Digital community governance management system and method based on artificial intelligence

Abstract

The invention discloses a digital community governance management system and method based on artificial intelligence, and relates to the technical field of big data analysis. A BIM modeling method is combined with a community building drawing to construct a community virtual model, residence units are divided in the virtual model, a database is established, mapping of the model and data is achieved through a Revit plug-in, and the system and the method have the advantages that the system and the method are simple and convenient, and the system and the method are easy to implement. The method comprises the steps of collecting resident behavior data through Internet of Things equipment, converting the resident behavior data into a third-order tensor, extracting spatial and temporal features by using a convolutional neural network and an LSTM network, optimizing a training model in combination with a cross entropy loss function, constructing a behavior feature library system and an activity label system, and realizing accurate pushing of information through matching degree calculation and dynamic pushing threshold setting. And finally, the data is updated according to the feedback information of the residents, a service closed loop is formed, the pertinence of activity pushing is improved, the community interaction experience is enhanced, and a data-driven intelligent solution is provided for intelligent community construction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data analysis, and particularly relates to a digital community governance management system and method based on artificial intelligence. BACKGROUND

[0002] With the construction of smart communities, traditional communities face multi-dimensional technical bottlenecks in activity pushing and resident behavior management. Resident spatial information and behavior data lack structured integration, and existing systems cannot map resident spatial information and building space models, resulting in insufficient visualization of community management, limited depth of resident behavior characteristic analysis, and inability to accurately analyze user behavior characteristics. Communities rely on the experience of staff when pushing activities, resulting in low interest in resident participation in community activities, low efficiency of community and resident interaction, low user experience, and lack of feedback iteration mechanism, making it difficult to update resident behavior based on resident feedback on activities, and making it difficult to continuously optimize the accuracy of activity pushing. SUMMARY

[0003] The present application aims to provide a digital community governance management system and method based on artificial intelligence to solve the problems in the prior art.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a digital community governance management method based on artificial intelligence, the method comprising the following steps:

[0005] Step S1, using a BIM modeling method to establish a virtual community model for the community;

[0006] Step S1-1, collecting community building plan, elevation, section, structure, building coordinate points, and layer height data, and converting the drawings to CAD files at a unified drawing scale;

[0007] Step S1-2, importing the building CAD drawing file into the BIM software, and dividing the residential units in the community building according to the building coordinate points and layer height data using the BIM software;

[0008] Step S1-3, setting each residential unit to correspond to one family, and each family containing n residents.

[0009] Step S2, collecting resident information, and mapping the resident information to the virtual community model based on the virtual community model;

[0010] Step S2-1, collecting resident information using an electronic registration form online and a paper registration form in the property center offline;

[0011] Step S2-2, setting a unique identification ID of a residential unit according to a building number, a unit number and a doorplate number, and setting a unique identification ID of a resident;

[0012] Step S2-3, creating a residential unit table Residence_Unit, a resident table Resident and an intermediate mapping table BIM_Unit_Mapping in the database according to the unique identification ID of the residential unit and the unique identification ID of the resident, and setting that one residential unit in the residential unit table can correspond to n residents;

[0013] Step S2-4, setting a field Unit_ID in the residential unit table and the resident table, in the residential unit table, Unit_ID is a primary key of the residential unit table, in the resident table, Unit_ID is a foreign key, pointing to the primary key in the residential unit table, and the corresponding relationship between the residential unit and the resident is established in the database according to Unit_ID;

[0014] Step S2-5, setting a field BIM_Element_ID in the intermediate mapping table, and establishing the mapping relationship between the residential unit and the resident and the community virtual model in the intermediate mapping table through BIM_Element_ID and Unit_ID;

[0015] Step S2-6, obtaining the component ElementID of each residential unit through a Revit API plug-in in a BIM software, and storing the ElementID as BIM_Element_ID in the BIM_Unit_Mapping table and binding the ElementID with the corresponding Unit_ID;

[0016] Step S2-7, importing the residential unit ID, the resident ID and the corresponding relationship into the residential unit table and the resident table through a SQL script;

[0017] Through the step, the system can intuitively show the resident information to an operator, and the operator can conveniently search.

