Household intelligent gateway system

Abstract

The invention relates to the technical field of communication and network control, and discloses a home intelligent gateway system which comprises an environment sensing module, an identity recognition module, a behavior prediction module, a strategy scheduling module and an edge calculation and feedback module. The environment sensing module is used for collecting and coding environment state data; the identity recognition module generates an identity vector through multi-modal fusion; the behavior prediction module predicts user behavior preference and equipment use intention in combination with the identity vector and the state parameter set; the strategy scheduling module generates an equipment control strategy according to the prediction result and performs energy consumption optimization; and the edge calculation and feedback module performs error correction and sampling frequency dynamic adjustment on the control result. According to the system, the self-adaptive optimization and energy consumption balance of equipment control are realized by constructing a user personalized behavior model, the intelligent response capability and the operation efficiency of a home automation system are effectively improved, and the system is suitable for various intelligent home network environments.

Description

Technical Field

[0001] This invention relates to the field of communication and network control technology, specifically to a home intelligent gateway system, and more particularly to efficient home device management under multimodal sensing, personalized behavior prediction and energy consumption control. Background Technology

[0002] With the rapid development of IoT technology and artificial intelligence, home automation systems are becoming increasingly popular worldwide. Existing smart home systems are mostly based on distributed control, enabling remote control and timed management of home appliances via mobile apps or voice assistants. However, these systems typically lack deep understanding of the home environment and user behavior, failing to automatically adjust device status based on user identity and real-time scenarios, resulting in limited intelligence and a poor user experience.

[0003] In existing technologies, environmental perception often relies on single or limited types of sensors, such as temperature, humidity, or light sensors, providing only static data feedback and failing to achieve multi-source data fusion processing. Furthermore, changes in environmental parameters often fail to trigger behavioral prediction mechanisms, resulting in equipment control strategies being unable to respond promptly to scene changes. In addition, existing systems often lack edge feedback and self-optimization capabilities, leading to unidirectional execution of equipment control decisions, a lack of closed-loop feedback, and low control accuracy and energy efficiency.

[0004] Some systems attempt to incorporate identity recognition and behavior prediction algorithms, but most are based on single-modal features (such as facial or voice) and lack multimodal data fusion and dynamic behavior modeling mechanisms, failing to accurately identify family members' personalized preferences and device usage intentions. Furthermore, traditional control strategies cannot be collaboratively optimized by combining user behavior preferences, environmental conditions, and device power consumption constraints, resulting in policy conflicts and energy waste. Therefore, there is an urgent need for an integrated home smart gateway system with multi-dimensional perception, identity recognition, behavior prediction, policy optimization, and edge feedback capabilities. Summary of the Invention

[0005] To achieve the above-mentioned objectives, the present invention provides the following technical solution: a home intelligent gateway system, comprising the following modules:

[0006] The environmental sensing module is used to collect indoor environmental data through multiple types of sensors. env The environmental data D env This includes temperature, humidity, light intensity, and air quality. The raw data undergoes preliminary filtering and normalization to generate a structured set of state parameters S. env ;

[0007] The identity recognition module is used to identify family members based on multimodal information and collect raw data D. id The original data Did This includes facial images, voiceprint data, and behavioral sequences, which are fused together to form a unified identity vector V. id ;

[0008] The behavior prediction module is used to predict behavior based on the identity vector V. id With the set of state parameters S env Predicting family members' behavioral preferences P beh With regard to the intended use of the equipment I dev And generate control command parameters C pred ;

[0009] The strategy scheduling module is used to schedule tasks based on control command parameters C. pred Generate the optimal equipment control strategy R ctrl And adjust the device power consumption allocation factor E according to the energy optimization algorithm. opt ;

[0010] Edge computing and feedback module, used for real-time processing of device control strategies. ctrl Compared with actual equipment response data R dev Generate feedback correction coefficient F adj And optimize the parameter set S env The update frequency.

