Empty room management method and system for intelligent transportation management platform of long-term-rental apartment

Abstract

The invention belongs to the technical field of empty room management, and particularly relates to an empty room management method and system for an intelligent transportation management platform of a long-rental apartment, and the method comprises the steps: collecting data from a multi-source terminal, carrying out the preprocessing of the data, outputting a real empty room probability through an empty room confidence model constructed through weighted logistic regression, and dividing the types of house resources; calling a housekeeper operation prediction model to output an abuse risk value to dynamically adjust the weight for potentially used empty rooms and non-empty rooms; for potential use empty rooms, a dynamic power transmission time length prediction model constructed by the bidirectional LSTM outputs a room inspection prediction time length, and power transmission logic is optimized; and finally, the electric power and intelligent door lock system is linked to execute the instruction. According to the invention, the problem that the existing long-rental apartment is difficult to manage in the aspect of empty room management can be solved.

Description

Technical Field

[0001] This invention belongs to the field of vacant room management technology, and in particular relates to a method and system for vacant room management on a smart operation and management platform for long-term rental apartments. Background Technology

[0002] With the acceleration of urbanization, long-term rental apartments have become one of the mainstream housing forms to meet long-term living needs, but they face multiple technical pain points in operation and management. Existing long-term rental apartment management platforms can only achieve basic property status labeling and lack refined management of vacant properties throughout their entire lifecycle: on the one hand, the long-term power supply to vacant properties leads to a waste of electricity resources, and it is difficult to trace the behavior of property managers abusing vacant properties to keep guests without permission, increasing operating costs and asset management risks; on the other hand, existing technologies do not cover the deep management needs of vacant properties, including the inability to adapt the fixed power supply time for viewing to actual booking scenarios, quality degradation problems such as mold on walls and frozen pipes caused by long-term power outages in vacant properties, the inability to identify abnormal behaviors such as furniture damage and unauthorized wiring during viewings, and the lack of closed-loop management of safety hazards such as gas leaks and electrical leaks. At the same time, there is a lack of a rental pricing mechanism that matches market dynamics, resulting in low rental efficiency.

[0003] In existing technologies, IoT power outage control can only implement a single start-stop logic, without combining timing prediction and environmental data to optimize power supply strategies; access control relies on manual auditing and lacks an immutable behavior evidence storage mechanism; vacant room maintenance relies on regular manual inspections and cannot predict wear and tear through digital means; security monitoring is limited to door lock switch detection and has not formed a multi-dimensional risk prevention and control system. Summary of the Invention

[0004] The technical problem solved by this invention is to provide a smart operation and management platform for long-term rental apartments, which provides a method and system for managing vacant rooms, in order to solve the problem of difficulty in managing vacant rooms in existing long-term rental apartments.

[0005] The basic solution provided by this invention is: a method for managing vacant rooms in a smart operation and management platform for long-term rental apartments, comprising: S1: Collect multi-source data from multiple terminals in long-term rental apartments, perform preprocessing, and generate preprocessed multi-source data; S2: Construct a vacancy confidence model based on weighted logistic regression. Use preprocessed multi-source data as input to the vacancy confidence model, output the probability that a property is truly vacant, and classify properties as absolutely vacant, potentially vacant, and non-vacant based on the probability that a property is truly vacant. S3: Construct a housekeeper operation prediction model for the abuse of housing listings. For housing listings of the type of potential vacant and non-vacant, call the housekeeper operation prediction model to output the risk value of the housekeeper abusing the housing listings and dynamically adjust the housekeeper's permissions. S4: Construct a dynamic power supply duration prediction model for vacant rooms based on a bidirectional LSTM model. Call the dynamic power supply duration prediction model for vacant rooms for rooms with the property type of potentially vacant rooms. Input preprocessed multi-source data, output the predicted viewing duration, and dynamically adjust the power supply logic for vacant rooms. S5: Based on the operation instructions generated by S2, S3, and S4, it links the long-term rental apartment's power control system and smart door lock system to execute the corresponding instruction operations.

[0006] Furthermore, S1 includes: S1-1: Real-time synchronization of the smart operation and management platform with the housing management system and internal audit system to collect basic housing information, usage status information and historical behavior data; S1-2: Based on the built-in sensors of the smart door lock, the operation record data of the smart door lock is transmitted to the smart operation and management platform through the Internet of Things; S1-3: Synchronize the interface between the smart operation and management platform and the contract signing system in real time to collect information on the contract signing status and signing time of the property; S1-4: Perform data preprocessing on the basic information of the property, usage status information, historical behavior data, smart lock operation record data, property contract status information, and contract time information.