[0018] Step S3, collecting resident behavior characteristic data, analyzing the resident behavior characteristic data by using a deep learning model, filtering user behavior characteristics, labeling the resident behavior characteristics, constructing a resident behavior characteristic library, collecting community activity information, extracting an activity subject, analyzing the activity subject and generating an activity label;

[0019] Step S3-1, collecting resident behavior characteristic data by using an Internet of Things device, eliminating information related to the resident identity in the behavior characteristic data, and only keeping the resident behavior characteristic data:

[0020] Step S3-2, defining an original feature data set, and the original feature data set containing multi-dimensional behavior characteristic data of each resident;

[0021] Step S3-3, normalizing the original data set;

[0022] By normalizing the household behavior feature data, the correlation of household behavior features is effectively captured, providing a data basis for subsequent data analysis;

[0023] Step S3-4, converting the normalized data into a three-order tensor, representing the three dimensions of households, time windows and behavior categories, to capture the spatio-temporal correlation of behavior features;

[0024] Step S3-5, processing the tensor data through a convolution layer to extract local correlation features of different behaviors in time series;

[0025] Step S3-6, using the gating mechanism of the long short-term memory network (LSTM) to process the time series data, specifically:

[0026] The forgetting gate calculates the weight according to the household state at the previous time and the behavior at the current time, and determines the degree of retaining the historical behavior features of the household according to the calculated value;

[0027] The input gate calculates the weight according to the behavior features of the household at a certain time, and when the calculated value is in [x, 1], the household behavior features are included in the long-term behavior features of the household, x is a judgment threshold set according to experience;

[0028] Set the threshold n of the number of times of user behavior features, the output gate judges the household behavior feature data with a number of times less than n as temporary feature data, and the output gate calculates the weight for the temporary feature data, filters the new behavior features of the household, and includes the filtered temporary feature data of the household in the long-term behavior feature category;

[0029] Step S3-7, manually label the household behavior feature data to generate corresponding real labels, build a labeling model, input the manually divided label data into the labeling model, train the labeling model, label the household behavior features according to the trained labeling model, and optimize the model parameters through the cross-entropy loss function. Specifically:

[0030]

[0031] In the formula, L is the multi-class cross-entropy loss value, which quantifies the difference between the model prediction result and the real label, y i,k is the real label of the kth behavior of the ith household, is the model prediction probability, K is the number of label categories, and the labeled behavior feature data is generated and integrated using the labeling model, and a user behavior feature library is constructed;

[0032] Step S3-8, processing the community activity text using natural language technology to identify the activity subject and generate activity labels.

[0033] The gating mechanism is used to filter and extract various behavior characteristic data of the household, thereby improving the effectiveness and accuracy of the household behavior characteristic library. The behavior standard library is constructed by combining artificial labeling and model training, thereby ensuring that the labels are highly consistent with the actual behavior of the household.

[0034] Step S4, integrating the household behavior label library and the activity label, and realizing the bidirectional accurate information push of the household and the community activity according to the label marking of the household behavior characteristic classification and the activity label;

[0035] Step S4-1, inputting the household behavior label and the activity label:

[0036] P i = [p i1 , p i2 ,..., p ik ];

[0037] In the formula, P i1 , P i2 , P i3 ,..., P ik represent the first, second,..., and kth activity labels of the household i;

[0038] T j = [t j1 , t j2 ,..., t jm};

[0039] In the formula, j represents the activity number, t j1 , t j2 ,..., t jm represent the first, second,..., and mth label of the jth activity, and T j is the label set of the jth activity;

[0040] Step S4-2, calculating the correlation degree of the household behavior label and the activity label, specifically:

[0041]

