Residual correction-based rental house over-occupancy identification system and method

Abstract

This application proposes a system and method for identifying overcrowding in rental housing based on residual correction, relating to the field of intelligent monitoring technology. The system includes: a sensing module for generating event indication data for the rental housing; an appliance monitoring module for collecting and outputting data on changes in appliance status; an environmental monitoring module for collecting and outputting environmental disturbance data; and an integrated smart gateway for generating an actual comprehensive assessment quantity and an event disturbance compensation quantity based on the event indication data, generating an equipment disturbance compensation quantity based on the appliance status change data, generating an environmental disturbance compensation quantity based on the environmental disturbance data, generating a counterfactual predicted assessment quantity under the condition of a constant registered occupancy based on the event disturbance compensation quantity, equipment disturbance compensation quantity, and environmental disturbance compensation quantity and the baseline usage quantity, and determining the overcrowding identification result based on the difference between the actual comprehensive assessment quantity and the counterfactual predicted assessment quantity. The solution proposed in this application can accurately and effectively identify overcrowding in rental housing.

Description

Technical Field

[0001] This application relates to the field of intelligent monitoring technology, and in particular to a rental housing overcrowding identification system and method based on residual correction. Background Technology

[0002] With the expansion of the urban floating population, the safety management of rental housing has become a core aspect of urban grassroots governance. Overcrowding (group rentals) is a significant contributing factor to fire hazards, security incidents, and public health risks. Currently, most solutions for identifying overcrowding in rental housing employ a logic of "registered occupants + water, electricity, and gas usage thresholds + time period comparison." This means that threshold ranges for water, electricity, and gas usage are pre-set based on factors such as the number of registered residents, apartment type, and season. When actual usage exceeds these thresholds, it is determined that overcrowding has occurred.

[0003] However, abnormal fluctuations in the consumption of resources such as water, electricity, and gas are not necessarily caused by an increase in the actual number of residents. In real-world scenarios, sudden temperature rises and falls within a short period can lead to a significant increase in energy consumption of temperature-controlled devices such as air conditioners, water heaters, and electric heaters; the addition, relocation, restoration of power after a power outage, malfunction, or change in the position of appliances can cause abrupt changes in the power consumption curve; leaking water pipes can create continuous water usage pulses, resulting in abnormally high water consumption; and short-term entry by visitors, maintenance personnel, cleaning staff, property managers, or temporarily authorized personnel can also generate lock-opening records and short-term water, electricity, and gas usage. These abnormal consumption patterns not caused by occupancy factors are all misjudged as an increase in the number of residents by existing threshold-based comparison schemes, leading to a high risk of false alarms in overcrowding identification and failing to meet the actual needs of refined management of rental housing. Summary of the Invention

[0004] In view of this, this application proposes a rental housing overcrowding identification system and method based on residual correction.

[0005] In the first aspect, this application provides a rental housing overcrowding identification system based on residual correction, including: a sensing module, an appliance monitoring module, an environmental monitoring module, and an integrated smart gateway; The sensing module is used to generate event indication data for the rental house. The sensing module includes an electronic lock for the entrance, a smart water meter, a smart electricity meter, a smart gas meter, and a water leakage sensor. The home appliance monitoring module is used to collect and output home appliance status change data, which includes home appliance power consumption data, posture recording data, and home appliance type labels. The environmental monitoring module is used to collect and output environmental disturbance data, which includes temperature changes over a predetermined time, humidity information, and weather conditions. The integrated smart gateway is used to generate actual comprehensive assessment quantity and event disturbance compensation quantity according to event indication data, generate equipment disturbance compensation quantity according to home appliance status change data, generate environmental disturbance compensation quantity according to environmental disturbance data, generate counterfactual prediction assessment quantity under the condition of unchanged number of registered residents by combining event disturbance compensation quantity, equipment disturbance compensation quantity, environmental disturbance compensation quantity and basic usage baseline, determine the overcrowding identification result based on the difference between actual comprehensive assessment quantity and counterfactual prediction assessment quantity, and send the overcrowding identification result to cloud server.

[0006] In one embodiment, the integrated smart gateway includes: a DMA controller, a dual-port RAM cache, an FPGA dedicated event pipeline, a Flash dedicated parameter storage area, an NPU inference module, and an illegal occupancy identification module; The DMA controller is used to directly write event indication data, home appliance status change data, and environmental disturbance data into a dual-port RAM cache. The FPGA-dedicated event pipeline is used to determine, based on event indication data, home appliance status change data and environmental disturbance data, when the event node meets the preset time window, identity type and status label conditions, generate causal edge labels between event nodes, encapsulate the event node and the corresponding causal edge labels into an event attribution input frame, and output a feature-ready hardware signal to the NPU inference module when the event attribution input frame meets the preset integrity conditions. The NPU inference module is used to respond to the feature-ready hardware signal, read the event attribution input frame and the parameters in the Flash dedicated parameter storage area, generate the actual comprehensive evaluation quantity, and generate the event disturbance compensation quantity, environmental disturbance compensation quantity and equipment disturbance compensation quantity according to the causal edge marker in the event attribution input frame. Based on the event disturbance compensation quantity, environmental disturbance compensation quantity, equipment disturbance compensation quantity and basic usage baseline, it generates the counterfactual prediction evaluation quantity under the condition of unchanged registered residents. Based on the difference between the actual comprehensive evaluation quantity and the counterfactual prediction evaluation quantity, it generates the residential behavior residual feature frame. Based on the residential behavior residual feature frame, it outputs the estimated value of the newly added residents. The event disturbance compensation quantity includes the leakage disturbance compensation quantity and the non-resident identity disturbance compensation quantity. The illegal occupancy identification module is used to generate an estimated number of actual residents based on the number of registered residents and the estimated number of newly added residents. When the estimated number of actual residents exceeds the rated number of residents in the rental house for a consecutive preset number of statistical periods, it is determined that the rental house has illegal overcrowding.

[0007] In one embodiment, the NPU inference module is further configured to extract normalized actual water consumption, normalized actual electricity consumption, and normalized actual gas consumption data calculated based on historical normal residential baselines within the current statistical period from the event attribution input frame; then read the configuration parameters related to water consumption, electricity consumption, and gas consumption weights pre-stored in the Flash dedicated parameter storage area; match the corresponding weight value rules based on the normalized actual water consumption, normalized actual electricity consumption, normalized actual gas consumption data, configuration parameters, and the operating condition attributes of the rental housing; determine the water consumption weight value, electricity consumption weight value, and gas consumption weight value; and complete the calculation based on each weight value according to the preset weighted calculation formula to generate the actual comprehensive evaluation quantity.

[0008] In one embodiment, the NPU inference module is further configured to extract monitoring data corresponding to environmental disturbances, appliance status, water leakage events, and non-resident entry events within the current statistical period based on the causal edge markers in the event attribution input frame; read the pre-stored environmental compensation coefficients, equipment disturbance compensation coefficients, water leakage disturbance compensation parameters, and non-resident entry disturbance compensation parameters in the Flash dedicated parameter storage area; match the corresponding calculation rules and coefficient value ranges based on the disturbance types corresponding to various monitoring data, configuration parameters, and causal edge markers; substitute the relevant known data into the preset calculation formulas corresponding to the compensation amounts of each type of disturbance to complete the calculation, thereby obtaining the environmental disturbance compensation amount, the equipment disturbance compensation amount, the water leakage disturbance compensation amount, and the non-resident entry disturbance compensation amount.

[0009] In one embodiment, the NPU inference module is further configured to read the basic usage baseline related configuration parameters pre-stored in the Flash dedicated parameter storage area, obtain the basic usage baseline value determined by the month, time segment, apartment type parameters and the number of registered residents, retrieve the calculated event disturbance compensation amount, environmental disturbance compensation amount and equipment disturbance compensation amount, substitute the basic usage baseline and various disturbance compensation amount data into the calculation formula corresponding to the counterfactual prediction assessment amount to complete the calculation, and generate the counterfactual prediction assessment amount under the condition that the number of registered residents remains unchanged.