[0007] Furthermore, S2 includes: S2-1: Construct a confidence model for vacant rooms based on weighted logistic regression. Collect historical data to label real vacant rooms as training data, use the XGBoost model to train the weights of each feature, and output the trained confidence model for vacant rooms. S2-2: Extract the basic state features, historical usage features, potential demand features, housekeeper behavior features, electricity consumption features, and time-series prediction features from the preprocessed multi-source data, and input them into the vacancy confidence model to map the true vacancy probability. The expression is:

[0008] in, Indicates the confidence level of vacant rooms. This represents the sigmoid activation function, which outputs probabilities in... , This represents the weight of the j-th feature. Indicates the bias term; This represents the j-th feature of the i-th property. S2-3: Based on the confidence level of vacant rooms Classify housing types and implement differentiated power control, including: when The room is completely empty and is in a deep power-off mode, cutting off all unnecessary circuits and only keeping the door lock powered. when The room is a potentially vacant room, so a shallow power outage mode will be implemented, while basic lighting power will be maintained, and the reservation signal will be monitored. when The room is not empty, so the power outage is lifted and the housekeeper's behavior is reviewed.

[0009] Furthermore, S3 includes: S3-1: Construct a housekeeper operation prediction model based on the LightGBM classification model to match housekeepers with properties; S3-2: For properties with the type of potentially vacant or non-vacant, extract the property manager's personal characteristics, property characteristics, and property manager-property interaction characteristics from the preprocessed multi-source data corresponding to the property. Output the probability that the property manager abuses the property, expressed as:

[0010] in, Indicates the butler For housing The probability of engaging in abuse operations. , This indicates the number of decision trees in LightGBM. Indicates the first Each decision tree for feature vectors The output; S3-3: Based on Dynamically adjust the administrator's permissions, including: when This is considered high-risk; the property manager is prohibited from generating a temporary password for this property, and any adjustments require approval from a higher-level authority to open the door. when This is classified as a medium-risk property. The number of times the property manager can open the door to the property without prior appointment each day will be limited, and the door opening records will be pushed to the administrator's log. when This is considered low-risk; the administrator's normal operating privileges are retained.

[0011] Furthermore, S4 includes: S4-1: Constructing a prediction model for the dynamic power supply duration of vacant rooms based on a bidirectional LSTM model; S4-2: For properties with the property type of potential vacancy, extract the temporal features of the t-th viewing of the property from the preprocessed multi-source data, including tenant type, viewing time, weather conditions, historical viewing duration and property layout complexity. S4-3: Input the temporal features into the prediction model for the dynamic power supply duration of vacant rooms, and capture the positive temporal dependencies through a forward LSTM layer. The expression is:

[0012] in, This represents the crowd matrix of the forward LSTM. For the forward LSTM in the 1st The hidden state of the time step. Let represent the temporal characteristics of the t-th house viewing. ; The total time step of the time series feature sequence; By capturing the reverse timing dependency through a backward LSTM layer, the expression is:

[0013] in, This represents the crowd matrix of the backward LSTM. For backward LSTM in the first The hidden state of the time step. Let represent the temporal characteristics of the t-th house viewing. ; The total time step of the time series feature sequence; By concatenating the forward and backward hidden states, we obtain the merged global hidden state, expressed as:

[0014] It contains both positive and negative time-series dependency information; The fused hidden state is mapped to the predicted power delivery duration using a fully connected layer, as expressed by:

[0015] in, This represents the predicted power transmission duration. For the weights of the fully connected layer, For bias terms, The activation function is a rectified linear unit, ensuring that the prediction duration is non-negative; Predicted power transmission duration The clip function is used to restrict the input to a preset range; the expression is:

[0016] in, The shortest and most reasonable time to view a property. To ensure the longest possible viewing time, In the function, if Then output ,like Then output Otherwise output ; S4-4: When the door lock detects an opening action, it calls the dynamic power supply duration prediction model for empty rooms in real time to output the predicted power supply duration value, and supplies power according to the predicted power supply duration value, while monitoring the door lock status in real time; if there is no opening action when the predicted power supply duration value ends, the power is automatically cut off.

[0017] Furthermore, S5 includes: S5-1: The smart operation and management platform for long-term rental apartments connects the power control system and smart door lock system of the long-term rental apartments; S5-2: When the smart operation and management platform receives a power outage command, it sends a power outage signal to the power control system corresponding to the long-term rental apartment. The power control system executes the power outage operation, cutting off all indoor power circuits except for the power supply to the door lock. S5-3: After receiving the temporary power supply instruction, the intelligent operation and management platform sends a power supply signal to the power control system to restore the power supply to the entire circuit in the apartment. If no door opening action is detected after the power supply time expires, a power outage signal is automatically sent to restore the power outage state; otherwise, power is restored. S5-4: The intelligent operation and management platform adjusts the association permissions between the housekeeper and the smart door lock system based on the housekeeper's permissions.