[0042] In the formula, a kj represents the correlation degree of the behavior category k and the activity j, and the value range is [0, 1]. C k is the behavior category keyword library predefined by natural language processing;

[0043] Step S4-3, calculating the matching degree of the household behavior characteristic and the activity label:

[0044]

[0045] wherein s ij is the matching degree of the household behavior characteristics and the activity label, w k is the weight of the behavior category k, p ik is the probability that the household i belongs to the behavior category k, a kj is the association degree between the behavior category k and the activity j, K is the total number of behavior categories, k is the behavior category index, indicating the kth behavior category, and the matching degree of the household behavior characteristics and each activity label is calculated;

[0046] Step S4-4, when the household pushes information to the activity, for the household i, the activity j with s ij > θ user is pushed in descending order of s ij , θ user is the push threshold, which is set by professionals through data analysis and can be dynamically adjusted according to the actual scene;

[0047] Step S4-5, when the activity pushes information to the household, for the activity j, the household i with s ij > θ event is pushed, θ event is the activity coverage threshold, which is set by professionals through data analysis and can be dynamically adjusted according to the actual scene;

[0048] Step S4-6, when the activity is pushed, both online and offline modes are included, online push is performed using the community APP, and offline push is performed by staff according to the household information marked by the community virtual model.

[0049] By quantifying the association degree and matching degree of the household behavior characteristics and the activity label, the bidirectional accurate matching of the household behavior characteristics and the activity is realized, the pertinence of the community activity accurate push is improved, the push efficiency is improved, the dynamic threshold is set to ensure the dynamic adjustment of the activity type and the household behavior characteristics, the information overload is avoided, the household interests and hobbies are mined through data analysis, the enthusiasm of the household to participate in the community activity is enhanced, and a good community service experience is provided for the household.

[0050] Step S5, according to the feedback of the household participating in the activity after the information push, the behavior characteristic data is updated and the updated data is re-labeled.

[0051] Step S5-1, collect user feedback data, eliminate information directly associated with user identity, and only keep feedback data;

[0052] Step S5-2, then update the household information of each table in the database table;

[0053] Step S5-3, merge the feedback data with the original behavior characteristic data to generate a new training set D', and re-standardize the data by using the method of step S3-3;

[0054] Step S5-4, generate a new third-order tensor by using step S3-4, and update the model parameters by steps S3-5 to S3-7;

[0055] Step S5-5, generate a new abstract feature vector through the output of the intermediate layer of the trained model;

[0056] Step S5-6, classify the new feature vector by step S3-8 to generate a new label;

[0057] Step S5-7, re-perform the accurate push of the policy and the household information according to step S4.

[0058] A digital community governance and management system based on artificial intelligence, the system comprises a model construction module, an information mapping module, a behavior analysis module, an accurate push module and a feedback updating module;

[0059] The model construction module comprises a drawing processing unit and a residential unit division unit;

[0060] The information mapping module comprises an information collection unit and a data mapping unit;

[0061] The behavior analysis module comprises a data preprocessing unit and a feature modeling unit;

[0062] The accurate push module comprises a matching calculation unit and a push optimization unit;

[0063] The feedback updating module is used for sending the collected information to the accurate push module.

[0064] Compared with the prior art, the beneficial effects of the present application are:

[0065] 1. Through the standardization processing of the community household behavior characteristic data and the deep modeling of the behavior characteristic data, the behavior characteristic data and the activity label have calculability through multi-dimensional data acquisition and space-time correlation analysis.

[0066] 2. The present application realizes the accurate push of information push by constructing a two-way accurate push mechanism and quantitatively calculating the behavior and activity matching degree of the household based on the label correlation.