[0011] Preferably, the environment sensing module includes:

[0012] Data acquisition unit: Employs temperature and humidity sensors, photosensors, and an air quality module to acquire raw environmental data in real time. env ={d1,d2,…,d n};

[0013] Data cleaning and anomaly removal unit: Identifies fluctuating data, when... At that time, remove data d i , where d i For the i-th set of data collected by the sensor, This represents the average value of the collected data.

[0014] State coding unit: through a normalization function The cleaned raw data is encoded into a structured set of state parameters S. env ={s1,s2,…,s n The structured state parameter set S is stored in a local cache and, after normalization, is processed according to the type of environmental parameters. env Label the parameters and arrange them according to the time sequence of parameter collection to ensure that the system prioritizes the identification of the latest high-weight environmental parameters when updating the status.

[0015] Preferably, the identity recognition module includes:

[0016] Multimodal feature acquisition unit: synchronously acquires image data I fac Speech waveform A voc With behavior vector trajectory B seq And transform it into a preliminary feature set {F} I ,F A ,F B};

[0017] Feature fusion and confidence generation unit: A gated residual fusion network is used to construct a fusion mapping function to generate the identity vector V. id :

[0018] V id =σ(W1F I +W2F A +W3F B +b);

[0019] Where W1, W2, and W3 are weight matrices, b is a bias term, and σ(·) is a normalized activation function;

[0020] Identity discrimination and matching unit: Calculates identity vector V id The Euclidean distance to each identity label V in the known vector library, if If the identity is identified, it is considered to be recognized; otherwise, an abnormal alarm mechanism is triggered. When performing identity matching, the system prioritizes the comparison of the identity vectors identified in the most recent three times. If the identity is known in all three consecutive identifications, the system automatically extends the validity period of the identity tag in the vector library. If two consecutive identifications fail, the system lock mechanism is triggered and an alarm message is sent to the administrator terminal.

[0021] Preferably, the behavior prediction module includes:

[0022] Input feature concatenation unit: This unit combines the identity vector V with the input feature concatenation unit. id With the set of state parameters S env Concatenate the vectors to construct the behavioral input vector X beh =[V id ,S env ];

[0023] Behavioral pattern modeling unit: Introducing temporal attention weight α t Spatial attention β s The behavioral preference probability vector P is output using the following formula. beh :

[0024]

[0025] in:

[0026] X beh The input vector is the behavior.

[0027] X beh T This is the transpose of the input vector.

[0028] V id -V represents the identity vector V. id The Euclidean distance to each identity label V in the known vector library;

[0029] α t Assuming time-based attention weights, if user behavior is more representative in the morning, then α will be used more heavily when predicting morning behavior. t =1, when predicting afternoon behavior, α t =0;

[0030] β s Spatial attention represents the degree of influence a user has across different spaces, reflecting spatial correlation. If the correlation is weak, then β... s =0;

[0031] Then calculate the device usage intention I using the following formula. dev :

[0032] I dev =V id ×β s ;

[0033] Control parameter generation unit: based on preference probability vector P beh and device usage intention I dev Combine and calculate control command parameters C pred :

[0034] C pred =f(P beh ,I dev );

[0035] Preferably, the strategy scheduling module includes:

[0036] Control strategy generation unit: Constructs control command mapping matrix M ctrl Calculate the equipment control strategy R according to the following formula. ctrl :

[0037] R ctrl =M ctrl ·C pred ;

[0038] Energy consumption optimization unit: Introduces an energy consumption minimization objective function to calculate the device power consumption allocation factor E. opt :

[0039]

[0040] Where m is the total number of devices. For the intended use of the i-th device, P i Let be the electrical power of the i-th device, and be the inherent parameters of the device;

[0041] Policy caching and conflict reconciliation unit: The control policy queue is optimized by using a timestamp conflict sorting algorithm to avoid rapid and repeated switching of device states. After executing the timestamp conflict sorting algorithm, a maximum state change threshold is set. When the number of control state changes of a device exceeds the threshold within a set time window, the execution of its subsequent control commands is suspended and a control conflict log is recorded for subsequent policy optimization.