[0018] A smart operation and management platform for long-term rental apartments provides a vacancy management system, applied to the aforementioned method for managing vacancy rooms on a smart operation and management platform for long-term rental apartments. The system includes a data acquisition module, a preprocessing module, a vacancy judgment module, a property manager behavior prediction module, a vacancy room power supply control module, and an execution module, wherein: The data acquisition module is used to collect multi-source data from multiple terminals in long-term rental apartments; the preprocessing module is used to preprocess the collected multi-source data to generate preprocessed multi-source data. The vacancy determination module constructs a vacancy confidence model based on weighted logistic regression. The vacancy determination module takes preprocessed multi-source data as input to the vacancy confidence model, outputs the probability that the property is truly vacant, and classifies the property as absolutely vacant, potentially vacant, and non-vacant based on the probability that the property is truly vacant. The housekeeper behavior prediction module constructs a housekeeper operation prediction model for the abuse of housing listings. The housekeeper behavior prediction module is used to call the housekeeper operation prediction model for housing listings of potential use of vacant and non-vacant listings, output the risk value of housekeeper abuse of housing listings, and dynamically adjust housekeeper permissions. The vacant room power supply module is based on a bidirectional LSTM model to build a dynamic power supply duration prediction model for vacant rooms. The vacant room power supply module is used to call the dynamic power supply duration prediction model for vacant rooms for rooms with the property type of potentially vacant rooms. It takes preprocessed multi-source data as input, outputs the house viewing prediction duration, and dynamically adjusts the vacant room power supply logic. The execution module receives operation instructions from the vacancy judgment module, the housekeeper behavior prediction module, and the vacancy power supply module, and links the long-term rental apartment power control system and smart door lock system to execute the corresponding instruction operations.

[0019] The principles and advantages of this invention are as follows: Based on multi-source terminal data acquisition and preprocessing, a three-level progressive intelligent decision-making system is constructed in this technical solution. First, a weighted logistic regression model is used to analyze the preprocessed data on property status, business processes, energy consumption, and other dimensions, outputting the true vacancy probability and achieving accurate property classification, providing a basis for differentiated management. Then, for potentially vacant and non-vacant properties, a housekeeper operation prediction model is used to mine the correlation between housekeeper behavior and property characteristics, quantifying the risk of abuse to dynamically adapt permissions. Simultaneously, for potentially vacant properties, a bidirectional LSTM model is used to capture the bidirectional dependency relationship of viewing time sequence characteristics, accurately predicting the viewing duration to optimize the power supply logic. Therefore, the collaborative decision-making of the three models ultimately generates operation instructions, linking the power control system and the smart door lock system to form a closed-loop management of the entire process from data input to instruction execution, ensuring that every step of the operation is supported by data and driven by the model.

[0020] The advantages are: the technical solution proposed in this application effectively solves the core pain points of difficult operation and maintenance management and the abuse of property managers in the management of vacant rooms in long-term rental apartments. I. By accurately classifying vacant rooms and implementing dynamic power outage logic, the ineffective power consumption of absolutely vacant rooms is significantly reduced, while avoiding potential power supply delays in vacant rooms. Second, by leveraging the risk prediction and dynamic adjustment of permissions of the steward, we have achieved early warning and precise control of violations, which has significantly reduced the cost of manual inspection and asset management risks. Third, the dynamic power supply duration adjustment driven by the bidirectional LSTM model not only ensures the user experience during the tenant's viewing process, but also avoids the energy waste caused by fixed power supply duration. The fully automated decision-making and system linkage significantly reduce the intensity of manual intervention, improve the efficiency and precision of vacancy management, and provide an integrated solution for long-term rental apartment operation that combines energy-saving benefits, risk control capabilities and user experience. Attached Figure Description

[0021] Figure 1 This is a flowchart of an embodiment of the present invention; Figure 2 This is a flowchart illustrating the approval process for changes in room usage according to an embodiment of the present invention. Detailed Implementation

[0022] The following detailed description illustrates the specific implementation method: The basic implementation examples are as follows: Figure 1 As shown: A method for managing vacant rooms on a smart operation and management platform for long-term rental apartments, comprising: S1: Collect multi-source data from multiple terminals in long-term rental apartments, perform preprocessing, and generate preprocessed multi-source data; wherein, S1 includes: S1-1: Real-time synchronization of the smart operation and management platform with the housing management system and internal audit system to collect basic housing information, usage status information and historical behavior data; S1-2: Based on the built-in sensors of the smart door lock, the operation record data of the smart door lock is transmitted to the smart operation and management platform through the Internet of Things; S1-3: Synchronize the interface between the smart operation and management platform and the contract signing system in real time to collect information on the contract signing status and signing time of the property; S1-4: Perform data preprocessing on the basic information of the property, usage status information, historical behavior data, smart lock operation record data, property contract status information, and contract time information.

[0023] In this embodiment, the interfaces of the intelligent operation and management platform with the housing management system and the internal audit system are synchronized in real time, once per hour. This synchronization collects basic housing attributes, current usage status, housing type change application and approval records, and historical behavior data. Basic housing attributes include housing ID, unit type, and location. Current usage status includes whether the property is rented or not. Housing type change application and approval records include... Figure 2 As shown, the data includes details of changes to the property type, the approver, and historical behavior data, including property manager operation data and records of misuse of vacant rooms.

[0024] As for the collection of door lock operation data, after the door lock's built-in sensor detects the door opening action, the operation data is immediately transmitted to the smart operation and management platform via the Internet of Things. The collected operation data includes the smart door lock's door opening action, door opening time, door opening method, door lock ID, temporary password verification status, etc.