[0067] 3. The present application actively excavates the interests and hobbies of the household by analyzing the household behavior characteristic data and the activity label, provides the household with a service experience that fits the behavior characteristics of the household, and improves the happiness index of the household. BRIEF DESCRIPTION OF DRAWINGS

[0068] Figure 1 A flowchart of a digital community governance management method based on artificial intelligence according to the present application;

[0069] Figure 2 A structural diagram of a digital community governance management system based on artificial intelligence according to the present application. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0071] Step S1, using a BIM modeling method to establish a virtual community model for the community;

[0072] Step S1-1, collecting community building plan, elevation, section, structure, building coordinate points, and layer height data, and converting the drawings into CAD files at a unified drawing scale;

[0073] Step S1-2, importing the building CAD drawing files into the BIM software, and dividing the residential units in the community building according to the building coordinate points and layer height data using the BIM software;

[0074] Step S1-3, setting each residential unit to correspond to one family, and each family to contain n residents.

[0075] Step S2, collecting resident information, mapping the resident information to the virtual community model based on the virtual community model;

[0076] Step S2-1, collecting resident information using an electronic registration form online and a paper registration form in the property center offline;

[0077] Step S2-2, setting a unique identification ID for the residential unit according to the building number, unit number, and house number, and setting a unique identification ID for the resident;

[0078] Step S2-3, creating a residential unit table Residence_Unit, a resident table Resident, and an intermediate mapping table BIM_Unit_Mapping in the database according to the unique identification ID of the residential unit and the unique identification ID of the resident, and setting that one residential unit in the residential unit table can correspond to n residents;

[0079] Step S2-4, setting a field Unit_ID in the residential unit table and the household table, in the residential unit table, Unit_ID is the primary key of the residential unit table, in the household table, Unit_ID is a foreign key, pointing to the primary key in the residential unit table, and the correspondence between the residential unit and the household in the database is established according to Unit_ID;

[0080] Step S2-5, setting a field BIM_Element_ID in the intermediate mapping table, and establishing a mapping relationship between the residential unit and the household and the community virtual model in the intermediate mapping table through BIM_Element_ID and Unit_ID;

[0081] Step S2-6, obtaining the component ElementID of each residential unit through the Revit API plug-in in the BIM software, and storing it as BIM_Element_ID in the BIM_Unit_Mapping table and binding it with the corresponding Unit_ID;

[0082] Step S2-7, importing the residential unit ID, household ID and corresponding relationship into the residential unit table and the household table through a SQL script;

[0083] Through this step, the system can intuitively show the household information to the operator, and facilitate the operator to search.

[0084] Step S3, collecting household behavior characteristic data, analyzing the household behavior characteristic data by using a deep learning model, filtering user behavior characteristics, labeling the household behavior characteristics, constructing a household behavior characteristic library, collecting community activity information, extracting activity subjects, and analyzing the activity subjects to generate activity labels;

[0085] Step S3-1, collecting household behavior characteristic data by using an Internet of Things device, eliminating information related to the household identity in the behavior characteristic data, and only retaining the household behavior characteristic data:

[0086] Step S3-2, defining an original feature data set, and the original feature data set contains multi-dimensional behavior characteristic data of each household;

[0087] Step S3-3, standardizing the original data set;

[0088] Through the standardization processing of the household behavior characteristic data, the correlation of the household behavior characteristics is effectively captured, and a data basis is provided for subsequent data analysis;

[0089] Step S3-4, converting the standardized data into a three-order tensor, representing three dimensions of households, time windows and behavior categories respectively, so as to capture the spatio-temporal correlation of behavior characteristics;

[0090] Step S3-5, processing the tensor data through a convolutional layer to extract local correlation features of different behaviors in the time series;

[0091] Step S3-6, processing the time series data using the gating mechanism of the long short-term memory network (LSTM), specifically:

[0092] The forget gate calculates the weight based on the resident state at the previous time and the behavior at the current time, and determines the degree of retention of the resident's historical behavior features based on the calculated value;

[0093] The input gate calculates the weight based on the behavior features of the resident at a certain time, and when the calculated value is in [x, 1], the resident behavior features are included in the long-term behavior features of the resident, x being a judgment threshold set according to experience;