[0042] Preferably, the edge computing and feedback module includes:

[0043] Real-time feedback acquisition unit: Acquires actual device response data R after actual execution. dev Equipment control strategy R ctrl Compare and calculate the feedback error ε:

[0044] ε=|R dev -R ctrl |;

[0045] Feedback correction coefficient generation unit: Estimates the feedback error and derives the feedback correction coefficient F. adj :

[0046] F adj =K t ·ε+(1-K t )·R dev ;

[0047] Among them, K t For adaptive weight gain, the initial value is set to 0.5. When ε is too large, compensation is performed according to the following formula:

[0048]

[0049] Acquisition Frequency Optimization Unit: Adjusts the sensor sampling period τ in the environmental perception module by combining correction coefficients. s The optimal condition is one that satisfies the following conditions:

[0050]

[0051] Where E(τ) is the energy consumption per unit time at the sampling frequency, and τ is the inherent parameter of the various types of sensors in the environmental perception module.

[0052] Preferably, during system operation, each module executes sequentially according to the following collaborative order:

[0053] First, the environmental sensing module acquires indoor environmental data D.env And generate a structured set of state parameters S env ;

[0054] Then, the identity recognition module uses multimodal data to construct an identity vector V. id ;

[0055] Then the behavior prediction module uses the set of state parameters S env With identity vector V id Calculate behavioral preferences P beh and equipment usage intention I dev Generate control command parameters C pred ;

[0056] Next, the strategy scheduling module determines the control command parameter C. pred Output device control strategy R ctrl And optimize the energy consumption factor E opt ;

[0057] Finally, the edge computing and feedback module obtains the actual device response data R. dev And generate feedback correction coefficient F adj Based on this coefficient, the set of state parameters S is dynamically optimized. env Sampling period τ of the environmental perception module s .

[0058] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0059] This invention achieves multimodal environmental perception and individual recognition linkage, improving the accuracy of intelligent response: By integrating multiple environmental sensors and identity recognition modules, and combining semantic fusion algorithms, the invention jointly models the state of the home environment and the identities of family members, enabling real-time matching of control strategies and significantly improving the system's response accuracy and adaptability to complex home scenarios.

[0060] Introducing user behavior prediction and energy consumption optimization mechanisms to enhance the system's intelligent decision-making capabilities: This invention adopts a user behavior modeling algorithm based on neural networks, combined with the usage habits and historical data of family members, to dynamically adjust the control strategy; at the same time, it integrates energy consumption prediction formulas to minimize the power consumption of equipment scheduling, improve the energy efficiency ratio, and reduce operating costs.

[0061] Constructing an edge intelligent feedback closed loop to enhance system stability and autonomous learning capabilities: This invention deploys an edge computing module locally, combining state feedback and optimization learning algorithms to evaluate the device's performance in real time and fine-tune strategies, constructing a "perception-prediction-decision-feedback" closed-loop control system, which significantly improves the system's stability, robustness, and continuous optimization capabilities. Attached Figure Description

[0062] Figure 1This application provides a schematic diagram of the system modules.

[0063] Figure 2 A schematic diagram of the environmental sensing module provided in this application;

[0064] Figure 3 A schematic diagram of the identity recognition module provided in this application;

[0065] Figure 4 A schematic diagram of the behavior prediction module provided in this application;

[0066] Figure 5 A schematic diagram of the strategy scheduling module provided in this application;

[0067] Figure 6 A schematic diagram of the edge computing and feedback module provided in this application. Detailed Implementation

[0068] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0069] refer to Figures 1-6 This invention provides a home smart gateway system, comprising the following modules:

[0070] The environmental sensing module is used to collect indoor environmental data through multiple types of sensors. env The environmental data D env This includes temperature, humidity, light intensity, and air quality. The raw data undergoes preliminary filtering and normalization to generate a structured set of state parameters S. env .