[0025] For the collection of business system data, the interface between the smart operation and management platform and the contract signing system is synchronized in real time. Data synchronization is completed within 3 seconds after the contract status changes. The collected business data includes property contract status, contract signing time, tenant information, etc.

[0026] Finally, the collected data is preprocessed, including removing invalid data, such as data with incorrect format or duplicate synchronization, while retaining complete data information.

[0027] S2: Construct a vacancy confidence model based on weighted logistic regression. The preprocessed multi-source data is used as input to the model, and the output is the probability that a property is truly vacant. Based on this probability, properties are categorized as absolutely vacant, potentially vacant, and not vacant. S2 includes: S2-1: Construct a confidence model for vacant rooms based on weighted logistic regression. Collect historical data to label real vacant rooms as training data, use the XGBoost model to train the weights of each feature, and output the trained confidence model for vacant rooms. S2-2: Extract the basic state features, historical usage features, potential demand features, housekeeper behavior features, electricity consumption features, and time-series prediction features from the preprocessed multi-source data, and input them into the vacancy confidence model to map the true vacancy probability. The expression is:

[0028] in, Indicates the confidence level of vacant rooms. This represents the sigmoid activation function, which outputs probabilities in... , This represents the weight of the j-th feature. Indicates the bias term; This represents the j-th feature of the i-th property. S2-3: Based on the confidence level of vacant rooms Classify housing types and implement differentiated power control, including: when The room is completely empty and is in a deep power-off mode, cutting off all unnecessary circuits and only keeping the door lock powered. when The room is a potentially vacant room, so a shallow power outage mode will be implemented, while basic lighting power will be maintained, and the reservation signal will be monitored. when The room is not empty, so the power outage is lifted and the housekeeper's behavior is reviewed.

[0029] In this embodiment, a vacancy confidence model is constructed using a weighted logistic regression model. During the construction process, at least six months of historical data are collected, including status, operation, and energy consumption records of over 5000 properties. Real vacant properties are labeled as 1, while non-vacant or pseudo-vacant properties are labeled as 0. This data serves as the training data for the vacancy confidence model. The model also pre-trains feature weights, including using an XGBoost model to train and output the importance of each feature, i.e., feature weights, ensuring that core features, such as basic status features, account for ≥40% of the weight. Simultaneously, the feature weights and biases are fine-tuned daily at midnight using the previous day's new data. To avoid model drift.

[0030] Next, the eigenvectors are defined, specifically: Basic status characteristics: including the property's rented status and internal audit approval status. In the rented status, based on one-hot coding, 1 represents rented and 0 represents unrented. In the internal audit approval status, based on one-hot coding, 1 represents compliant management of the property and 0 represents unapproved or non-compliant use of the property. Historical usage characteristics: including the percentage of vacancy time in the past 7 days and the number of viewings in the past 30 days. The percentage of vacancy time in the past 7 days is expressed by the following expression: and normalized to To characterize the meaning of this feature; the number of property viewings in the past 30 days can be represented by integers. Potential demand characteristics include the regional rental popularity index and the length of time properties have been listed. The regional rental popularity index is calculated by dividing the number of daily bookings in the region by the total number of properties in the region, and then normalized. The listing duration is calculated by dividing the number of days the property has been listed by 90 and then normalizing to the nearest integer. ; Housekeeper behavior characteristics: including the number of violations by the associated housekeeper in the past 30 days and the frequency of the housekeeper's operations on this property. The number of violations by the associated housekeeper in the past 30 days is represented by integers, with 0 for no violations; the frequency of the housekeeper's operations on this property is represented by the formula: number of operations in the past 15 days / 15, and normalized. Electricity consumption characteristics: including average energy consumption during periods of vacancy, expressed by the formula: (Nearly 7 vacancy periods / 7) / rated energy consumption. Abnormally high energy consumption may indicate abuse. The time-series prediction features include the probability of viewing a house in the next 3 days, which is predicted using an existing LSTM model, and the predicted values ​​are normalized.

[0031] Based on the above characteristics, the weighted features are linearly combined into probabilities using the sigmoid function. In the calculation process, feature weights The features were trained on historical data using XGBoost. For example, the feature weight for the "rented" status was set to 0.25, which is the highest weight and reflects the core role of the "rented" status. Other features, such as the "internal review and approval" status, were trained with a weight of 0.20. The feature weight for the "vacancy period in the past 7 days" was trained with a weight of 0.15. The feature weight for the number of viewings in the past 30 days was trained with a weight of 0.10, and so on.

[0032] Finally, according to The system categorizes housing types based on numerical values ​​and implements differentiated power control, including the aforementioned classification of absolutely vacant rooms, potentially vacant rooms, and non-vacant rooms. Based on their differentiated power control logic, different housing types generate corresponding operation instructions, such as power-off instructions and power-off cancellation instructions, which are transmitted to the corresponding long-term rental apartment's power control system for power control.