[0094] Set a threshold n for the number of times a user behavior feature appears, and the output gate judges the resident behavior feature data with a number of times less than n as temporary feature data. The output gate calculates the weight for the temporary feature data and filters the new behavior features of the resident, and the filtered temporary feature data of the resident is included in the long-term behavior feature category;

[0095] Step S3-7, manually label the resident behavior feature data to generate corresponding real labels, build a labeling model, input the manually labeled data into the labeling model, train the labeling model, and label the resident behavior features according to the trained labeling model. The model parameters are optimized through the cross-entropy loss function, specifically:

[0096]

[0097] In the formula, L is the multi-class cross-entropy loss value, which quantifies the difference between the model prediction result and the real label, y i,k is the real label of the kth behavior of the ith resident, is the model prediction probability, and K is the number of label categories. The labeling model is used to generate labeled behavior feature data and integrate and store it to build a user behavior feature library.

[0098] Step S3-8, use natural language technology to process community activity text and identify activity subjects to generate activity labels.

[0099] The gating mechanism is used to filter and extract various behavior feature data of the residents, improving the effectiveness and accuracy of the resident behavior feature library. The combination of manual labeling and model training is used to build a behavior standard library, ensuring that the labels are highly consistent with the actual behavior of the residents.

[0100] Step S4, integrate the household behavior tag library and the activity tag, and realize the two-way accurate information push of the household and the community activity according to the label marking of the household behavior characteristic classification tag and the activity tag;

[0101] Step S4-1, input the household behavior tag and the activity tag:

[0102] P i = [p i1 , p i2 , …, p ik ];

[0103] In the formula, P i1 , P i2 , P i3 , …, P ik represent the first, second, …, and kth activity tags of the household i;

[0104] T j = [t j1 , t j2 , …, t jm};

[0105] In the formula, j represents the activity number, t j1 , t j2 , …, t jm represent the first, second, …, and mth tag of the jth activity, and T j is the tag set of the activity j;

[0106] Step S4-2, calculate the correlation degree of the household behavior tag and the activity tag, specifically:

[0107]

[0108] In the formula, a kj represents the correlation degree of the behavior category k and the activity j, the value range is [0, 1], C k is the behavior category keyword library predefined through natural language processing;

[0109] Step S4-3, calculate the matching degree of the household behavior characteristics and the activity tag:

[0110]

[0111] In the formula, s ij is the matching degree of the household behavior characteristics and the activity tag, w k is the weight of the behavior category k, p ik is the probability that the household i belongs to the behavior category k, and a kjis the correlation between behavior category k and activity j, K is the total number of behavior categories, k is the behavior category index, indicating the kth behavior category, by calculating the matching degree between the resident behavior characteristics and each activity label;

[0112] Step S4-4: When the resident pushes information to the activity, filter s for resident i ij >θ user Activity j, press s ij Descending push, θ user To push thresholds, the thresholds are set by professionals through data analysis and can be dynamically adjusted according to actual scenarios;

[0113] Step S4-5: When the activity pushes information to the residents, filter s for activity j ij >θ event Resident i, targeted push, θ event The threshold for activity coverage is set by professionals through data analysis and can be dynamically adjusted according to actual scenarios;

[0114] Step S4-6: When pushing activities, there are two modes, online and offline. Online, the community APP is used for pushing, and offline, the staff pushes according to the resident information marked by the community virtual model.

[0115] By quantifying the correlation and matching degree between residents' behavioral characteristics and activity tags, we achieve two-way precise matching between residents' behavioral characteristics and activities, improve the targeted push of community activities, and improve push efficiency. Setting dynamic thresholds ensures dynamic adjustment of activity types and residents' behavioral characteristics, avoiding information overload. Through data analysis, we explore residents' interests and hobbies, enhance their enthusiasm for participating in community activities, and provide them with a good community service experience.

[0116] Step S5: Update the behavioral feature data based on the feedback from the residents participating in the activity after the information is pushed, and re-label the updated data.