[0071] The system first constructs a basic environmental sensing network using temperature and humidity sensors, light sensors, and air quality modules installed in each room. Each sensor updates its signal every τ... s = Data collection is performed every 10 seconds, and the resulting raw environmental data set is represented as follows:

[0072] D env ={d1,d2,…,d n};

[0073] The system performs initial screening of the dataset using a data cleaning and anomaly removal unit. If any sampled data d is detected... i Meet the conditions Data deemed abnormal is then discarded to avoid misjudgments caused by occasional sensor errors. Based on this, the state coding unit performs normalization processing on the remaining data using the formula: The normalized set of structured state parameters is S env ={s1,s2,…,s n The data is further sorted in ascending order by acquisition time to ensure that the latest parameters have higher weight in subsequent analyses. In addition, the system attaches labels to different environmental parameters to identify their source and category during multimodal fusion. This module uses local caching to achieve real-time updates and historical tracking of state data, ensuring the continuity and stability of system response.

[0074] The identity recognition module is used to identify family members based on multimodal information and collect raw data D. id The original data D id This includes facial images, voiceprint data, and behavioral sequences, which are fused together to form a unified identity vector V. id .

[0075] The identity recognition module is deployed in the main areas of the home entrance hall and living room, and works in conjunction with voice interaction devices and surveillance cameras. The system simultaneously collects three types of data through a multimodal feature acquisition unit: image data I... fac Speech waveform A voc With behavioral sequence trajectory B seq Then convert it into the corresponding feature vector set {F} I ,F A ,F B The feature fusion and confidence generation unit uses a gated residual fusion network to integrate the above features into a unified identity vector V. id The fusion formula is as follows:

[0076] V id =σ(W1F I +W2F A +W3F B +b);

[0077] Where W1, W2, and W3 are the training weight matrices for each modality, and σ(·) is the normalized activation function. The identity discrimination and matching unit uses the Euclidean distance algorithm to determine the minimum distance between the vector and the identity label set V in the system vector library: If the identity is determined to be identified, the system further introduces an identity verification mechanism: if the user's identity is identified by the same tag in the last three identifications, the validity period of the identity in the database will be automatically extended; if identification fails twice in a row, a locking mechanism will be triggered and an alarm message will be pushed to the administrator terminal.

[0078] The behavior prediction module is used to predict behavior based on the identity vector V.id With the set of state parameters S env Predicting family members' behavioral preferences P beh With regard to the intended use of the equipment I dev And generate control command parameters C pred .

[0079] The system will use the identity vector V id With the set of environmental state parameters S env Concatenate to construct the behavior input vector:

[0080] X beh =[V id ,S env ];

[0081] Then, temporal and spatial attention mechanisms are introduced to construct weights α respectively. t With β s The behavioral preference probability vector P is calculated using the following relationship. beh :

[0082]

[0083] Wherein, time weight α t Automatically set based on time of day, for example, behavior weighting is higher in the morning than in the afternoon; spatial weight β s This is learned from user behavior data in different rooms. After generating the user's current behavioral preferences, the system combines this with intent calculation to determine device activation metrics.

[0084] I dev =V id ×β s ;

[0085] Finally, the control parameters are obtained through the control parameter generation unit:

[0086] C pred =f(P beh ,I dev );

[0087] Function f is implemented using a neural network regression function and has been deployed locally.

[0088] The strategy scheduling module is used to schedule tasks based on control command parameters C. pred Generate the optimal equipment control strategy R ctrl And adjust the device power consumption allocation factor E according to the energy optimization algorithm. opt .

[0089] First, the system is based on the control command mapping matrix M ctrl Computing device control strategy:

[0090] Rctrl =M ctrl ·C pred ;

[0091] Next, an energy consumption optimization objective function is introduced to calculate the power consumption allocation factor for each device:

[0092]

[0093] Among them, P i Let i be the rated power of the i-th device. This determines its current usage intent. The policy caching and conflict reconciliation unit uses a timestamp conflict sorting algorithm to sort the current control queue and sets a maximum state switching threshold to prevent malfunctions and energy consumption caused by repeated device switching. For example, if a smart light switches more than 5 times within 5 minutes, the system will automatically pause its subsequent control commands and record a control conflict log.