[0033] S3: Construct a housekeeper operation prediction model to detect housekeeper abuse of listings. For listings classified as potentially vacant or not vacant, call the housekeeper operation prediction model to output a risk value for housekeeper abuse and dynamically adjust housekeeper permissions. S3 includes: S3-1: Construct a housekeeper operation prediction model based on the LightGBM classification model to match housekeepers with properties; S3-2: For properties with the type of potentially vacant or non-vacant, extract the property manager's personal characteristics, property characteristics, and property manager-property interaction characteristics from the preprocessed multi-source data corresponding to the property. Output the probability that the property manager abuses the property, expressed as:

[0034] in, Indicates the butler For housing The probability of engaging in abuse operations. , This indicates the number of decision trees in LightGBM. Indicates the first Each decision tree for feature vectors The output; S3-3: Based on Dynamically adjust the administrator's permissions, including: when This is considered high-risk; the property manager is prohibited from generating a temporary password for this property, and any adjustments require approval from a higher-level authority to open the door. when This is classified as a medium-risk property. The number of times the property manager can open the door to the property without prior appointment each day will be limited, and the door opening records will be pushed to the administrator's log. when This is considered low-risk; the administrator's normal operating privileges are retained.

[0035] In this embodiment, a housekeeper operation prediction model is constructed based on the LightGBM classification model to match housekeepers with properties. The model predicts the probability of a housekeeper abusing a property, based on the property confidence model's classification of potentially occupied vacant and non-vacant properties. The input features of the model include housekeeper personal features, property features, and interaction features between the housekeeper and the property. Specifically: Housekeeper's personal characteristics include: number of violations in the past 90 days, risk behavior tendency index, and vacancy rate deviation of managed properties. The number of violations in the past 90 days is expressed as an integer and includes unauthorized opening of doors, unauthorized changes to property types, etc. The risk behavior tendency index is expressed by the formula: number of unauthorized door openings in the past 30 days / total number of door openings, normalized to the nearest integer. The deviation in vacancy rate of managed properties is expressed by the formula: (Vacancy rate under the manager's jurisdiction - Average vacancy rate of the apartment) Excessive deviation may indicate that vacant rooms were intentionally reserved.

[0036] Property characteristics include rental grade and location sensitivity coefficient. Rental grade is coded using one-hot encoding, with 1 indicating high rent (more than 120% of the market average), 2 indicating medium rent, and 3 indicating low rent. Location sensitivity coefficient is also based on one-hot encoding, with 1 indicating a prime business district and 2 indicating a non-prime business district. Misuse of prime business districts yields higher returns.

[0037] The interaction characteristics between the housekeeper and the property include the frequency of the housekeeper's operations on the property and the proportion of non-appointment operations. The frequency of the housekeeper's operations on the property includes the number of operations in the past 15 days, such as opening the door and checking the status. The proportion of non-appointment operations is calculated by dividing the number of non-appointment operations in the past 15 days by the total number of operations.

[0038] Based on the above features, the decision tree ensemble outputs the probability through the LightGBM model. The number of decision trees, M, is initially set to 100, balancing accuracy and speed; finally, based on the output... Dynamically adjusting housekeeper permissions, including the aforementioned division based on different risk levels, allows for the adjustment of different management permissions. This breaks through the traditional approach of "only evaluating housekeeper risk" and combines housekeeper-property pairing characteristics. For example, a high-risk housekeeper paired with a high-rent property has a higher probability of abuse, greatly improving the accuracy of predicting housekeeper abuse.

[0039] S4: A dynamic power supply duration prediction model for vacant properties is constructed based on a bidirectional LSTM model. This model is applied to properties classified as potentially vacant, taking preprocessed multi-source data as input and outputting the predicted viewing duration. The power supply logic for vacant properties is dynamically adjusted accordingly. S4 includes: S4-1: Constructing a prediction model for the dynamic power supply duration of vacant rooms based on a bidirectional LSTM model; S4-2: For properties with the property type of potential vacancy, extract the temporal features of the t-th viewing of the property from the preprocessed multi-source data, including tenant type, viewing time, weather conditions, historical viewing duration and property layout complexity. S4-3: Input the temporal features into the prediction model for the dynamic power supply duration of vacant rooms, and capture the positive temporal dependencies through a forward LSTM layer. The expression is:

[0040] in, This represents the crowd matrix of the forward LSTM. For the forward LSTM in the 1st The hidden state of the time step. Let represent the temporal characteristics of the t-th house viewing. ; The total time step of the time series feature sequence; By capturing the reverse timing dependency through a backward LSTM layer, the expression is:

[0041] in, This represents the crowd matrix of the backward LSTM. For backward LSTM in the first The hidden state of the time step. Let represent the temporal characteristics of the t-th house viewing. ; The total time step of the time series feature sequence; By concatenating the forward and backward hidden states, we obtain the merged global hidden state, expressed as:

[0042] It contains both positive and negative time-series dependency information; The fused hidden state is mapped to the predicted power delivery duration using a fully connected layer, as expressed by:

[0043] in, This represents the predicted power transmission duration. For the weights of the fully connected layer, For bias terms, The activation function is a rectified linear unit, ensuring that the prediction duration is non-negative; Predicted power transmission duration The clip function is used to restrict the input to a preset range; the expression is:

[0044] in, The shortest and most reasonable time to view a property. To ensure the longest possible viewing time, In the function, if Then output ,like Then output Otherwise output ; S4-4: When the door lock detects an opening action, it calls the dynamic power supply duration prediction model for empty rooms in real time to output the predicted power supply duration value, and supplies power according to the predicted power supply duration value, while monitoring the door lock status in real time; if there is no opening action when the predicted power supply duration value ends, the power is automatically cut off.