[0010] In one embodiment, the FPGA-dedicated event pipeline is further used to standardize the event indication data to generate event nodes, retrieve preset time window parameters, identity type matching rules, and status flag configuration parameters, and verify each event node data and corresponding configuration parameter to see if they meet the preset time window, identity type, and status flag judgment conditions. After matching the corresponding association rules, causal edge tags between event nodes are generated. Then, the event nodes that have been judged and their corresponding causal edge tags are retrieved and integrated and packaged according to the preset frame structure and encapsulation rules. The event nodes and their corresponding causal edge tags are uniformly encapsulated into the event attribution input frame.

[0011] In one embodiment, the FPGA-dedicated event pipeline generates causal edge tags according to the following rules: If the entry event node corresponding to the permanent tenant's identity is time-related to the water, electricity and gas event nodes within the first preset time window, then the entry-caused water mark, entry-caused electricity mark and entry-caused gas mark will be generated sequentially. If the environmental disturbance event node corresponding to the temperature change at a preset time is temporally correlated with the power consumption change event node within the second preset time window, an environmental energy consumption change marker is generated. If an event node indicating a change in the status of a household appliance is time-related to an event node indicating a sudden change in power consumption within the third preset time window, a marker indicating a change in energy consumption caused by the device is generated. If the leakage status signal coincides with the time of the continuous water use event node, or if the continuous water use event node is not time-correlated with the permanent tenant's entry event node within the fifth preset time window, a leakage causing continuous water use marker will be generated. If the entry event node corresponding to a non-resident identity forms a time correlation with the water, electricity and gas event nodes within the fourth preset time window, a short-term usage marker caused by non-resident identity is generated. If the event node corresponding to the temporary authorized identity appears repeatedly in multiple consecutive statistical periods and is temporally related to the residential behavior residual, a temporary authorization long-term marker is generated. The residential behavior residual is the difference between the actual comprehensive assessment quantity and the counterfactual predicted assessment quantity. If multiple independent residential behavior residual fragments appear in the event node corresponding to the same identity identifier, a suspected shared identity tag is generated.

[0012] In one embodiment, the NPU inference module is further configured to determine an identity-behavior consistency score based on the event attribution input frame, the residential behavior residual feature frame, and historical event statistics, identify two abnormal scenarios—temporary authorization becoming permanent and suspected multiple-person shared identity—based on the identity-behavior consistency score, and adjust the non-residential identity disturbance compensation amount based on the identity-behavior consistency score.

[0013] In one embodiment, the NPU inference module is further configured to extract, from the event attribution input frame, the residential behavior residual feature frame, and historical event statistics, the cross-day recurrence frequency, the number of nighttime residual occurrences, the number of morning and evening peak residual occurrences, the stability of the first usage residual delay after unlocking, the visitor-type marker, and the non-residential identity marker corresponding to the target identity, respectively. The module then performs normalization processing on the first four indicators according to preset rules, reads the identity-behavior scoring weight coefficients pre-stored in the Flash dedicated parameter storage area, matches the corresponding calculation rules based on various feature data, identity marker data, and weight coefficients, substitutes the relevant known data into the preset calculation formula corresponding to the identity-behavior consistency score to complete the calculation, and outputs the identity-behavior consistency score corresponding to the target identity.

[0014] Secondly, this application also provides a method for identifying overcrowded rental housing based on residual correction, wherein the method is executed based on the rental housing overcrowding identification system based on residual correction as described in the first aspect; the method includes: Acquire event indication data, appliance status change data, and environmental disturbance data; Based on the event indication data, generate the actual comprehensive assessment quantity and event disturbance compensation quantity; based on the appliance status change data, generate the equipment disturbance compensation quantity; based on the environmental disturbance data, generate the environmental disturbance compensation quantity. The counterfactual prediction assessment quantity is generated by combining the event disturbance compensation quantity, equipment disturbance compensation quantity, and environmental disturbance compensation quantity with the basic usage baseline, under the condition that the number of registered residents remains unchanged. The overcrowding identification result is determined based on the difference between the actual comprehensive assessment quantity and the counterfactual predicted assessment quantity, and the overcrowding identification result is sent to the cloud server.

[0015] The rental housing overcrowding identification system based on residual correction proposed in this application has the following advantages over related technologies: 1. The rental housing overcrowding identification system based on residual correction proposed in this application, through the division of labor and cooperation of the sensing module, home appliance monitoring module and environmental monitoring module, can comprehensively collect various types of data such as rental housing entry behavior, water, electricity and gas usage, water leakage, equipment status and environmental changes, etc., and fully cover all factors that may cause fluctuations in energy consumption. The data collection is comprehensive and highly targeted, thereby ensuring that the analysis results are accurate and effective.

[0016] 2. The integrated smart gateway generates the actual comprehensive assessment quantity based on the event indication data, calculates various disturbance compensation quantities according to type, and generates the counterfactual prediction assessment quantity in combination with the basic usage baseline. Based on the difference between the actual comprehensive assessment quantity and the counterfactual prediction assessment quantity, the overcrowding situation can be identified, which can effectively distinguish the usage anomalies caused by non-human factors and reduce the probability of misjudgment. At the same time, the gateway uploads the final results to the cloud server to realize local intelligent analysis and centralized cloud management, which facilitates the unified supervision of rental housing by management personnel. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the structure of a rental housing overcrowding identification system based on residual correction in one embodiment of this application; Figure 2 This is a schematic diagram of the internal structure of an FPGA-dedicated event pipeline in one embodiment of this application; Figure 3 This is a schematic diagram of causal attribution of event nodes and causal edge markers in one embodiment of this application; Figure 4 This is a schematic diagram of the structure of an event attribution input frame in one embodiment of this application; Figure 5 This is a flowchart illustrating a method for identifying overcrowding in rental housing based on residual correction in one embodiment of this application.

[0019] Explanation of reference numerals in the attached figures: 1-Sensing module, 11-Entrance electronic lock, 12-Smart water meter, 13-Smart electricity meter, 14-Smart gas meter, 15-Leakage sensor, 2-Home appliance monitoring module, 3-Environmental monitoring module, 4-Integrated smart gateway, 41-DMA controller, 42-Dual-port RAM cache, 43-FPGA dedicated event pipeline, 431-Event node input register group, 432-Event node encapsulation circuit, 433-Timestamp difference calculator array, 434-Time window comparator array, 435-Identity type matcher, 436-State flag register group, 437-Causal edge flag state machine, 438-Frame encapsulation circuit, 439-Frame verification circuit, 44-Flash dedicated parameter storage area, 45-NPU inference module, 46-Illegal occupancy identification module. Detailed Implementation

[0020] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0021] In some embodiments, such as Figure 1 As shown, this application provides a rental housing overcrowding identification system based on residual correction, including: a sensing module 1, an appliance monitoring module 2, an environmental monitoring module 3, and an integrated smart gateway 4.

[0022] Sensing module 1 is used to generate event indication data for the rental property. Sensing module 1 includes an electronic lock 11, a smart water meter 12, a smart electricity meter 13, a smart gas meter 14, and a leak sensor 15. The electronic lock 11 collects and outputs entry event data carrying an identity identifier, identity type marker, unlocking timestamp, and unlocking method. The identity type marker includes at least the identity of a permanent tenant and a non-resident. The smart water meter 12 collects and outputs water event data carrying a sampling timestamp, including time-of-use water pulses, cumulative water consumption, and continuous water consumption duration. The smart electricity meter 13 collects and outputs electricity event data carrying a sampling timestamp, including time-of-use electricity consumption, instantaneous power, power surge events, and continuous high-power periods. The smart gas meter 14 collects and outputs gas event data carrying a sampling timestamp, including time-of-use gas consumption, gas consumption period, and continuous gas consumption duration. The leak sensor 15 collects and outputs leak event data carrying a trigger timestamp, including leak status, leak duration, and leak location.

[0023] It should be noted that the electronic lock 11 is installed on the entrance door of the rental property and supports multiple unlocking methods such as fingerprint, password, NFC, and facial recognition. Each unlocking event generates corresponding entry event data, with the data format being: event type + identity identifier + identity type tag + unlocking timestamp + unlocking method. In this embodiment, the identity type tag is divided into 6 categories: 01-permanent tenant, 02-temporary authorized, 03-visitor, 04-cleaning, 05-repair, and 06-housekeeper. Different identity types correspond to different permissions and attribution rules.

