set the time threshold as: if the door is opened frequently during non-normal time periods (such as from 1 am to 5 am), it is abnormal usage status data and a warning is triggered; Set the duration threshold as: if the door opening duration is too long (such as more than 5 minutes), it is abnormal usage status data and a warning is triggered.

[0084] Furthermore, appropriate anomaly detection algorithms are selected according to data characteristics and application scenarios, including anomaly detection algorithms based on statistics or machine learning. The anomaly detection algorithm based on statistics is: using the Z-score (i.e., standard score or standardized value) to detect anomaly points; for example, when the Z-score of a certain data point exceeds the set threshold (such as 3 times the standard deviation), this data point is considered abnormal. Specifically, for each feature (such as door opening time, frequency, duration), its mean and standard deviation are calculated respectively; a threshold (usually plus or minus 3) is set. If the absolute value of the Z-score of a certain data point exceeds this threshold, this point is considered abnormal.

[0085] The anomaly detection algorithms based on machine learning include:

[0086] For high-dimensional data (such as abnormal behavior trajectory data), the Isolation Forest algorithm is used to construct isolation trees to identify abnormal usage status data.

[0087] Learn the low-dimensional representation of data by reconstructing the input data with an autoencoder. When the input data contains abnormal behaviors, the reconstruction error of the autoencoder will increase significantly, thus triggering an anomaly alarm. The autoencoder is used for anomaly detection. During the data input and reconstruction phase, various types of sensor data are normalized or standardized (i.e., convert the opening time, frequency, and duration of the smart door lock into time series data; extract video frames of the camera as image features; convert the readings of the water and electricity meters and the alarm data of the smoke sensor into numerical sequences). The input data (including the usage records of the smart door lock, video frames of the camera, readings of the smart water and electricity meters, and alarm data of the smoke sensor, etc.) is reconstructed in the autoencoder to generate a reconstruction result. Whether there is an anomaly is judged by calculating the reconstruction error. The autoencoder compresses the input data into a low-dimensional space through the encoder, and then tries to reconstruct the input data through the decoder. The reconstruction result is an approximate representation of the input data by the autoencoder. The reconstruction error is usually calculated through a loss function, such as the mean square error (MSE) or the mean absolute error (MAE). Judge whether the reconstruction error increases significantly according to the threshold (the average error plus one standard deviation) by analyzing the distribution of the reconstruction errors of normal data. If the reconstruction error of a certain data point exceeds this threshold, it is considered that there is an anomaly in this data point.

[0088] Apply the recurrent neural network (RNN) or its variants (such as LSTM, GRU) to time series data to capture time dependence and analyze the usage status pattern of residents for public rental housing. First, convert the opening time, frequency, duration of the smart door lock, as well as the readings of the water and electricity meters and the alarm of the smoke sensor into time series format according to time. Use the trained RNN or its variants (such as LSTM or GRU) to extract hidden states as features from the time series data. The hidden features represent the dynamic changes and context information in the time series. For example: Analyze the anomaly of the opening frequency according to the hidden features. If the feature deviates significantly from the normal range at a certain time point, it indicates an abnormal usage state; judge whether there is abnormal usage by analyzing the time series features of the readings of the water and electricity meters. Set a threshold to judge anomalies according to the feature distribution of normal data; or dynamically analyze the abnormal usage status pattern by analyzing the change trend of the features over time. Train the recurrent neural network (RNN) or its variants (such as LSTM, GRU) with a pre-built training set.

[0089] Furthermore, when performing anomaly analysis on the usage status features, dynamically adjust the reconstruction threshold and the status threshold, including:

[0090] Dynamically adjust the threshold according to the dynamic characteristics of the data using a sliding window or quantile calculation. By monitoring the changes in the data in real time and adaptively setting the threshold according to the data distribution, improve the accuracy and adaptability of anomaly detection to meet the anomaly detection requirements in different scenarios.

[0091] It should be understood that an autoencoder is a neural network model for unsupervised learning, mainly used to learn effective representations of data. By compressing the input data into a low-dimensional latent space and then attempting to reconstruct the original input data, functions such as feature extraction, dimensionality reduction, denoising, and anomaly detection can be achieved. Recurrent neural networks (RNNs) or their variants (such as LSTMs, GRUs) are important architectures in deep learning for processing sequential data, capturing the temporal dependencies and dynamic changes in sequences, and are widely used in fields such as natural language processing, time series analysis, and speech recognition.

[0092] In the embodiments of the present invention, the visual data captured by the camera can be fused with the biometric data of the intelligent door lock (such as fingerprint and face recognition), and the face features and usage status patterns can be analyzed simultaneously through deep learning algorithms. For example, it can identify whether the user is a legitimate user and determine whether they are in an abnormal state (such as being coerced).