[0045] In this embodiment, for properties classified as potentially vacant, targeted power supply is required. However, existing strategies using a fixed 30-minute power supply after the door is opened suffer from insufficient or wasted time. To address this, this application constructs a dynamic power supply duration prediction model for vacant properties using a bidirectional LSTM model. The model takes multi-source time-series features from before the viewing as input and outputs the estimated viewing duration. Specifically, for the input features: Tenant type uses one-hot coding: 1 for single, 2 for couple, and 3 for family. It is determined by the number of people registering when viewing the property. The reason for determining the tenant type is that families tend to have the longest viewing time, which may exceed the original fixed 30-minute power supply time, while singles generally have a shorter viewing time.

[0046] Viewing times are based on one-hot coding: 1 indicates early (8:00 AM to 12:00 PM), 2 indicates midday (12:00 PM to 6:00 PM), and 3 indicates late (6:00 PM to 10:00 PM).

[0047] Weather conditions are based on one-hot encoding, where 1 represents sunny, 2 represents rain, and 3 represents snow. Different weather conditions will affect the viewing time; for example, the viewing time may be shortened on rainy days.

[0048] Historical viewing duration includes the actual duration of the last 5 viewings, such as 15 minutes, 20 minutes, 30 minutes, etc.

[0049] In terms of the complexity of the property layout, 1 represents a one-bedroom apartment, 2 represents a two-bedroom apartment, 3 represents a three-bedroom apartment, etc. The more complex the layout, the longer the viewing time will be.

[0050] Subsequently, based on the aforementioned input features, the forward LSTM layer of the vacant room dynamic power supply duration prediction model captures the positive temporal dependencies of the input features, such as the pattern of "early period - short duration"; the backward LSTM layer captures the negative temporal dependencies of the input features, such as the pattern of "family tenants - long duration"; then, the forward and backward hidden states are concatenated, and the predicted duration is output through a fully connected layer, and the predicted duration is limited to a preset duration range, such as a minimum of 5 minutes and a maximum of 60 minutes. Power is dynamically supplied according to the output predicted duration, and the dynamic power supply generates operation instructions that are transmitted to the power control system of the corresponding long-term rental apartment, such as power supply signals and power outage signals.

[0051] S5: Based on the operation instructions generated by S2, S3, and S4, it coordinates with the long-term rental apartment's power control system and smart door lock system to execute the corresponding instruction operations; among which, S5 includes: S5-1: The smart operation and management platform for long-term rental apartments connects the power control system and smart door lock system of the long-term rental apartments; S5-2: When the smart operation and management platform receives a power outage command, it sends a power outage signal to the power control system corresponding to the long-term rental apartment. The power control system executes the power outage operation, cutting off all indoor power circuits except for the power supply to the door lock. S5-3: After receiving the temporary power supply instruction, the intelligent operation and management platform sends a power supply signal to the power control system to restore the power supply to the entire circuit in the apartment. If no door opening action is detected after the power supply time expires, a power outage signal is automatically sent to restore the power outage state; otherwise, power is restored. S5-4: The intelligent operation and management platform adjusts the association permissions between the housekeeper and the smart door lock system based on the housekeeper's permissions.

[0052] In this embodiment, based on the operation instructions in steps S2, S3, and S4, the IoT control executes as follows: Power outage execution: The smart operation and management platform receives the power outage command and sends a power outage signal to the power control system of the corresponding long-term rental apartment through the communication protocol. The power control system executes the power outage operation, cutting off all indoor power circuits except for the power supply to the door lock.

[0053] Temporary power supply execution: After receiving the temporary power supply instruction, the smart operation and management platform sends a power supply signal to the power control module to restore power supply to all indoor circuits; if no door opening is detected after the power supply time reaches the preset value, such as 30 minutes, a power outage signal is automatically sent to restore the power outage state; if door opening is detected again during the period, the 30-minute timer is reset. Release power outage execution: After receiving the power outage release command, send a permanent power supply signal and remove the property from the vacant property management list, and no longer execute the vacant property power control logic.

[0054] In addition, the intelligent operation and management platform of this application also receives abnormal early warning information and pushes it to the administrator to ensure the safety of the housing resources.