[0024] Smart water meter 12, smart electricity meter 13, and smart gas meter 14 are installed in the water inlet, electricity inlet, and gas inlet pipes / lines of the rental house, respectively. The sampling frequency is configurable; in this embodiment, the sampling frequency is 1 time / minute. This value is only for illustrative purposes and can be flexibly adjusted according to the actual monitoring needs of the rental house, and is not limited to this. The three types of smart meters output time-sharing data on water, electricity, and gas usage events, respectively. The data carries accurate sampling timestamps, time-sharing usage, duration, power / pulse characteristics, and other information. Leakage sensor 15 is installed in areas prone to leakage, such as the kitchen, bathroom, and balcony. It uses a water immersion electrode detection method. When a leak is detected, it immediately outputs leakage event data, carrying trigger timestamps, leakage location, and leakage duration, among other information.

[0025] The appliance monitoring module 2 is used to collect and output appliance status change data, which includes appliance power consumption data, posture recording data, and appliance type tags. Specifically, appliance power consumption data can include appliance power outage markers, vibration operation markers, and power surge markers. The appliance monitoring module 2 can be deployed in the main appliances (air conditioner, water heater, refrigerator, washing machine, etc.) in a rental house, and can be implemented through smart sockets, built-in Bluetooth modules, etc. It can collect appliance status data such as power outage status, position changes, posture changes, operating vibration, and power surges in real time, and output appliance status change event data.

[0026] The environmental monitoring module 3 is used to collect and output environmental disturbance data, including temperature changes, humidity information, and weather conditions over a predetermined time period. For example, the environmental monitoring module 3 may include an indoor temperature and humidity sensor and an outdoor weather data interface module, capable of collecting real-time data on indoor and outdoor temperature, humidity, weather, and rainfall. It primarily outputs environmental disturbance data such as temperature changes and indoor / outdoor temperature differences within one hour. It should be noted that this time interval is merely an example and can be adjusted according to the actual scenario; it is not limited to this.

[0027] The integrated smart gateway 4 is used to generate actual comprehensive assessment quantities and event disturbance compensation quantities based on event indication data, generate equipment disturbance compensation quantities based on appliance status change data, generate environmental disturbance compensation quantities based on environmental disturbance data, generate counterfactual prediction assessment quantities under the condition of unchanged number of registered residents based on event disturbance compensation quantities, equipment disturbance compensation quantities, environmental disturbance compensation quantities and basic usage baseline, determine the overcrowding identification result based on the difference between actual comprehensive assessment quantities and counterfactual prediction assessment quantities, and send the overcrowding identification result to the cloud server.

[0028] It should be noted that the cloud management platform is deployed at the city's rental housing management service center. It is used to receive the overcrowding identification results uploaded by the integrated smart gateway 4 of each rental house. The overcrowding identification results can include overcrowding warning information, estimated actual number of residents and attribution explanation results. It displays the visualized identification data to the management personnel. At the same time, it supports sending basic configuration parameters, updating compensation coefficients and model parameters to the gateway. The cloud platform does not participate in the real-time judgment process of illegal overcrowding, but is only responsible for management and display.

[0029] Understandably, the integrated smart gateway 4 first uses the received event indication data to calculate and simultaneously generate an actual comprehensive assessment quantity reflecting the real usage level on site, as well as the corresponding event disturbance compensation quantity. Then, it performs special calculations on the data of changes in the status of home appliances and environmental disturbance data to obtain the equipment disturbance compensation quantity and the environmental disturbance compensation quantity in turn. Subsequently, it combines the three types of disturbance compensation quantities with the preset basic usage baseline and calculates the corresponding counterfactual prediction assessment quantity under the premise that the number of registered residents remains unchanged. By comparing the numerical difference between the actual comprehensive assessment quantity and the counterfactual prediction assessment quantity, it determines whether the rental house is overcrowded. After completing the overcrowding status identification, the resulting overcrowding identification results are finally pushed to the cloud server.

[0030] The aforementioned rental housing overcrowding identification system based on residual correction, through the collaborative efforts of sensor module 1, appliance monitoring module 2, and environmental monitoring module 3, comprehensively collects various data such as rental housing entry behavior, water, electricity, and gas usage, leaks, equipment status, and environmental changes. It fully covers all factors that can cause fluctuations in energy consumption, ensuring comprehensive and targeted data collection and accurate and effective analysis results. The integrated smart gateway 4 generates actual comprehensive assessment quantities based on event indication data and calculates various disturbance compensation quantities according to type. Combined with the baseline usage, it generates counterfactual predictive assessment quantities. Based on the difference between the actual comprehensive assessment quantity and the counterfactual predictive assessment quantity, it identifies overcrowding situations, effectively distinguishing usage anomalies caused by factors other than human numbers, reducing the probability of misjudgment. Simultaneously, the gateway uploads the final results to a cloud server, achieving local intelligent analysis and centralized cloud management, facilitating unified supervision of rental housing by management personnel.

[0031] In some embodiments, such as Figure 1As shown, the integrated smart gateway 4 includes: a DMA (Direct Memory Access) controller 41, a dual-port RAM (Random Access Memory) cache 42, a dedicated FPGA (Field Programmable Gate Array) event pipeline 43, a dedicated Flash parameter storage area 44, an NPU (Neural Processing Unit) inference module 45, and an illegal occupancy identification module 46.

[0032] The DMA controller 41 is used to directly write event indication data, appliance status change data, and environmental disturbance data into the dual-port RAM buffer 42. The DMA controller 41 interfaces with various sensing terminals via RS485, LoRa, WiFi, Bluetooth, etc., directly writing the collected multi-source event data into the dual-port RAM buffer 42 without going through the main processor, reducing the transmission delay caused by the main processor's scheduling and ensuring the real-time performance of data acquisition.

[0033] The FPGA-dedicated event pipeline 43 is used to determine, based on event indication data, home appliance status change data and environmental disturbance data, when the event node meets the preset time window, identity type and status label conditions, generate causal edge labels between event nodes, encapsulate the event node and the corresponding causal edge labels into an event attribution input frame, and output a feature-ready hardware signal to the NPU inference module 45 when the event attribution input frame meets the preset integrity conditions.

[0034] For example, the internal structure of the FPGA dedicated event pipeline 43 is as follows: Figure 2 As shown, the pipeline includes an event node input register group 431, an event node encapsulation circuit 432, a timestamp difference calculator array 433, a time window comparator array 434, an identity type matcher 435, a status flag register group 436, a causal edge marking state machine 437, a frame encapsulation circuit 438, and a frame verification circuit 439. In this embodiment, the FPGA-dedicated event pipeline 43 adopts a fully hardware pipeline design, with each processing stage corresponding to an independent hardware circuit. It requires no software instruction scheduling and can perform pipelined processing according to a fixed hardware clock cycle.

[0035] Specifically, the event node input register group 431 consists of multiple sets of parallel registers, each set of registers corresponding to the event data of a type of sensing terminal, used to cache the multi-source event data output by the dual-port RAM buffer 42, providing stable input for subsequent encapsulation processing.

[0036] The event node encapsulation circuit 432 encapsulates multi-source event data into event nodes according to a preset unified format. The unified format of the event node may include event type, event timestamp, event source, identity type, usage value, duration, status flag, and location information. Event types may include entry event nodes, water usage event nodes, electricity usage event nodes, gas usage event nodes, water leakage event nodes, appliance status event nodes, environmental disturbance event nodes, and non-residential identity event nodes. In this embodiment, the unified format of the event node is: 8-bit event type + 32-bit event timestamp + 8-bit event source + 8-bit identity type + 32-bit usage value + 16-bit duration + 8-bit status flag + 8-bit location information. The event types are divided into 8 categories: 01-Entry event node, 02-Water usage event node, 03-Electricity usage event node, 04-Gas usage event node, 05-Water leakage event node, 06-Appliance status event node, 07-Environmental disturbance event node, and 08-Non-resident identity event node. The above coding format and classification are only illustrative examples and can be expanded and adjusted according to actual needs, and are not limited thereto.

[0037] The timestamp difference calculator array 433 consists of multiple parallel hardware subtractors, which can simultaneously calculate the difference between timestamps of multiple event nodes and output the time interval between events, providing data support for subsequent time window comparison.

[0038] The time window comparator array 434 consists of multiple parallel hardware comparators and presets multiple sets of associated time windows.