[0093] Preferably, the separately performing anomaly analysis on multiple usage status features to obtain multiple abnormal usage statuses further includes:

[0094] Applying association rules to reveal the associations between different attributes, analyzing the associations between the door opening time and the pre-divided living areas, device usage (such as data monitoring of devices such as door locks, cameras, smoke alarms, etc.), and obtaining the usage status patterns within a specific time period. For example, for a housing unit located in a certain business district, the data shows that the household has regular access from Monday to Friday and no one enters on weekends, and by matching the age range of the tenant, it is identified as a housing type for urban young commuters.

[0095] Using regression analysis to establish a mathematical model to predict the relationship between any one usage status feature and other usage status features. For example, the relationship between the internal and external door opening frequencies and the visitor reception frequency, analyzing the patterns of residents' usage status of public rental housing, and then identifying abnormal behavior trajectories. Specifically, the visitor reception frequency is the number of times the tenant opens the door from the inside within a certain time period divided by two. Perform a regression analysis on the door opening frequency and the visitor reception frequency to generate a mathematical model. According to the preset judgment rules of the mathematical model (such as the average number of visitors received in a certain room on non-holiday days), determine whether there are abnormal visitors. If the tenant's reception frequency is significantly higher than the average level, it is concluded that abnormal visitors have appeared, and then the detailed data of this room is retrieved for further observation. The user's behavior trajectory is generated based on the data collected by the door lock (such as data on temporary authorization for visitors to open the door and visitor voice conversations of intelligent door locks).

[0096] In the embodiments of the present invention, the abnormal usage statuses for the special analysis of abnormal subletting include vacancy and subletting, etc. When the analysis result is vacancy, the administrator contacts the tenant to understand the situation and makes a plan such as applying for early lease termination.

[0097] Preferably, in the analysis unit, according to a plurality of the abnormal usage status data, early warning category division is performed to obtain a housing early warning result, including:

[0098] Classify a plurality of the abnormal usage status data through a clustering algorithm to obtain a plurality of early warning categories, and respectively match the plurality of early warning categories with a set early warning mode to obtain a housing early warning result.

[0099] It should be understood that the K-Means clustering (k-means clustering algorithm) is an iterative clustering analysis algorithm. If the data is divided into K groups, then K objects are randomly selected as the initial clustering centers, and the distance between each object and each seed clustering center is calculated, and each object is assigned to the nearest clustering center to achieve classification.

[0100] Specifically, the data is divided into K clusters through the K-Means clustering algorithm. The data points within each cluster are similar. For example, it is used to divide the door opening time series data into different patterns, such as leaving early and returning late, leaving late and returning early, vacant on weekends, vacant in winter, etc. The early warning mode can be set to SMS information prompts or warning pop-up windows, etc.

[0101] In the embodiment of the present invention, clustering analysis is performed according to the abnormal usage status data to generate abnormal types, and then the abnormal usage of public rental housing is managed.

[0102] Preferably, as Figure 4 shown, in the early warning unit, it is specifically used for:

[0103] Timely notify the management personnel by means of SMS and third-party APP push, etc.; push notification SMS by binding the tenant's mobile phone number, and trigger two-way calls in case of emergency for direct communication between the administrator and the tenant. Realize data interaction between users and the platform through the interface of a pre-configured third-party APP (such as user management software), including information query, device control, early warning notification, etc. The management personnel take timely measures according to the housing early warning result to handle abnormal situations. The set early warning mode includes directly triggering an artificial customer service call for events that need to be processed immediately, pushing SMS for silent processing matters, and synchronously notifying in the third-party APP.

[0104] In the embodiment of the present invention, when an abnormal usage status is detected, an alarm is triggered according to the set early warning mode, and relevant personnel are notified.

[0105] Preferably, as Figure 5 shown, data sharing is realized through the API interface among the public rental housing management system, the detection and early warning system (this device) and other management systems (such as the fund management system, the maintenance management system, the property management system and the community management system), avoiding data singularity and improving management efficiency.

[0106] Database storage uses common data formats such as JSON or XML to ensure data consistency and comparability, so that data can be connected in structure and format in various systems. Through the data integration platform (data warehouse, data lake or data middle platform), the data generated by various business systems can be centrally managed, which can support data sources in multiple data formats, including rental management, maintenance management and other systems, avoid single information, quickly connect various data sources, and efficiently integrate data. Using API (application programming interface) and microservice architecture, data can be easily exchanged between different systems and services. Share data from smart housing systems to smart community and smart property systems in real time. Formulate clear data policies and standards, including data classification, data quality and data security specifications, ensure compliance with relevant laws, regulations and privacy policies, and use encryption, desensitization and other technologies to protect sensitive data. Guide regulatory units to encourage communication and cooperation between different departments, establish cross-functional teams, and focus on data integration projects.