[0055] In another embodiment of this example, a smart operation and management platform for long-term rental apartments is further included, which is applied to the above-mentioned method for managing vacant rooms on a smart operation and management platform for long-term rental apartments. The system includes a data acquisition module, a preprocessing module, a vacant room judgment module, a housekeeper behavior prediction module, a vacant room power supply control module, and an execution module, wherein: The data acquisition module is used to collect multi-source data from multiple terminals in long-term rental apartments; the preprocessing module is used to preprocess the collected multi-source data to generate preprocessed multi-source data. The vacancy determination module constructs a vacancy confidence model based on weighted logistic regression. The vacancy determination module takes preprocessed multi-source data as input to the vacancy confidence model, outputs the probability that the property is truly vacant, and classifies the property as absolutely vacant, potentially vacant, and non-vacant based on the probability that the property is truly vacant. The housekeeper behavior prediction module constructs a housekeeper operation prediction model for the abuse of housing listings. The housekeeper behavior prediction module is used to call the housekeeper operation prediction model for housing listings of potential use of vacant and non-vacant listings, output the risk value of housekeeper abuse of housing listings, and dynamically adjust housekeeper permissions. The vacant room power supply module is based on a bidirectional LSTM model to build a dynamic power supply duration prediction model for vacant rooms. The vacant room power supply module is used to call the dynamic power supply duration prediction model for vacant rooms for rooms with the property type of potentially vacant rooms. It takes preprocessed multi-source data as input, outputs the house viewing prediction duration, and dynamically adjusts the vacant room power supply logic. The execution module receives operation instructions from the vacancy judgment module, the housekeeper behavior prediction module, and the vacancy power supply module, and links the long-term rental apartment power control system and smart door lock system to execute the corresponding instruction operations.

[0056] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A long-lease apartment intelligent operation and management platform empty room management method, characterized by: The method comprises the following steps: S1: Collecting multi-source data from the multi-source terminal of the long-term apartment, and preprocessing to generate preprocessed multi-source data; S2: Building an empty house confidence model based on weighted logistic regression, taking the preprocessed multi-source data as the input of the empty house confidence model, outputting the probability of the house source being a real empty house, and dividing the absolute empty house, potential use empty house and non-empty house according to the probability of the house source being a real empty house; S3: Building a housekeeper operation prediction model for house source abuse, calling the housekeeper operation prediction model for house sources of potential use empty house and non-empty house, outputting the risk value of housekeeper abuse of house source, and dynamically adjusting the housekeeper authority; S4: Building an empty house dynamic power supply time length prediction model based on a bidirectional LSTM model, calling the empty house dynamic power supply time length prediction model for house sources of potential use empty house, inputting the preprocessed multi-source data, and outputting the predicted viewing time length, and dynamically adjusting the empty house power supply logic; S5: According to the operation instructions generated by S2, S3 and S4, the long-term apartment power control system and the intelligent door lock system are linked to execute the corresponding instruction operation. 2.The long-lease apartment intelligent operation and management platform empty room management method of claim 1, wherein: The S1 comprises: S1-1: Real-time synchronization of the interface of the intelligent operation and management platform with the house source management system and the internal audit system, collecting the basic information, use state information and historical behavior data of the house source; S1-2: Based on the detection of the built-in sensor of the intelligent door lock, the intelligent door lock operation record data is transmitted to the intelligent operation and management platform through the Internet of Things; S1-3: Real-time synchronization of the interface of the intelligent operation and management platform with the signing system, collecting the house source signing state information and signing time information; S1-4: Data preprocessing of the basic information, use state information, historical behavior data, intelligent door lock operation record data, house source signing state information and signing time information of the house source. 3.The long-term rental apartment intelligent operation and management platform empty room management method of claim 1, wherein: The S2 comprises: S2-1: Building an empty house confidence model based on weighted logistic regression, collecting historical data for real empty house label annotation as training data, training the feature weights using the XGBoost model, and outputting the trained empty house confidence model; S2-2: Extracting the basic state features, historical use features, potential demand features, housekeeper behavior features, power consumption features and time sequence prediction features from the preprocessed multi-source data, inputting them into the empty house confidence model to map out the probability of real empty house, and the expression is: where, represents the empty room confidence, represents the sigmoid activation function, output probability in , represents the jth feature weight, represents the bias term; represents the jth feature of the ith room. S2-3: Based on the vacant room confidence Differentiate the execution of power control by room type, where: When , it belongs to absolute empty room, executes deep power-off mode, cuts off all unnecessary circuits, and only keeps door lock power supply; When , belonging to potential use empty room, execute shallow power-off mode, retain basic lighting power supply, while monitoring reservation signal; When , the room is not empty, the power is turned off, and the housekeeper behavior review is triggered.