[0039] The identity type matcher 435 is implemented through a hardware comparison circuit. It reads the identity type marker in the event node, compares it with the preset identity type code, distinguishes between permanent tenants and non-residents, and outputs the identity type matching result.

[0040] The status flag register group 436 includes a water leakage flag register, a device status flag register, an environmental disturbance flag register, and a non-resident identity flag register, which are used to latch the status flags of the corresponding events. When the corresponding event is triggered, the register is set and cleared after the event ends, providing a status basis for causal edge marking.

[0041] The causal edge labeling state machine 437 is the core of the FPGA-dedicated event pipeline 43, used to generate causal edge labels between event nodes to clarify the causes of abnormal usage. In this embodiment, the causal edge label is a 32-bit code, with each bit corresponding to a causal relationship, specifically including: bit0 - water usage caused by entry into the home, bit1 - electricity usage caused by entry into the home, bit2 - gas usage caused by entry into the home, bit3 - energy consumption change caused by environmental factors, bit4 - energy consumption change caused by equipment, bit5 - continuous water usage caused by water leakage, bit6 - short-term usage caused by non-resident identity, bit7 - temporary authorization becoming long-term, bit8 - suspected shared identity by multiple people, and the remaining bits are reserved for expansion. The above encoding method is only an example and can be adjusted according to actual needs, and is not limited to this.

[0042] like Figure 3 As shown, Figure 3 This is a causal attribution diagram for event nodes and causal edge markers. The FPGA-dedicated event pipeline 43 generates causal edge markers according to the following rules: If the entry event node corresponding to the permanent tenant's identity forms a time association with the water, electricity, and gas event nodes within the first preset time window, entry-induced water usage markers, entry-induced electricity usage markers, and entry-induced gas usage markers are generated sequentially; if the environmental disturbance event node corresponding to the temperature change within the preset time window forms a time association with the power consumption change event node within the second preset time window, an environmental-induced energy consumption change marker is generated; if the appliance status change event node forms a time association with the power consumption change event node within the third preset time window, an appliance-induced energy consumption change marker is generated; if the water leakage status signal coincides with the continuous water usage event node, or if the continuous water usage event node does not coincide with the permanent tenant's identity within the fifth preset time window... When a guest's entry event node forms a temporal correlation, a water leakage-induced continuous water usage marker is generated. If an entry event node corresponding to a non-resident identity forms a temporal correlation with water, electricity, and gas event nodes within a fourth preset time window, a non-resident identity-induced short-term usage marker is generated. If an event node corresponding to a temporary authorized identity repeatedly appears in multiple consecutive statistical periods and has a temporal correlation with the residential behavior residual, a temporary authorization long-term marker is generated. The residential behavior residual is the difference between the actual comprehensive assessment quantity and the counterfactual predicted assessment quantity (used to characterize the remaining usage change that cannot be explained by environmental disturbances, equipment disturbances, water leakage disturbances, and short-term disturbances of non-resident identities). If multiple independent residential behavior residual fragments appear in the event nodes corresponding to the same identity identifier, a suspected shared identity marker is generated.

[0043] In this embodiment, the first preset time window is 1 to 120 minutes after a resident tenant enters the premises, preferably 5 to 30 minutes; the second preset time window is 5 to 360 minutes after an environmental disturbance event, preferably 15 to 180 minutes; the third preset time window is 1 to 180 minutes after an appliance status change event, preferably 5 to 60 minutes; the fourth preset time window is 1 to 240 minutes after a non-resident enters the premises, preferably 15 to 120 minutes; and the fifth preset time window is 10 to 360 minutes after a water leakage status signal is triggered, preferably 30 to 180 minutes. These time windows can be configured according to the size of the rental unit, the equipment layout, and management needs, and are not limited to the above preferred values.

[0044] Frame encapsulation circuit 438 encapsulates the event node and its corresponding causal edge marker into an event attribution input frame. In this embodiment, the structure of the event attribution input frame is as follows: Figure 4 As shown, the frame structure includes: a 32-bit frame header field, a 16-bit event node count field, an event node set field, a 32-bit causal edge marker field, a 16-bit perturbation type field, and a 32-bit frame checksum field. The frame header field uses a fixed code of 0xAA55AA55 for frame synchronization; the event node set field contains all encapsulated event nodes; and the frame checksum field uses a CRC32 checksum to verify the integrity of the frame data. This frame structure is only an example and can be expanded according to actual needs; it is not limited to this.

[0045] The frame verification circuit 439 reads the frame verification field of the event attribution input frame, performs CRC32 verification on the frame data, and checks whether key fields such as the number of event nodes, timestamp, identity type, and causal edge marker are complete. When the verification passes and all key fields are written, the event attribution input frame is determined to meet the preset integrity conditions. When the event attribution input frame meets the integrity conditions, a high-level pulse feature-ready hardware signal is output to the NPU inference module 45 to trigger the NPU to start the inference process, achieving hardware-level process triggering without CPU interrupt scheduling.

[0046] The dedicated Flash parameter storage area 44 uses non-volatile Flash memory to store baseline parameters, environmental compensation coefficients, equipment disturbance compensation coefficients, water leakage disturbance compensation parameters, non-resident identity disturbance compensation parameters, rated resident population, registered resident population, statistical period parameters, etc. The dedicated Flash parameter storage area 44 is fixedly mapped to the data read channel of the NPU inference module 45 through a hardware address decoding circuit. The NPU can directly read parameters through the hardware address without the need for operating system drivers and file system calls, which helps reduce the impact of parameter read latency. The frame output end of the FPGA dedicated event pipeline 43 is directly connected to the feature input end of the NPU inference module 45 through an AXI-Stream point-to-point dedicated physical channel. Event attribution input frames are directly transmitted to the NPU's input buffer through this channel, without going through CPU reading, CPU task scheduling, shared memory buffer, or operating system driver layer forwarding during transmission, which helps reduce data transmission scheduling latency and improve processing timing determinism.

[0047] The NPU inference module 45, in response to a feature-ready hardware signal, reads the parameters from the event attribution input frame and the dedicated Flash parameter storage area 44, generates the actual comprehensive evaluation quantity, and generates event disturbance compensation quantity, environmental disturbance compensation quantity, and equipment disturbance compensation quantity according to the causal edge markers in the event attribution input frame. Based on the event disturbance compensation quantity, environmental disturbance compensation quantity, equipment disturbance compensation quantity, and the baseline usage, it generates a counterfactual prediction evaluation quantity under the condition of unchanged registered resident population. Based on the difference between the actual comprehensive evaluation quantity and the counterfactual prediction evaluation quantity, it generates a resident behavior residual feature frame, and outputs an estimated value of the newly added resident population based on the resident behavior residual feature frame. The event disturbance compensation quantity includes leakage disturbance compensation quantity and non-resident identity disturbance compensation quantity. The NPU inference module 45 can employ an embedded neural network processor.

[0048] It is understandable that after receiving the feature-ready hardware signal, the NPU inference module 45 starts data processing and calculation. First, it reads the various raw monitoring data carried by the event attribution input frame and the pre-configured calculation parameters in the Flash dedicated parameter storage area 44. It completes the calculation through the established algorithm and generates the actual comprehensive evaluation quantity. Then, it combines the causal edge markers in the event attribution input frame to distinguish and calculate different types of disturbances, and obtains the event disturbance compensation quantity, environmental disturbance compensation quantity, and equipment disturbance compensation quantity in sequence. The event disturbance compensation quantity is composed of two parts: water leakage disturbance compensation quantity and non-resident identity disturbance compensation quantity. On this basis, the three types of disturbance compensation quantities are combined with the preset basic usage baseline. The counterfactual prediction evaluation quantity is calculated by strictly following the constraint that the number of registered residents remains unchanged. Then, the difference between the actual comprehensive evaluation quantity and the counterfactual prediction evaluation quantity is calculated. Based on the calculation result, a residential behavior residual feature frame that can reflect the energy consumption characteristics of additional personnel is generated. Finally, the system completes the analysis and judgment based on the residential behavior residual feature frame and outputs the corresponding estimated value of the number of newly added residents.