[0094] Edge computing and feedback module, used for real-time processing of device control strategies. ctrl Compared with actual equipment response data R dev Generate feedback correction coefficient F adj And optimize the parameter set S env The update frequency.

[0095] The system uses edge computing and feedback modules to verify the execution status and adaptively adjust the strategy. The system first collects the actual operating status R of the device. dev and the original strategy R ctrl Comparison error:

[0096] ε=|R dev -R ctrl |;

[0097] If the error value exceeds the set threshold, the feedback correction coefficient calculation module will compensate and correct it according to the following formula:

[0098] F adj =K t ·ε+(1-K t )·R dev ;

[0099] Initially K t =0.5, automatically adjusted to: when the error is too large.

[0100]

[0101] Subsequently, the system based on F adj Update the sampling period τ in the environment perception module s Optimize the following objective function:

[0102]

[0103] Where E(τ) is the unit energy consumption function under different sampling periods.

[0104] A home smart gateway system, firstly, the environmental sensing module starts and runs its subordinate data acquisition unit, which collects environmental data in real time through temperature and humidity sensors, light sensors, and air quality modules installed in different locations indoors, obtaining the raw environmental dataset D. env ={d1,d2,…,d n After data collection is complete, the data enters the data cleaning and anomaly removal unit, where the system processes the data according to set standards. Determine if it is an outlier, and then set the outlier data d. i To avoid bias in subsequent analysis, the cleaned data is then removed. Next, the cleaned data enters the state coding unit and is normalized using a normalization function. The data of various types are standardized and labeled according to the collection time sequence and data type to generate a structured state parameter set S. env ={s1,s2,…,s n} and temporarily store it in the local cache for later use.

[0105] Subsequently, the system enters the identity recognition module's processing stage. First, the multimodal feature acquisition unit synchronously utilizes the image acquisition camera, voice microphone, and displacement sensor to acquire image data. fac Speech waveform A voc With behavior vector trajectory B seq And convert it into a preliminary feature set {F} I ,F A ,F B The feature fusion and confidence generation unit then uses a gated residual fusion network to perform feature mapping, and calculates the fused identity vector V through the activation function σ. id =σ(W1F I +W2F A +W3F B +b). Finally, V is calculated by the identity discrimination and matching unit. id The Euclidean distance to each vector in the local identity vector library, when there exists The matching vector indicates a known identity; if two consecutive matches fail, an alarm process is automatically triggered, locking the system and notifying the administrator.

[0106] Next, the behavior prediction module is entered, where the input feature concatenation unit generates the current identification vector V. id With the set of state parameters S env Concatenate into a behavior input vector X beh =[V id ,S envThe input is fed into the behavior pattern modeling unit. This unit introduces a temporal attention weight α. t Spatial attention β s Spatiotemporal modeling of user behavior characteristics is performed to calculate the probability vector of behavioral preferences. And further based on P beh Spatial attention weight β s Intended Use of Computing Devices I dev =V id ×β s Finally, the control parameter generation unit integrates the two to calculate and output the control command parameter C. pred =f(P beh ,I dev This serves as the core decision-making basis for intelligent control of equipment.

[0107] Subsequently, the system switched to the strategy scheduling module. The regulation strategy generation unit in this module first calls the control command parameter C. pred Mapped to the preset control command matrix M ctrl Perform matrix multiplication to generate the device control strategy R. ctrl =M ctrl ·C pred Next, the energy optimization unit evaluates the power consumption of all devices to be controlled using the formula... The module calculates the total energy consumption index and optimizes the energy allocation of each device. To avoid conflicts between control commands, the strategy cache and conflict reconciliation unit in the module adjusts the execution priority of each device command through a timestamp conflict sorting algorithm and introduces a maximum state change limit mechanism. If the number of device state changes exceeds the threshold within a set time window, its subsequent control commands are suspended and a conflict log is recorded.