4. The long-term apartment intelligent operation and management platform empty room management method of claim 3, wherein: The S3 comprises: S3-1: Building a housekeeper operation prediction model for house source based on LightGBM classification model; S3-2: For house sources of potential use empty house and non-empty house, extracting the housekeeper personal features, house source features and housekeeper and house source interaction features from the preprocessed multi-source data corresponding to the house source, outputting the probability of housekeeper abuse operation on the house source, and the expression is: wherein, denotes the housekeeper on the housing the probability of performing a misuse operation, , denotes the number of decision trees of LightGBM, denotes the th decision tree output of the feature vector ; S3-3: Based on Dynamic adjustment of housekeeping permissions, wherein: When , it belongs to high risk, prohibit the housekeeper to generate the temporary password of the house source, adjust the need to be approved by the superior to open the door; When , the risk is medium, limit the number of non-appointment opening times of the property by the housekeeper every day, and push the opening record log to the administrator; When , belongs to low risk, keep the housekeeper normal operation permission.

5. The long-term apartment intelligent operation and management platform empty room management method of claim 4, wherein: The S4 comprises: S4-1: Building an empty house dynamic power supply time length prediction model based on a bidirectional LSTM model; S4-2: For house sources of potential use empty house, extracting the time sequence features of the tth viewing of the house source from the preprocessed multi-source data, including tenant type, viewing time period, weather condition, historical viewing time length and house source house type complexity; S4-3: input the time sequence feature into the empty house dynamic power supply time length prediction model, capture the forward time sequence dependence through the forward LSTM layer, and the expression is: wherein, denotes the crowd matrix of the forward LSTM, is the hidden state of the forward LSTM at the time step, denotes the time-series feature of the t-th house visit, is the total time step of the time-series feature sequence;​ capture the reverse time sequence dependence through the backward LSTM layer, and the expression is: wherein, is a crowd matrix representing a backward LSTM, is a hidden state of the backward LSTM at the time step, is a time-series feature of the tth visit, ; is a total time step of the time-series feature sequence; concatenate the forward and backward hidden states to obtain the fused global hidden state, and the expression is: Simultaneously contains forward and reverse timing-dependent information; map the fused hidden state to the power supply time length prediction value through the fully connected layer, and the expression is: wherein, represents the power transmission duration prediction value, is a full connection layer weight, is a bias term, is a rectified linear unit activation function, ensuring that the predicted duration is non-negative; Predicting the power transmission duration The expression is limited in a preset interval by a clip function. wherein, the shortest reasonable house viewing duration, is the longest reasonable house viewing duration, In the function, if then output if then output otherwise output ; S4-4: when the door lock detects the opening action, the empty house dynamic power supply time length prediction model is called in real time to output the power supply time length prediction value, and the power supply time length prediction value is supplied according to the power supply time length prediction value, and the door lock state is monitored in real time; if the power supply time length prediction value ends without opening action, the power supply is automatically turned off.

6. The long-term apartment intelligent operation and management platform empty room management method of claim 5, wherein: The S5 comprises: S5-1: the intelligent operation management platform of the long-term apartment is connected with the power control system and the intelligent door lock system of the long-term apartment; S5-2: when the intelligent operation management platform receives the power-off instruction, the power-off signal is sent to the corresponding power control system of the long-term apartment, the power control system executes the power-off operation, and all indoor power loops except the door lock power supply are cut off; S5-3: after the intelligent operation management platform receives the temporary power supply instruction, the power supply signal is sent to the power control system, and the indoor full-loop power supply of the apartment is restored; if no opening action is detected after the power supply time length is cut off, the power-off signal is automatically sent to restore the power-off state, otherwise the power supply is restarted; S5-4: the intelligent operation management platform adjusts the association permission of the housekeeper and the intelligent door lock system according to the housekeeper permission.

7. A long-lease apartment intelligent operation and management platform empty room management system applied to the long-lease apartment intelligent operation and management platform empty room management method according to any one of claims 1-6. The system comprises a data acquisition module, a preprocessing module, an empty house judgment module, a housekeeper behavior prediction module, an empty house power supply control module, and an execution module, wherein: The data acquisition module is used to collect multi-source data from the multi-source terminal of the long-term apartment; the preprocessing module is used to generate preprocessed multi-source data after preprocessing the collected multi-source data; In the empty house judgment module, an empty house confidence model is constructed based on weighted logistic regression, the preprocessed multi-source data is input into the empty house confidence model, the probability that the house source is a real empty house is output, and the house source is divided into absolute empty house, potential use empty house and non-empty house according to the probability that the house source is a real empty house; In the housekeeper behavior prediction module, a housekeeper operation prediction model for housekeeper abuse of house source is constructed, the housekeeper behavior prediction module is used to call the housekeeper operation prediction model for house source of potential use empty house and non-empty house, output the risk value of housekeeper abuse of house source, and dynamically adjust the housekeeper permission; In the empty house power supply module, an empty house dynamic power supply time length prediction model is constructed based on a bidirectional LSTM model, the empty house power supply module is used to call the empty house dynamic power supply time length prediction model for house source of potential use empty house, input the preprocessed multi-source data, output the house viewing prediction time length, and dynamically adjust the empty house power supply logic; The execution module is used to receive the operation instructions in the empty house judgment module, the housekeeper behavior prediction module and the empty house power supply module, link the long-term apartment power control system and the intelligent door lock system, and execute the corresponding instruction operation.

Images