[0049] The illegal occupancy identification module 46 is used to generate an estimated actual occupancy number based on the registered occupancy number and the estimated number of newly added occupants. If the estimated actual occupancy number exceeds the rental property's rated occupancy number for a consecutive preset number of statistical periods, it determines that the rental property has engaged in illegal overcrowding. The illegal occupancy identification module 46 can send relevant overcrowding identification information via a communication module that can use Ethernet, 4G / 5G, WiFi, or other communication methods to achieve data interaction between the integrated smart gateway 4 and the cloud management platform. In this embodiment, the integrated smart gateway 4 supports local closed-loop operation even when the external network is interrupted, continuously executing all event processing, inference calculations, and overcrowding determination processes, and caching warning information and identification results. After the external network is restored, it only uploads overcrowding warning information, the estimated actual occupancy number, and attribution explanation tags to the cloud management platform, without transmitting the original collected data, which helps protect data privacy and security.

[0050] In the application, the illegal occupancy identification module 46 reads the estimated number of new occupants ΔP output by the NPU and the rated number of occupants Prated for the rental property, and calculates the estimated number of actual occupants Pactual according to the following formula: Pactual = Pregistered + ΔP Wherein, Pactual is the estimated number of actual residents, Precaged is the registered number of residents in the rental property, and ΔP is the estimated number of new residents output by the NPU inference module 45; the value range of Precaged is 1 to 20, the value range of ΔP is 0 to 20, and the value range of Pactual is 1 to 40; when Pactual > Prated, and this relationship is established for a consecutive preset number of statistical periods, the illegal occupancy identification module 46 determines that there is illegal overcrowding behavior, where Prated is the rated number of residents in the rental property, and the value range of Prated is 1 to 20.

[0051] In this embodiment, the rated number of occupants can be determined based on local rental housing management regulations, housing area, unit type parameters, or management platform configuration parameters; the value range of the continuous statistical period N is 2 to 30, and can be selected as 3 to 7; the duration of each statistical period is 15 minutes to 24 hours, preferably 1 hour, 6 hours, 12 hours, or 24 hours. For example, when Pactual > Prated and this relationship holds true for three consecutive 24-hour statistical periods, it is determined that the rental housing has an illegal overcrowding situation, and an overcrowding warning message and corresponding attribution explanation results are generated.

[0052] In some embodiments, the NPU inference module 45 is further configured to extract normalized actual water consumption, normalized actual electricity consumption, and normalized actual gas consumption data calculated based on historical normal residential baselines within the current statistical period from the event attribution input frame; then read the configuration parameters related to water consumption, electricity consumption, and gas consumption weights pre-stored in the Flash dedicated parameter storage area 44; match the corresponding weight value rules based on the normalized actual water consumption, normalized actual electricity consumption, normalized actual gas consumption data, configuration parameters, and the operating condition attributes of the rental housing; determine the water consumption weight value, electricity consumption weight value, and gas consumption weight value; and complete the calculation based on each weight value according to the preset weighted calculation formula to generate the actual comprehensive evaluation quantity.

[0053] To ensure the comparability of parameters across different rental unit types, areas, registered occupants, and equipment configurations, this embodiment and subsequent embodiments normalize water, electricity, gas, environment, equipment status, and identity / behavioral characteristics. Unless otherwise specified, the normalized variable values ​​range from 0 to 1; when using a percentage-based normalization, the variable values ​​range from 0 to 100.

[0054] In application, the actual comprehensive consumption Ut can be obtained by weighting the normalized actual water consumption Wwater,t, the normalized actual electricity consumption Wpower,t, and the normalized actual gas consumption Wgas,t, as shown in the formula: Ut=μ1×Wwater,t+μ2×Wpower,t+μ3×Wgas,t Among them, the normalized actual water consumption Wwater,t is determined by the ratio of the actual water consumption within the statistical period t to the baseline water consumption of the rental property during the same time period under normal historical occupancy conditions; the normalized actual electricity consumption Wpower,t is determined by the ratio of the actual electricity consumption within the statistical period t to the baseline electricity consumption of the rental property during the same time period under normal historical occupancy conditions; the normalized actual gas consumption Wgas,t is determined by the ratio of the actual gas consumption within the statistical period t to the baseline gas consumption of the rental property during the same time period under normal historical occupancy conditions; μ1, μ2, and μ3 are the weights of water consumption, electricity consumption, and gas consumption in the actual comprehensive consumption Ut, respectively. In general scenarios, the value of μ1 ranges from 0.35 to 0.55, the value of μ2 ranges from 0.30 to 0.50, and μ3... The value ranges from 0.05 to 0.25, and μ1+μ2+μ3=1; when the rental house is not connected to the smart gas meter 14 or the gas usage frequency is lower than the preset frequency threshold, μ3=0, the value range of μ1 is from 0.45 to 0.60, the value range of μ2 is from 0.40 to 0.55, and μ1+μ2=1; when the rental house is a centralized heating or centralized hot water scenario, the value range of μ1 is from 0.45 to 0.65, the value range of μ2 is from 0.25 to 0.45, and the value range of μ3 is from 0 to 0.15; when the rental house has a scenario where high-power electrical appliances are continuously running, the value range of μ2 is from 0.40 to 0.60, the value range of μ1 is from 0.25 to 0.45, and the value range of μ3 is from 0.05 to 0.20.

[0055] For water, electricity, and gas consumption, normalization can be achieved using historical maximum values, quantile normalization, or normalization within the rated range based on the apartment layout. For temperature variations, normalization can be achieved using a preset temperature variation range; for the frequency of identity-related behaviors, normalization can be achieved using the maximum number of occurrences within the statistical period. Each weighting coefficient can be preset, manually configured, or obtained from historical data training. However, the weighting coefficients must not be determined solely by a generalization range of 0 to 1, but should meet the specific value ranges and scenario-based adjustment rules defined in this specification for environment, equipment, comprehensive consumption, and identity-behavior consistency scoring.

[0056] In some embodiments, the NPU inference module 45 is further configured to extract monitoring data corresponding to environmental disturbances, appliance status, water leakage events, and non-residential entry events within the current statistical period based on the causal edge markers in the event attribution input frame; read the pre-stored environmental compensation coefficients, equipment disturbance compensation coefficients, water leakage disturbance compensation parameters, and non-residential disturbance compensation parameters in the Flash dedicated parameter storage area 44; match the corresponding calculation rules and coefficient value ranges based on the disturbance types corresponding to various monitoring data, configuration parameters, and causal edge markers; substitute the relevant known data into the preset calculation formulas corresponding to the compensation amounts of each type of disturbance to complete the calculation and obtain the environmental disturbance compensation amount, equipment disturbance compensation amount, water leakage disturbance compensation amount, and non-residential disturbance compensation amount.

[0057] In the application, the environmental disturbance compensation amount Et is calculated based on the environmental-induced energy change markers and environmental disturbance data in the event attribution input frame. The calculation formula is as follows: Et=α1×ΔTt+α2×|ΔTt|+α3×Ht+α4×Wt Wherein, ΔTt is the short-term temperature change within the statistical period t, ranging from -30℃ to 30℃; Ht is the normalized humidity value, ranging from 0 to 1; Wt is the normalized weather value, ranging from 0 to 1; α1, α2, α3, and α4 are environmental compensation coefficients, where α1 ranges from 0.10 to 0.35, α2 from 0.10 to 0.30, α3 from 0.05 to 0.20, and α4 from 0.05 to 0.20, and α1+α2+α3+α4≤1. In areas with centralized heating in winter, α1 and α2 can take lower values ​​within the above ranges; in high-temperature summer or electric heating scenarios in winter, α1 and α2 can take higher values ​​within the above ranges.

[0058] The equipment disturbance compensation amount Dt is calculated based on the equipment-induced energy consumption change marker and appliance status change data in the event attribution input frame. The calculation formula is as follows: Dt=∑(k=1 to K)βk×ak,t Where K represents the number of appliance state change types, ranging from 1 to 20; ak,t represents the state change marker of the k-th type of appliance within the statistical period t, with a value of 0 or 1, or a normalized intensity value between 0 and 1; βk represents the equipment disturbance compensation coefficient corresponding to the k-th type of appliance state change. When the k-th type of appliance is a temperature-controlled device such as an air conditioner, water heater, or electric heater, the value of βk ranges from 0.20 to 0.45; when the k-th type of appliance is a short-term high-load device such as a washing machine or dryer, the value of βk ranges from 0.10 to 0.30; when the k-th type of appliance is a normally-on low-power device such as a refrigerator or router, the value of βk ranges from 0.02 to 0.15; when the k-th type of appliance is other ordinary electrical equipment, the value of βk ranges from 0.01 to 0.20; and ∑βk≤1.