[0117] Step S5-1: Collect user feedback data, remove information directly related to the user's identity, and retain only the feedback data;

[0118] Step S5-2: then update the household information in each table in the database;

[0119] Step S5-3: Merge the feedback data with the original behavior feature data to generate a new training set D', and re-standardize the data using the method of step S3-3;

[0120] Step S5-4: Generate a new third-order tensor using step S3-4, and update the model parameters through steps S3-5 to S3-7;

[0121] Step S5-5, generate a new abstract feature vector through the intermediate layer output of the trained model;

[0122] Step S5-6, generate a new label by classifying the new feature vector in step S3-8;

[0123] Step S5-7, re-perform the accurate push of the policy and the household information according to step S4.

[0124] A digital community governance and management system based on artificial intelligence, the system comprises a model construction module, an information mapping module, a behavior analysis module, an accurate push module and a feedback updating module;

[0125] The model construction module comprises a drawing processing unit and a residential unit division unit;

[0126] The information mapping module comprises an information collection unit and a data mapping unit;

[0127] The behavior analysis module comprises a data preprocessing unit and a feature modeling unit;

[0128] The accurate push module comprises a matching calculation unit and a push optimization unit;

[0129] The feedback updating module is used to send the collected information to the accurate push module.

[0130] Embodiment one: collect the building drawings of the community, combine the building coordinate points and the floor height data to establish a three-dimensional model, import the Revit plug-in to establish a three-dimensional model to divide 360 residential units, set each residential unit to correspond to a family, the number of households is 3, collect the household information, build a database table, enter the household information, realize the mapping of the three-dimensional model and the household information through BIM_Element_ID and Unit_ID, collect the household behavior feature data through the Internet of Things equipment, construct it into a three-order tensor, filter the household behavior feature data using the LSTM model, input gate: continuous 21 days of fitness behavior, the result is > threshold value x = 0.7, the household behavior feature is included in the long-term behavior feature of the household, output gate: the occurrence of single late-night entry is <3, the behavior is filtered, use artificial annotation + trained model to label the long-term fitness behavior as a sports lover, use cross-entropy loss optimization, automatically label it as a new household behavior, such as basketball activity, use NLP to extract the label, then input the household label and activity label, calculate the correlation degree through keyword matching, and calculate the matching degree to obtain s ij=0.8, and finally, according to the threshold, the qualified households are pushed to participate in the basketball activity through the APP, and according to the household information in the community virtual model, the activity leaflets are sent to the households offline; after the households complete the activity, the system collects feedback information, updates the database according to the feedback information, and then reanalyzes the behavior characteristics of new households.

[0131] It will be apparent to those skilled in the art that the application is not limited to the details of the above-exemplified embodiments and that the application can be implemented in other particular forms without departing from the spirit or essential characteristics thereof. The embodiments should therefore be considered in all respects as illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the above description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. No reference signs in the claims should be considered as limiting the scope of the claims to the features to which the reference signs are attached. The application covers the following aspects and their combinations.

Claims

1. A digital community governance and management method based on artificial intelligence, characterized by: The method comprises the following steps: Step S1: Use BIM modeling method to establish a virtual community model for the community; Step S2: Collect household information and map the household information to the virtual community model based on the virtual community model; Step S3: Collect resident behavior feature data, analyze the resident behavior feature data using a deep learning model, filter user behavior features, label the resident behavior features, build a resident behavior feature library, collect community activity information, extract activity subjects, analyze the activity subjects and generate activity labels; Step S4: Integrate the resident behavior tag library and activity tags, and implement two-way accurate information push between residents and community activities based on the tags classified by resident behavior characteristics and activity tags; Step S5: Update the behavioral feature data based on the feedback from the residents participating in the activity after the information is pushed, and re-label the updated data.