[0107] Preferably, the selected models (including face recognition models, age recognition models, mathematical models constructed by regression analysis, convolutional neural networks, deep neural networks, autoencoders and recurrent neural networks) are trained using historical data, and the model performance is evaluated by cross-validation. Common evaluation indicators include precision, recall and F1 score.

[0108] like Figure 9 As shown, an abnormal early warning method for rental housing based on the Internet of Things provided by an embodiment of the present invention includes the following steps:

[0109] Data from the public housing is collected by using various sensors pre-arranged in the public housing to obtain a monitoring data set;

[0110] Preprocessing the monitoring data set, performing feature conversion on the preprocessed monitoring data set to obtain a house status data set;

[0111] Extracting features from the house status data set to obtain a plurality of usage status features, performing abnormal analysis on the plurality of usage status features to obtain a plurality of abnormal usage status data, and classifying warning categories according to the plurality of abnormal usage status data to obtain a house warning result;

[0112] The house early warning result is sent to the administrator according to the preconfigured communication interface.

[0113] The beneficial effects of the present invention include: integrating data from multiple IOT devices (such as smart door locks, cameras, environmental sensors, etc.), including data cleaning, formatting, and multi-source data integration, which can effectively improve the integrity and accuracy of data, providing a reliable data basis for subsequent intelligent identification and early warning. Intelligent identification and abnormal early warning based on big data analysis and artificial intelligence technology can achieve accurate identification of the usage status of rental housing and real-time early warning of abnormal behaviors, including but not limited to abnormal situations such as subleasing, group renting, and long-term unoccupied. A highly integrated rental housing management system architecture that can achieve seamless docking with other systems such as rent management, maintenance management, and security monitoring. A monitoring and early warning mechanism for special groups such as the elderly and single elderly people, which can timely detect abnormal situations of special groups through data analysis and provide targeted early warning and emergency response. Through data cleaning algorithms, formatting methods, and multi-source data integration mechanisms, efficient data preprocessing and fusion technologies ensure the accuracy and real-time nature of data, and an abnormal early warning technology with fast response and low false alarm rate.

[0114] By integrating data from multiple IOT devices such as smart door locks, cameras, and environmental sensors, comprehensive monitoring of the usage status of rental housing is achieved. The integration of multi-dimensional data can provide more comprehensive and accurate information, reducing information loss and misjudgment. Using big data analysis and artificial intelligence technology to deeply mine and analyze the collected multi-dimensional data, intelligent identification of the usage status of rental housing and real-time early warning of abnormal behaviors are realized, which can improve the accuracy and real-time nature of abnormal detection, reduce the false alarm rate, and timely discover and handle problems. A highly integrated abnormal early warning management system (i.e., this device) that realizes docking with other systems such as rent management, maintenance management, and security monitoring can break data islands, achieve data sharing and functional collaboration, and improve management efficiency and collaboration ability. For special groups such as the elderly and single elderly people, abnormal situations are timely detected through data analysis, and a targeted early warning and emergency response mechanism is provided, which can improve the humanized level of rental housing management, reduce potential safety hazards, and meet the social care needs for special groups. By adopting efficient data preprocessing and fusion technologies to ensure the accuracy and real-time nature of data and achieve rapid response to abnormal situations, the response speed can be improved, the false alarm rate can be reduced, and the timeliness and accuracy of early warning can be ensured. By reducing the false alarm rate, improving management efficiency and collaboration ability, the management cost can be significantly reduced, and timely discovery and handling of problems can reduce potential economic losses.

[0115] For the above-mentioned method for abnormal early warning of rental housing based on the Internet of Things, reference can be made to the implementation content and its beneficial effects specifically described above for an abnormal early warning device for rental housing based on the Internet of Things, which will not be elaborated here.

[0116] It should be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0117] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described devices and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0118] In several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0119] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiments of the present invention.

[0120] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An abnormal early warning device for rental housing based on the Internet of Things, characterized in that: include: A collection unit is used to collect data from the public rental housing through a plurality of different types of sensors pre-arranged in the public rental housing to obtain a monitoring data set; A preprocessing unit, used for preprocessing the monitoring data set, performing feature conversion on the preprocessed monitoring data set to obtain a house status data set; An analysis unit is used to extract features from the house status data set to obtain a plurality of usage status features, perform abnormal analysis on the plurality of usage status features to obtain a plurality of abnormal usage status data, and classify warning categories according to the plurality of abnormal usage status data to obtain a house warning result; The early warning unit is used to send the house early warning result to the administrator according to the preconfigured communication interface.