[0059] The leakage disturbance compensation amount Lt is calculated based on the leakage-induced continuous water use marker in the event attribution input frame, and its value ranges from 0 to 1, or from 0 to 80% of the actual comprehensive water use Ut. When the leakage marker is triggered, Lt can be the normalized cumulative water use of the corresponding continuous water use event; when the leakage sensor 15 is not triggered and the continuous water use event is time-related to the resident tenant's entry event, Lt=0; when the leakage sensor 15 is triggered, or the continuous water use duration exceeds the leakage judgment duration threshold of 10 minutes to 180 minutes and is not time-related to the resident tenant's entry event, Lt is greater than 0.

[0060] The non-resident identity disturbance compensation amount Qt is calculated based on the short-term usage marker caused by the non-resident identity in the event attribution input frame, and its value ranges from 0 to 1, or from 0 to 60% of the actual comprehensive usage Ut. When there are no in-home events corresponding to visitor, cleaning, maintenance, housekeeper, or temporary authorized identities within the statistical period t, Qt=0; when there are in-home events of the above non-resident identities and short-term usage events occur within the fourth preset time window of 1 minute to 240 minutes thereafter, Qt is greater than 0.

[0061] In some embodiments, the NPU inference module 45 is also used to read the basic usage baseline related configuration parameters pre-stored in the Flash dedicated parameter storage area 44, obtain the basic usage baseline value determined by the month, time segment, unit type parameters and the number of registered residents, retrieve the calculated event disturbance compensation amount, environmental disturbance compensation amount and equipment disturbance compensation amount, substitute the basic usage baseline and various disturbance compensation amount data into the calculation formula corresponding to the counterfactual prediction assessment amount to complete the calculation, and generate the counterfactual prediction assessment amount under the condition that the number of registered residents remains unchanged.

[0062] In response to the feature-ready hardware signal, the NPU inference module 45 reads the parameters from the event attribution input frame and the Flash dedicated parameter storage area 44, and generates a counterfactual prediction assessment quantity under the condition that the number of registered residents remains unchanged. The calculation formula is as follows: Ûcf_t=Bm,s,h,n+Et+Dt+Lt+Qt Wherein, Ûcf_t is the counterfactual prediction assessment quantity within the statistical period, Bm,s,h,n is the basic usage baseline determined by month m, time period s, apartment type parameter h, and registered resident number n, and Et, Dt, Lt, and Qt are the environmental disturbance compensation quantity, equipment disturbance compensation quantity, leakage disturbance compensation quantity, and non-resident identity disturbance compensation quantity, respectively. Et, Dt, Lt, and Qt are not output as independent judgment results, but are written into the counterfactual prediction assessment quantity as usage components that can be explained by environment, equipment, leakage, and non-resident identity. The basic usage baseline Bm,s,h,n and various disturbance compensation quantities together represent the comprehensive usage that the system believes should occur when the registered resident number remains unchanged. The basic usage baseline Bm,s,h,n is generated through pre-training using historical normal occupancy data of rental housing and stored in the Flash dedicated parameter storage area 44.

[0063] In this embodiment, the duration of the statistical period is 15 minutes to 24 hours, preferably 1 hour, 6 hours, 12 hours or 24 hours; the month m is 1 to 12; the time segment number s is 1 to 96; the number of registered residents n is 1 to 20; h includes at least one or more of the following: building area, number of rooms, number of water points, number of electrical circuits; the baseline of basic usage Bm,s,h,n is 0 to 1 when it is a normalized value, or 0 to 100 when it is a percentage normalized value.

[0064] Based on this, the residential behavior residual Rt is obtained by subtracting the counterfactual prediction assessment amount Ûcf_t from the actual comprehensive usage Ut, using the following formula: Rt=Ut-Ûcf_t Where Rt represents the residual of residential behavior within the statistical period t. Since the counterfactual prediction assessment already includes the baseline usage and the disturbance compensation for environment, equipment, leakage, and non-resident status, Rt represents the remaining usage change that cannot be explained after deducting the above non-personnel factors. This remaining usage change is used as a residential behavior feature caused by suspected new residents and input into the subsequent population estimation inference model. Rt can be positive, zero, or negative. When Rt is less than 0, the NPU inference module 45 can truncate Rt to 0 or retain a negative value as a marker that is lower than the counterfactual prediction assessment. When used for calculating the estimated number of new residents, R+t=max(Rt,0) is preferably used as the effective residential behavior residual, and the value of R+t ranges from 0 to 1.

[0065] In some embodiments, the FPGA-dedicated event pipeline 43 is also used to standardize event indication data to generate event nodes, retrieve preset time window parameters, identity type matching rules, and status flag configuration parameters, verify each event node data and corresponding configuration parameter to see if they meet the preset time window, identity type, and status flag judgment conditions, generate causal edge tags between event nodes after matching the corresponding association rules, and then retrieve the event nodes that have been judged and their corresponding causal edge tags, integrate and package them according to the preset frame structure and encapsulation rules, and encapsulate the event nodes and their corresponding causal edge tags into an event attribution input frame.

[0066] The preset integrity conditions for the event attribution input frame are as follows: the event node quantity field, event type field, event timestamp field, identity type field, causal edge marker field, disturbance type field, and frame verification field are all written, and the verification value of the frame verification field meets the preset verification conditions.

[0067] It is understandable that the FPGA-dedicated event pipeline 43 will perform unified and standardized processing on the event indication data transmitted from the front end, converting the raw data into event nodes with standardized formats. Then, it will call the system's pre-configured time window parameters, identity type matching rules, and status flag configuration parameters, and perform condition verification one by one based on the raw data of each event node and the called configuration parameters to determine whether it meets the various judgment requirements of the preset time window, identity type, and status flag. After matching the corresponding event association rules, it will generate causal edge tags between different event nodes. Then, it will collect all event nodes that have completed condition verification and their matching causal edge tags, and perform integration and packaging operations in strict accordance with the frame structure and encapsulation rules set by the system. Finally, it will encapsulate the event nodes and their corresponding causal edge tags into an event attribution input frame that can be called by subsequent modules.

[0068] In some embodiments, the NPU inference module 45 is further configured to determine an identity-behavior consistency score based on the event attribution input frame, the residential behavior residual feature frame, and historical event statistics; to identify two abnormal scenarios—temporary authorization becoming permanent and suspected multiple-person shared identity—based on the identity-behavior consistency score; and to adjust the non-residential identity disturbance compensation amount based on the identity-behavior consistency score. Specifically, when the identity-behavior consistency score is higher than a first identity threshold, the non-residential identity disturbance compensation amount corresponding to that identity is reduced or canceled, and its corresponding residual is included in the estimation of newly added residents; when the score is lower than a second identity threshold, the non-residential identity disturbance compensation amount is retained or increased.

[0069] In the application, the NPU inference module 45 calls the event nodes, causal edge markers and other related data contained in the event attribution input frame, the additional water, electricity and gas usage deviations and time series feature data recorded in the residential behavior residual feature frame, and the historical records such as the target identity's cross-day activity frequency and past behavior patterns stored in the historical event statistics. It extracts the core behavioral features associated with the identity from the three types of data sources, and combines them with the preset scoring weight coefficients in the Flash dedicated parameter storage area 44. Through the operation logic of positive feature weighted accumulation and abnormal identity mark deduction, the identity-behavior consistency score is determined (the higher the score, the more the identity behavior fits the residential characteristics of permanent tenants). Then, based on the scoring threshold and judgment rules, it accurately judges the "temporary authorization becoming long-term" scenario where the temporary authorized identity appears repeatedly for multiple periods and the behavior fits the permanent characteristics, and the "suspected multiple people sharing identity" scenario where multiple independent residential behavior residuals and contradictory behavior patterns appear under the same identity. In addition, the degree of association between the non-residential attribute and the identity is quantified based on the score. The higher the score, the weaker the non-residential attribute and the stronger the actual residential attribute. Based on this, the value of the non-residential identity disturbance compensation is dynamically adjusted to ensure that the compensation amount matches the actual usage scenario of the identity and improve the accuracy of overall usage calculation and overcrowding identification.