2. The artificial intelligence-based digital community governance and management method according to claim 1, characterized in that: The specific steps of step S1 are as follows: Step S1-1: Collect community building plans, elevations, sections, structural drawings, building coordinate points, and floor height data, and convert the drawings into CAD files at a unified drawing scale; Step S1-2: Import the building CAD drawing file into the BIM software, and use the BIM software to divide the residential units in the community building according to the building coordinate points and floor height data; Step S1-3: Assume that each residential unit corresponds to a household, and each household contains n residents.

3. The artificial intelligence-based digital community governance and management method according to claim 2, characterized in that: The specific steps of step S2 are as follows: Step S2-1: Collect household information using the electronic registration form online and the paper registration form at the property management center offline; Step S2-2: Set the unique ID of the residential unit and the unique ID of the resident according to the building number, unit number and house number; Step S2-3: Create a residential unit table Residence_Unit, a resident table Resident, and an intermediate mapping table BIM_Unit_Mapping in the database according to the residential unit unique identification ID and the resident unique identification ID, and set that one residential unit in the residential unit table can correspond to n residents; Step S2-4: Set the field Unit_ID in the residential unit table and the household. In the residential unit table, Unit_ID is the primary key of the residential unit table. In the household table, Unit_ID is a foreign key pointing to the primary key in the residential unit table. The correspondence between residential units and households is established in the database based on Unit_ID. Step S2-5: Set the field BIM_Element_ID in the intermediate mapping table, and establish a mapping relationship between the residential unit and the resident and the community virtual model in the intermediate mapping table through BIM_Element_ID and Unit_ID; Step S2-6: Obtain the component ElementID of each residential unit through the Revit API plug-in in the BIM software, store it as BIM_Element_ID in the BIM_Unit_Mapping table, and bind it to the corresponding Unit_ID; Step S2-7: Import the residential unit ID, household ID and corresponding relationship into the residential unit table and household table through SQL script.

4. The artificial intelligence-based digital community governance and management method according to claim 3 is characterized by: The specific steps of step S3, collecting household behavior feature data, analyzing the household behavior feature data using a deep learning model, and filtering user behavior features are as follows: Step S3-1: Collect resident behavioral characteristic data using IoT devices, remove information related to resident identity from the behavioral characteristic data, and retain only the resident behavioral characteristic data; Step S3-2: defining an original feature dataset, where the original feature dataset includes multi-dimensional behavioral feature data of each household; Step S3-3, standardizing the original data set; Step S3-4: Convert the standardized data into a third-order tensor, representing the three dimensions of household, time window, and behavior category, to capture the spatiotemporal correlation of behavioral characteristics; Step S3-5: Process the tensor data through the convolution layer to extract the local correlation features of different behaviors in the time series; Step S3-6: Use the gating mechanism of the long short-term memory network (LSTM) to process time series data. Specifically: The forget gate calculates the weights based on the resident's previous state and current behavior, and determines the degree to which the resident's historical behavior characteristics are retained based on the calculated value; The input gate sets weights based on the household's behavioral characteristics at a certain moment and performs calculations. When the calculated value is in [x, 1], the household's behavioral characteristics are included in the household's long-term behavioral characteristics. x is the judgment threshold set based on experience. A threshold n of the number of occurrences of user behavior characteristics is set, and the output gate determines the household behavior characteristic data with a number of occurrences less than n as temporary characteristic data. The output gate sets weights for the temporary characteristic data and performs calculations to filter the new behavior characteristics of the residents, and then includes the filtered temporary characteristic data of the residents into the long-term behavior characteristic category of the residents.

5. The artificial intelligence-based digital community governance and management method according to claim 4 is characterized by: The specific steps of step S3 for labeling the resident behavior characteristics and building the resident behavior characteristics database are as follows: Step S3-7: Manually label the household behavior feature data, generate corresponding real labels, build a labeling model, input the manually labeled data into the labeling model, train the labeling model, label the household behavior features according to the trained labeling model, and optimize the model parameters using the cross-entropy loss function. Specifically: Where L is the multi-classification cross entropy loss value, which quantifies the difference between the model prediction result and the true label, and y i,k is the true label of the k-th behavior of the i-th household, is the model prediction probability, K is the number of label categories, and the annotation model is used to generate labeled behavioral feature data and integrate and store them to build a user behavior feature library.