[0108] Finally, the system enters the edge computing and feedback module. First, the real-time feedback acquisition unit monitors the actual response data R of the device. dev The device control strategy R output in the previous stage ctrl Perform a comparison and calculate the feedback error ε=|R dev -R ctrl Next, the feedback correction coefficient generation unit calculates the feedback correction coefficient F. adj =K t ·ε+(1-K t )·R dev K t For adaptive weights, if the error is significant, the system will adjust accordingly. The correction weights are dynamically adjusted. Finally, the acquisition frequency optimization unit uses a correction coefficient F. adj Using the input variable and combining it with the energy consumption characteristic function E(τ) of each sensor, the solution is obtained. Calculate the optimal sampling period τ s The system adjusts the data acquisition frequency of the environmental perception module in real time to achieve adaptive optimization and dynamic feedback closed loop in system operation.

[0109] It should be noted that, unless otherwise specified, the embodiments and features and technical solutions in the present invention can be combined with each other.

[0110] Obviously, the embodiments described above are merely some embodiments of the present invention, not all embodiments. The accompanying drawings show preferred embodiments of the present invention, but do not limit the patent scope of the present invention. The present invention can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the patent protection scope of this invention.

Claims

1. A home intelligent gateway system, characterized by, The system comprises the following modules: An environment perception module is used to collect indoor environment data D through multiple types of sensors env , wherein the environment data D env includes temperature and humidity, illumination, and air quality, and the original data is preliminarily filtered and normalized to generate a structured state parameter set S env ; An identity recognition module is configured to identify the identity of the family member based on the multi-modal information, collect original data D id , wherein the original data D id comprise a face image, voiceprint data and a behavior sequence, and a unified identity vector V id is formed by fusing feature vectors. The behavior prediction module is used to predict behavior based on the identity vector V. id With the set of state parameters S env Predicting family members' behavioral preferences P beh With regard to the intended use of the equipment I dev And generate control command parameters C pred ; A policy scheduling module is configured to generate a control instruction parameter C based on the energy consumption data and the energy optimization algorithm pred An optimal device regulation strategy R is generated ctrl , and the device power consumption distribution factor E is adjusted according to the energy optimization algorithm opt ; Edge computing and feedback module for real-time processing of device regulation strategy R ctrl with actual device response data R dev , generating feedback correction factors F adj and optimizing the update frequency of the parameter set S env and the sampling period τ of the environment perception module s .

2. The home smart gateway system of claim 1, wherein, The environment perception module comprises: Data acquisition unit: adopt temperature and humidity sensor, photosensitive sensor, air quality module, real-time acquisition of environmental original data D env = {d1, d2,..., d n}; Data cleaning and outlier elimination unit: identify fluctuation data, when eliminate data d i , wherein d i is the i-th set of data collected by the sensor, is the average value of the collected data; State encoding unit: by a normalization function Encode the cleaned raw data into a structured set of state parameters S env = {s1, s2,..., s n} and store in local cache.

3. The home smart gateway system of claim 1, wherein, The identity recognition module comprises: Multimodal feature acquisition unit: synchronously acquiring image data I fac , speech waveform A voc and behavior vector trajectory B seq and converted into a preliminary feature set {F I ,F A ,F B} Feature fusion and confidence generation unit: a gated residual fusion network is used to construct a fusion mapping function to generate an identity vector V id : V id = σ(W1F I + W2F A + W3F B + b); Wherein, W1, W2, W3 are weight matrices, b is a bias term, and σ(·) is a normalized activation function. Identity discrimination and matching unit: calculate identity vector V id Euclidean distance with each identity label V in the known vector library, if If it is determined that the identity is recognized, otherwise, an abnormal alarm mechanism is triggered.