[0070] In some embodiments, the NPU inference module 45 is further configured to extract from the event attribution input frame, the residential behavior residual feature frame, and historical event statistics the cross-day recurrence frequency, nighttime residual occurrence frequency, morning and evening peak residual occurrence frequency, first-time usage residual delay stability after unlocking, visitor type marker, and non-residential identity marker corresponding to the target identity, respectively. It then performs normalization processing on the first four indicators according to preset rules, reads the identity-behavior scoring weight coefficients pre-stored in the Flash dedicated parameter storage area 44, matches the corresponding calculation rules based on various feature data, identity marker data, and weight coefficients, substitutes the relevant known data into the preset calculation formula corresponding to the identity-behavior consistency score to complete the calculation, and outputs the identity-behavior consistency score corresponding to the target identity.

[0071] The residential behavior residual feature frame may include the actual comprehensive assessment quantity field, the counterfactual prediction assessment quantity field, the residential behavior residual field, the environmental disturbance compensation field, the equipment disturbance compensation field, the water leakage disturbance compensation field, the non-residential identity disturbance compensation field, the identity-behavior consistency score field, the cross-period repeated residual field, the nighttime residual field, the morning and evening peak residual field, the temporary authorization long-term marker field, the suspected multiple-shared identity marker field, the estimated value of newly added residents field, and the attribution explanation marker field. The residential behavior residual field is the usage deviation data obtained by calculating the difference between the actual comprehensive assessment and the counterfactual predicted assessment. The cross-period repeated residual field records the frequency, duration, and amplitude of the residential behavior residual in multiple consecutive statistical periods to determine whether the usage deviation has long-term characteristics. The nighttime residual field specifically collects residential behavior residual data detected in the system's preset nighttime period, distinguishing between temporary visits and long-term residence based on nighttime energy consumption characteristics. The morning and evening peak residual fields conduct residual statistics for the two peak periods of resident activity, using the energy consumption patterns during peak periods to determine the residential attributes and activity characteristics of residents. The Flash dedicated parameter storage area 44 is used to store basic usage baseline parameters, environmental compensation coefficients, equipment disturbance compensation coefficients, leakage disturbance compensation parameters, non-resident identity disturbance compensation parameters, comprehensive usage weights, identity-behavior scoring weights, continuous statistical period parameters, preset time window parameters, and dynamic overcrowding judgment parameters.

[0072] NPU inference module 45 calculates the identity-behavior consistency score according to the following formula: Sp=γ1×Fp+γ2×Np+γ3×Mp+γ4×Lp-γ5×Vp-γ6×Qp Wherein, Fp, Np, Mp, and Lp are normalized values ​​from 0 to 1, and Vp and Qp are state flag values ​​of 0 or 1; γ1 to γ6 are preset weight coefficients, where γ1 ranges from 0.15 to 0.30, γ2 ranges from 0.15 to 0.30, γ3 ranges from 0.10 to 0.25, γ4 ranges from 0.05 to 0.20, γ5 ranges from 0.05 to 0.25, and γ6 ranges from 0.05 to 0.25, and γ1+γ2+γ3+γ4+γ5+γ6=1. Sp ranges from -1 to 1. When Sp is greater than the first identity threshold configured in the range of 0.3 to 0.8, the behavior corresponding to the identity is determined to be closer to permanent resident behavior; when Sp is less than the second identity threshold configured in the range of -0.8 to 0.3, the behavior corresponding to the identity is determined to be closer to non-resident short-term behavior.

[0073] Based on this, the NPU, using the residential behavior residual feature frames and a pre-trained population estimation inference model, outputs an estimated value ΔP for the number of new residents. The population estimation inference model can be a lightweight neural network model, a regression model, a decision tree model, or a rule-based and model-combined inference model. The input to this model is the core feature dimensions in the residential behavior residual feature frames, including the magnitude of the residential behavior residual, duration, cross-cycle repetition, living time period matching degree, and identity-behavior consistency score. The output is the estimated number of new residents, which can be an integer value, a discrete level, or a rounded value. The value of ΔP ranges from 0 to 20, preferably from 0 to 10. When ΔP is a discrete level, its level can include level 0, level 1, level 2, level 3, level 4, and level 5, corresponding to no new residents, suspected 1 new resident, suspected 2 new residents, suspected 3 new residents, suspected 4 new residents, and suspected 5 or more new residents, respectively.

[0074] In some embodiments, the rental housing overcrowding detection system based on residual correction may further include a local linkage control module. This module interfaces with the execution terminal and, upon determining that there is unauthorized overcrowding, drives the execution terminal to perform a preset local linkage operation. In this embodiment, the execution terminal may include an audible and visual alarm and an access control module. The linkage operation may include triggering an audible and visual alarm, restricting temporary unlocking permissions for unregistered individuals, etc. The linkage strategy can be configured according to actual needs and is not limited to these.

[0075] In some embodiments, please refer to Figure 5 This application provides a method for identifying overcrowded rental housing based on residual correction. The method for identifying overcrowded rental housing based on residual correction is executed based on any of the above schemes of the system for identifying overcrowded rental housing based on residual correction. The method for identifying overcrowded rental housing based on residual correction includes the following steps S501 to S504.

[0076] S501: Acquire event indication data, appliance status change data, and environmental disturbance data.

[0077] S502: Generate the actual comprehensive assessment quantity and event disturbance compensation quantity according to the event indication data, generate the equipment disturbance compensation quantity according to the appliance status change data, and generate the environmental disturbance compensation quantity according to the environmental disturbance data.

[0078] S503: Based on the event disturbance compensation amount, equipment disturbance compensation amount, and environmental disturbance compensation amount, and the baseline usage, a counterfactual prediction assessment amount is generated under the condition that the number of registered residents remains unchanged.

[0079] S504: Determine the overcrowding identification result based on the difference between the actual comprehensive assessment quantity and the counterfactual predicted assessment quantity, and send the overcrowding identification result to the cloud server.

[0080] It should be noted that the method for identifying overcrowded rental housing based on residual correction provided in this application embodiment and the system for identifying overcrowded rental housing based on residual correction provided in this application embodiment are based on the same inventive concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned system for identifying overcrowded rental housing based on residual correction, and the repeated parts will not be described again.

[0081] In some embodiments, an electronic device provided in this application includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the above-described method for identifying overcrowded rental housing based on residual correction.

[0082] Specifically, the processor may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0083] Memory can be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory can include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and also random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0084] This application also provides a non-transitory computer storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned method for identifying overcrowded rental housing based on residual correction. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method as described in the embodiments of this application.

[0085] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0086] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application, even if such combinations or combinations are not explicitly described in this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A rental housing overcrowding identification system based on residual correction, characterized in that, include: Sensing module, home appliance monitoring module, environmental monitoring module, and integrated smart gateway; The sensing module is used to generate event indication data for the rental house. The sensing module includes an electronic lock for the entrance, a smart water meter, a smart electricity meter, a smart gas meter, and a water leakage sensor. The home appliance monitoring module is used to collect and output home appliance status change data, which includes home appliance power consumption data, posture recording data, and home appliance type labels. The environmental monitoring module is used to collect and output environmental disturbance data, which includes temperature changes over a predetermined time, humidity information, and weather conditions. The integrated smart gateway is used to generate actual comprehensive assessment quantity and event disturbance compensation quantity according to event indication data, generate equipment disturbance compensation quantity according to home appliance status change data, generate environmental disturbance compensation quantity according to environmental disturbance data, generate counterfactual prediction assessment quantity under the condition of unchanged number of registered residents by combining event disturbance compensation quantity, equipment disturbance compensation quantity, environmental disturbance compensation quantity and basic usage baseline, determine the overcrowding identification result based on the difference between actual comprehensive assessment quantity and counterfactual prediction assessment quantity, and send the overcrowding identification result to cloud server.