6. The artificial intelligence-based digital community governance and management method according to claim 5, characterized in that: The specific steps of step S3, which collects community activity information, extracts activity subjects, and analyzes the activity subjects to generate activity tags, are as follows: Step S3-8: Use natural language technology to process community activity text, identify activity subjects and generate activity tags.

7. The artificial intelligence-based digital community governance and management method according to claim 6, characterized in that: The specific steps of step S4 integrating the resident behavior tag library and the activity entity tag library are as follows: Step S4-1: Enter resident behavior tags and activity tags: P i =[p i1 ,p i2 ,...,p ik ]; Where, P i1 、P i2 、P i3 ,...,P ik represents the 1st, 2nd, ..., kth activity labels of household i; T j ={t j1 ,t j2 ,...,t jm }; Where j represents the activity number, t j1 , t j2 ,...,t jm represents the 1st, 2nd, ..., mth labels of the jth activity, T j is the tag set of activity j; Step S4-2: Calculate the correlation between the household behavior label and the activity label, specifically: Where a kj represents the correlation between behavior category k and activity j, with a value range of [0,1], C k A keyword library for predefined behavior categories through natural language processing.

8. The artificial intelligence-based digital community governance and management method according to claim 7, characterized in that: The specific steps of step S4 for implementing bidirectional accurate information push of residents and community activities based on the label marking of resident behavior characteristics and activity entity labels are as follows: Step S4-3: Calculate the matching degree between household behavior characteristics and activity tags: Where s ij is the matching degree between household behavior characteristics and activity labels, w k is the weight of behavior category k, p ik is the probability that household i belongs to behavior category k, a kj is the correlation between behavior category k and activity j, K is the total number of behavior categories, k is the behavior category index, indicating the kth behavior category, by calculating the matching degree between the resident behavior characteristics and each activity label; Step S4-4: When the resident pushes information to the activity, filter s for resident i ij >θ user Activity j, press s ij Descending push, θ user To push thresholds, the thresholds are set by professionals through data analysis and can be dynamically adjusted according to actual scenarios; Step S4-5: When the activity pushes information to the residents, filter s for activity j ij >θ event Resident i, targeted push, θ event The threshold for activity coverage is set by professionals through data analysis and can be dynamically adjusted according to actual scenarios; Step S4-6: When pushing activities, there are two modes, online and offline. Online, the community APP is used for pushing, and offline, the staff pushes according to the resident information marked by the community virtual model.

9. The artificial intelligence-based digital community governance and management method according to claim 8, characterized in that: The specific steps of step S5 are as follows: Step S5-1: Collect user feedback data, remove information directly related to the user's identity, and retain only the feedback data; Step S5-2: then update the household information in each table in the database; Step S5-3: Merge the feedback data with the original behavior feature data to generate a new training set D', and re-standardize the data using the method of step S3-3; Step S5-4: Generate a new third-order tensor using step S3-4, and update the model parameters through steps S3-5 to S3-7; Step S5-5: Generate a new abstract feature vector through the output of the trained model intermediate layer; Step S5-6: classify the new feature vector through step S3-8 to generate a new label; Step S5-7: Re-produce accurate push of policies and resident information according to step S4.

10. A digital community governance and management system based on artificial intelligence, characterized by: The system includes the following modules: The system includes a model building module, an information mapping module, a behavior analysis module, a precision push module and a feedback update module; The model building module includes a drawing processing unit and a residential unit division unit; The information mapping module includes an information collection unit and a data mapping unit; The behavior analysis module includes a data preprocessing unit and a feature modeling unit; The precise push module includes a matching calculation unit and a push optimization unit; The feedback update module is used to send the collected information to the precision push module.

Images