4. The home smart gateway system of claim 1, wherein, The behavior prediction module comprises: Input feature concatenation unit: concatenates the identity vector V id with the state parameter set S env to build the action input vector X beh = [V id , S env ]; Behavior pattern modeling unit: introduce temporal attention weight α t with spatial attention β s , output behavior preference probability vector P by the following formula beh : Wherein: X beh is a behavior input vector; X beh T is the transpose of the behavior input vector; V id - V is an identity vector V id Euclidean distance to each identity label V in the known vector library; a t is the time attention weight, if the user behavior is more representative in the morning, then a t = 1 when predicting the morning behavior, and a t = 0 when predicting the afternoon behavior. β s For spatial attention, it represents the degree of influence of the user between different spaces, reflecting the relevance in space. If the relevance is weak, β s = 0; The device usage intention I is calculated again by the following formula dev : I dev = V id × β s ; Control parameter generating unit: from preference probability vector P beh and device usage intention I dev combination, calculate control instruction parameter C pred : C pred = f(P beh , I dev ).

5. The home smart gateway system of claim 1, wherein, The policy scheduling module comprises: A regulation policy generation unit: a control instruction mapping matrix M is constructed ctrl The device regulation policy R is calculated according to the following formula ctrl : R ctrl = M ctrl · C pred ; Energy consumption optimization unit: Introduce energy minimization objective function to calculate device power consumption distribution factor E opt : wherein m is the total number of devices, is the usage intention of the i-th device, P i is the electric power of the i-th device, and is a device-specific parameter; The policy cache and conflict reconciliation unit adopts a timestamp conflict sorting algorithm to optimize the control policy queue, avoiding rapid repeated switching of device states.

6. The home smart gateway system of claim 1, wherein, The edge computing and feedback module comprises: Real-time feedback acquisition unit: acquire actual device response data R after actual execution dev with device regulation strategy R ctrl Compare, calculate feedback error ε: ε = |R dev - R ctrl |; feedback correction coefficient generating unit: estimates the feedback error to obtain a feedback correction coefficient F adj : F adj = K t · ε + (1 - K t ) · R dev ; where K t is an adaptive weight gain, initially set to 0.5, and compensated when ε is too large according to the following formula: The collection frequency optimization unit adjusts the sensor sampling period τ in the environment perception module in combination with the correction coefficient s , and meets the following optimal conditions: Wherein, E(τ) is the energy consumption per unit time under the sampling frequency, and is the inherent parameter of the multi-type sensor of the environment perception module.

7. The home smart gateway system of claim 2, wherein, In the environment perception module, the state coding unit further comprises, after the normalization processing is completed, coding the structured state parameter set S env The label identification is performed, and arrangement is performed based on the parameter collection time sequence, so as to ensure that the system identifies the latest high-weight environment parameter in priority when performing state updating.

8. The home smart gateway system of claim 3, wherein, In the identity recognition module, the identity discrimination and matching unit further comprises, when performing the identity matching operation, preferentially using the identity vectors of the last three recognitions for comparison, when the known identity appears for three times in succession, the system automatically extends the validity period of the identity label in the vector library, if the recognition fails for two times in succession, the system locking mechanism is triggered and the alarm information is sent to the administrator terminal.

9. The home smart gateway system of claim 5, wherein, In the policy scheduling module, after the policy cache and conflict reconciliation unit executes the timestamp conflict sorting algorithm, by setting a maximum state change number threshold, when the control state change number of a certain device within a set time window exceeds the number threshold, the subsequent regulation and control command execution is suspended and the regulation and control conflict log is recorded for subsequent policy optimization.

10. The home smart gateway system of any one of claims 1 to 9, wherein, During the system operation, the modules are executed in the following collaborative order: First, the indoor environment data D is acquired by the environment perception module env and a structured state parameter set S is generated env ; Then the identity recognition module constructs an identity vector V using the multi-modal data id ; from the behavior prediction module based on the state parameter set S env with the identity vector V id computing the behavior preference P beh and the device use intention I dev generating the control instruction parameter C pred ; Then the policy scheduling module schedules the control instruction parameter C pred The output device regulation policy R ctrl And optimizes the energy consumption factor E opt ; Finally, the actual device response data R is obtained by the edge computing and feedback module dev and the feedback correction coefficient F is generated adj , based on which the state parameter set S is dynamically optimized env and the sampling period τ of the environment perception module s .

Images