2. The rental housing overcrowding identification system based on residual correction as described in claim 1, characterized in that, The integrated smart gateway includes: a DMA controller, a dual-port RAM cache, an FPGA dedicated event pipeline, a Flash dedicated parameter storage area, an NPU inference module, and an illegal occupancy identification module; The DMA controller is used to directly write event indication data, home appliance status change data, and environmental disturbance data into a dual-port RAM cache. The FPGA-dedicated event pipeline is used to determine, based on event indication data, home appliance status change data and environmental disturbance data, when the event node meets the preset time window, identity type and status label conditions, generate causal edge labels between event nodes, encapsulate the event node and the corresponding causal edge labels into an event attribution input frame, and output a feature-ready hardware signal to the NPU inference module when the event attribution input frame meets the preset integrity conditions. The NPU inference module is used to respond to the feature-ready hardware signal, read the event attribution input frame and the parameters in the Flash dedicated parameter storage area, generate the actual comprehensive evaluation quantity, and generate the event disturbance compensation quantity, environmental disturbance compensation quantity and equipment disturbance compensation quantity according to the causal edge marker in the event attribution input frame. Based on the event disturbance compensation quantity, environmental disturbance compensation quantity, equipment disturbance compensation quantity and basic usage baseline, it generates the counterfactual prediction evaluation quantity under the condition of unchanged registered residents. Based on the difference between the actual comprehensive evaluation quantity and the counterfactual prediction evaluation quantity, it generates the residential behavior residual feature frame. Based on the residential behavior residual feature frame, it outputs the estimated value of the newly added residents. The event disturbance compensation quantity includes the leakage disturbance compensation quantity and the non-resident identity disturbance compensation quantity. The illegal occupancy identification module is used to generate an estimated number of actual residents based on the number of registered residents and the estimated number of newly added residents. When the estimated number of actual residents exceeds the rated number of residents in the rental house for a consecutive preset number of statistical periods, it is determined that the rental house has illegal overcrowding.

3. The rental housing overcrowding identification system based on residual correction as described in claim 2, characterized in that, The NPU inference module is also used to extract normalized actual water consumption, normalized actual electricity consumption, and normalized actual gas consumption data calculated based on historical normal residential baselines within the current statistical period from the event attribution input frame. Then, it reads the configuration parameters related to water consumption, electricity consumption, and gas consumption weights pre-stored in the Flash dedicated parameter storage area. Based on the normalized actual water consumption, normalized actual electricity consumption, normalized actual gas consumption data, configuration parameters, and the operating condition attributes of the rental house, it matches the corresponding weight value rules to determine the water consumption weight value, electricity consumption weight value, and gas consumption weight value. Based on each weight value, it completes the calculation according to the preset weighted calculation formula to generate the actual comprehensive evaluation quantity.

4. The rental housing overcrowding identification system based on residual correction as described in claim 2, characterized in that, The NPU inference module is also used to extract monitoring data corresponding to environmental disturbances, appliance status, water leakage events, and non-resident entry events within the current statistical period based on the causal edge markers in the event attribution input frame. It reads the pre-stored environmental compensation coefficients, equipment disturbance compensation coefficients, water leakage disturbance compensation parameters, and non-resident entry disturbance compensation parameters in the Flash dedicated parameter storage area. Based on the disturbance types corresponding to various monitoring data, configuration parameters, and causal edge markers, it matches the corresponding calculation rules and coefficient value ranges, substitutes the relevant known data into the preset calculation formulas corresponding to the compensation amounts of each type of disturbance to complete the calculation, and obtains the environmental disturbance compensation amount, the equipment disturbance compensation amount, the water leakage disturbance compensation amount, and the non-resident entry disturbance compensation amount.

5. The rental housing overcrowding identification system based on residual correction as described in claim 2, characterized in that, The NPU inference module is also used to read the basic usage baseline related configuration parameters pre-stored in the Flash dedicated parameter storage area, obtain the basic usage baseline value determined by the month, time segment, unit type parameters and the number of registered residents, retrieve the calculated event disturbance compensation amount, environmental disturbance compensation amount and equipment disturbance compensation amount, substitute the basic usage baseline and various disturbance compensation amount data into the calculation formula corresponding to the counterfactual prediction assessment amount to complete the calculation, and generate the counterfactual prediction assessment amount under the condition that the number of registered residents remains unchanged.

6. The rental housing overcrowding identification system based on residual correction as described in claim 2, characterized in that, The FPGA-dedicated event pipeline is also used to standardize the event indication data to generate event nodes, retrieve preset time window parameters, identity type matching rules, and status flag configuration parameters, and verify each event node data and corresponding configuration parameter to see if they meet the preset time window, identity type, and status flag judgment conditions. After matching the corresponding association rules, causal edge tags between event nodes are generated. Then, the event nodes that have been judged and their corresponding causal edge tags are retrieved and integrated and packaged according to the preset frame structure and encapsulation rules. The event nodes and their corresponding causal edge tags are uniformly encapsulated into the event attribution input frame.

7. The rental housing overcrowding identification system based on residual correction as described in claim 6, characterized in that, The FPGA-dedicated event pipeline generates causal edge tags according to the following rules: If the entry event node corresponding to the permanent tenant's identity is time-related to the water, electricity and gas event nodes within the first preset time window, then the entry-caused water mark, entry-caused electricity mark and entry-caused gas mark will be generated sequentially. If the environmental disturbance event node corresponding to the temperature change at a preset time is temporally correlated with the power consumption change event node within the second preset time window, an environmental energy consumption change marker is generated. If an event node indicating a change in the status of a household appliance is time-related to an event node indicating a sudden change in power consumption within the third preset time window, a marker indicating a change in energy consumption caused by the device is generated. If the leakage status signal coincides with the time of the continuous water use event node, or if the continuous water use event node is not time-correlated with the permanent tenant's entry event node within the fifth preset time window, a leakage causing continuous water use marker will be generated. If the entry event node corresponding to a non-resident identity forms a time correlation with the water, electricity and gas event nodes within the fourth preset time window, a short-term usage marker caused by non-resident identity is generated. If the event node corresponding to the temporary authorized identity appears repeatedly in multiple consecutive statistical periods and is temporally related to the residential behavior residual, a temporary authorization long-term marker is generated. The residential behavior residual is the difference between the actual comprehensive assessment quantity and the counterfactual predicted assessment quantity. If multiple independent residential behavior residual fragments appear in the event node corresponding to the same identity identifier, a suspected shared identity tag is generated.

8. The rental housing overcrowding identification system based on residual correction as described in claim 2, characterized in that, The NPU inference module is also used to determine the identity-behavior consistency score based on the event attribution input frame, the residential behavior residual feature frame and historical event statistics, to identify two abnormal scenarios, namely temporary authorization becoming long-term and suspected multiple people sharing the same identity, based on the identity-behavior consistency score, and to adjust the non-residential identity disturbance compensation amount based on the identity-behavior consistency score.

9. The rental housing overcrowding identification system based on residual correction as described in claim 8, characterized in that, The NPU inference module is also used to extract the following from the event attribution input frame, the residential behavior residual feature frame, and historical event statistics: the frequency of cross-day recurrence, the number of nighttime residual occurrences, the number of morning and evening peak residual occurrences, the stability of the first usage residual delay after unlocking, the visitor type marker, and the non-resident identity marker. The module then performs normalization processing on the first four indicators according to preset rules, reads the identity-behavior scoring weight coefficients pre-stored in the Flash dedicated parameter storage area, matches the corresponding calculation rules based on various feature data, identity marker data, and weight coefficients, substitutes the relevant known data into the preset calculation formula corresponding to the identity-behavior consistency score to complete the calculation, and outputs the identity-behavior consistency score corresponding to the target identity.

10. A method for identifying overcrowding in rental housing based on residual correction, characterized in that, The method for identifying overcrowded rental housing based on residual correction is performed based on the system for identifying overcrowded rental housing based on residual correction as described in any one of claims 1 to 9; The method for identifying overcrowding in rental housing based on residual correction includes: Acquire event indication data, appliance status change data, and environmental disturbance data; Based on the event indication data, generate the actual comprehensive assessment quantity and event disturbance compensation quantity; based on the appliance status change data, generate the equipment disturbance compensation quantity; based on the environmental disturbance data, generate the environmental disturbance compensation quantity. The counterfactual prediction assessment quantity is generated by combining the event disturbance compensation quantity, equipment disturbance compensation quantity, and environmental disturbance compensation quantity with the basic usage baseline, under the condition that the number of registered residents remains unchanged. The overcrowding identification result is determined based on the difference between the actual comprehensive assessment quantity and the counterfactual predicted assessment quantity, and the overcrowding identification result is sent to the cloud server.

Images