2. The rental housing abnormality warning device according to claim 1 is characterized in that: In the preprocessing unit, preprocessing the monitoring data set includes: The monitoring data is selected from the monitoring data set according to the set data category through the ETL tool, the selected monitoring data is cleaned, the format of the cleaned monitoring data is converted, the converted monitoring data is loaded and stored, and so on, until all categories of monitoring data in the monitoring data set are processed.

3. The rental housing abnormality warning device according to claim 1, characterized in that: The pre-processed monitoring data set includes door lock data, monitoring data and environmental data; In the preprocessing unit, feature conversion is performed on the preprocessed monitoring data set to obtain a house status data set, including: The door lock data is feature-converted according to a time series to obtain door lock sequence features, key frames of the monitoring data are screened out according to a spatial sequence to obtain image features, and the environment data is feature-converted according to a time series to obtain environment sequence features.

4. The rental housing abnormality warning device according to claim 3 is characterized in that: In the analysis unit, after the step of obtaining a plurality of abnormal usage status data, the step further includes: The image features are identified by a face recognition model to obtain face features, and the image features are identified by an age recognition model to obtain age features. Character types are divided according to the face features and the age features, and based on the character types, the multiple abnormal usage states are matched with the set warning rules to obtain character warning results.

5. The rental housing abnormality warning device according to claim 1, characterized in that: In the preprocessing unit, after the step of obtaining the house status data set, the following steps are further included: Calculating the Pearson correlation coefficient between each house status data in the house status data set to obtain multiple statistical indicators; Predicting the contribution of the house status dataset by using the Shapley addition interpretation model to obtain multiple contribution degrees; The corresponding house status data is weighted according to the multiple statistical indicators and the multiple contribution degrees to obtain multiple analysis weight values, and each of the analysis weight values ​​is respectively assigned to the corresponding house status data.

6. The rental housing abnormality warning device according to claim 1, characterized in that: In the analysis unit, feature extraction is performed on the house status data set to obtain a plurality of usage status features, including: The house status data set is subjected to feature extraction through a convolutional neural network to obtain a plurality of local features, the house status data set is subjected to feature extraction through a deep neural network to obtain a plurality of hidden features, and the plurality of local features and corresponding hidden features are subjected to feature splicing to obtain a plurality of usage status features.

7. The rental housing abnormality warning device according to claim 1, characterized in that: In the analysis unit, abnormal analysis is performed on the plurality of usage status features respectively to obtain a plurality of abnormal usage status data, including: The usage status feature is compressed by an autoencoder to obtain a low-dimensional feature, and the low-dimensional feature is reconstructed to obtain an approximate usage status feature. Loss calculation is performed on the usage status feature and the approximate usage status feature to obtain a reconstruction error. If the reconstruction error is greater than a set reconstruction threshold, the usage status feature is determined to be abnormal usage status data. In this process, multiple usage status features are analyzed for abnormality to obtain multiple abnormal usage status data.

8. The rental housing abnormality warning device according to claim 1, characterized in that: In the analysis unit, abnormal analysis is performed on the plurality of usage status features respectively to obtain a plurality of abnormal usage status data, and further includes: If the usage status feature is time series data, the time series feature of the usage status feature is extracted through a recurrent neural network to obtain a state feature, and a state error between the state feature and a set normal state feature is calculated. If the state error is greater than a set state threshold, the usage status feature is determined to be abnormal usage status data. In this process, multiple usage status features are analyzed for abnormality to obtain multiple abnormal usage status data.

9. The rental housing abnormality warning device according to claim 1, characterized in that: In the analysis unit, warning categories are divided according to the plurality of abnormal usage status data to obtain a house warning result, including: The plurality of abnormal usage status data are classified by a clustering algorithm to obtain a plurality of warning categories, and the plurality of warning categories are matched with set warning modes respectively to obtain a house warning result.

10. A rental housing abnormality warning method based on the Internet of Things, characterized in that: The steps include: Data from the public rental housing is collected by using various types of sensors pre-arranged in the public rental housing to obtain a monitoring data set; Preprocessing the monitoring data set, performing feature conversion on the preprocessed monitoring data set to obtain a house status data set; Extracting features from the house status data set to obtain a plurality of usage status features, performing abnormal analysis on the plurality of usage status features to obtain a plurality of abnormal usage status data, and classifying warning categories according to the plurality of abnormal usage status data to obtain a house warning result; The house early warning result is sent to the administrator according to the preconfigured communication interface.

Images