Digital twin virtual-real mapping and intelligent control scheduling platform for home scenario

Abstract

The application provides a digital twin virtual-real mapping and intelligent control scheduling platform for a family scene, relates to the technical field of intelligent control, and comprises a virtual-real dynamic mapping engine, a behavior causal inference engine, a virtual pre-execution scheduling mechanism and a self-evolution scheduling mechanism.The platform calculates a four-dimensional credibility field vector through sensor data, realizes malignant distortion filtering and state self-healing mapping, constructs a family behavior causal graph by using a dynamic Bayesian network and a physical information neural network, realizes short and long double-cycle forward probability risk and macro-environment evolution prediction, in a trusted execution environment, uses a space-time correlation conflict matrix and a multi-objective game model to perform sandbox simulation and pre-performance and trajectory audit interception on concurrent instructions, solves a Pareto optimal path, and finally realizes online strategy incremental reconstruction and rule automatic emergence evolution by combining user fine-tuning feedback and environment long-lag response time lag through a space-time decay reinforcement learning algorithm.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, specifically to a digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios. Background Technology

[0002] As the Internet of Things (IoT) in the home evolves towards high-density concurrency of multiple devices, deep integration of scenarios, and autonomous control, higher demands are placed on the performance, security, and long-term economic efficiency of home intelligent control and scheduling platforms. In particular, real-time and accurate synchronization, security conflict interception, and long-term adaptive control strategies in complex and dynamic environments with multiple members and intertwined tasks have become the focus of attention in the smart home and edge control industries. For example, smart elderly care scenarios require highly reliable multimodal behavior perception to achieve proactive defense against sudden risks to the elderly; multi-device collaborative humidity control scenarios require precise coordination of multiple devices with zero energy consumption offsetting; and green and low-carbon homes require a control center to mitigate long-tail energy waste from a monthly overall perspective. Currently, IoT digital twin technology has begun to be gradually applied to the digital modeling and basic status display of smart homes.

[0003] However, existing smart home and IoT control and scheduling platform technologies face the following key technical challenges in fulfilling these requirements: 1. Insufficient robustness of virtual-real mapping alignment and hardware dirty data filtering capabilities: Traditional smart home digital mapping often relies on threshold triggering of a single sensor, lacking multi-dimensional correlation review of data. For example, temperature and humidity sensors based on traditional wireless communication are prone to reporting transient and abrupt dirty data when encountering network jitter or component drift. Due to the lack of cross-verification and credibility calculation in dimensions such as time, data, behavior, and environment, the digital twin space is directly missynchronized, which leads to frequent mis-triggering and control oscillation of subsequent control strategies. Moreover, the existing system lacks the ability to roll back anomalies and self-heal its state, making it difficult to ensure the high credibility of the mapping layer.

[0004] 2. Lack of causal spatiotemporal depth and multi-device collaborative auditing in behavioral risk prediction: Existing platform linkage control is mostly based on isolated, post-feedback "if-then" static rules, lacking a deep causal understanding of multi-member behavior and forward-looking sandbox exercises. For example, traditional systems cannot predict secondary high-risk events such as kitchen dry burning caused by discrete forgetting behavior of members by probabilistic propagation along spatiotemporal movement within short cycles of minutes. At the same time, due to the lack of modeling of continuous physical laws such as thermal inertia of building walls, the system cannot predict long-term environmental degradation and equipment fatigue trajectory within long cycles of days or months. Furthermore, it cannot perform spatiotemporal conflict detection and multi-objective game optimization for concurrent instructions from multiple devices before physical execution, which can easily lead to hidden dangers such as hot and cold control offsetting or circuit breaker tripping caused by multiple high-power electrical appliances.

[0005] 3. Lack of online generalization and lifelong self-evolutionary learning mechanisms for control strategies: Existing smart home scheduling strategies mostly rely on factory presets or tedious manual programming, which are extremely unsuitable for adapting to the personalized characteristics of families and the drift of long-term habits. For example, traditional control platforms cannot quantitatively extract the inherent long-term environmental response delay of a specific residential building, resulting in algorithmic mismatch between control signal issuance and environmental feedback. Furthermore, they lack feedback stream monitoring and non-linear penalty mechanisms for users' daily proactive fine-tuning behavior, and cannot dynamically transform users' implicit dissatisfaction and long-tail preferences into the driving force for upgrading the strategy library. This results in a rigid rule system, which cannot achieve autonomous incremental evolution and lossless pruning and merging of collaborative scheduling strategies, thus limiting the applicability of the platform for long-term refined control in different families and across seasons.

[0006] To address the aforementioned issues, there is an urgent need for a digital twin-based home self-evolving intelligent collaborative control and scheduling platform and method to achieve highly reliable mapping and self-healing of physical data, causal prospective extrapolation of short- and long-term dual-cycle risks, spatiotemporal sandbox compliance auditing of concurrent instructions, and lifelong online self-evolving rule emergence of control strategies. Summary of the Invention

[0007] Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides a digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios, solving the problems mentioned in the background section.

[0009] Technical solution To achieve the above objectives, this invention provides the following technical solution: a digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios, wherein the scheduling platform includes a virtual-real dynamic mapping engine, a behavior causal inference engine, a virtual pre-execution scheduling mechanism, and a self-evolving scheduling mechanism, wherein: The virtual-real dynamic mapping engine is used to collect device status, environmental data and multimodal behavior data in the home physical space. By dynamically constructing a high-dimensional virtual-real credibility field composed of time credibility, data credibility, behavior credibility and environmental credibility for each device, it evaluates the virtual-real deviation value between the physical space and the digital twin space in real time, and triggers a self-healing control mechanism when the virtual-real deviation value exceeds a preset threshold, so as to achieve high-credibility dynamic alignment from the physical space to the twin space. The behavioral causal inference engine is used to construct a time-sliding family behavioral causal graph based on family historical behavioral sequences, dynamic environmental changes, multi-member interaction information, and device status correlation. It also uses the causal graph to predict the causal inference of family future short-term events, and combines it with long-term future state evolution schemes to infer macro trends. The virtual pre-execution scheduling mechanism is used to intercept control commands and project them into the digital twin space to perform virtual control simulation before the control commands are sent to the physical space. It makes full predictions on multi-device conflicts, user comfort, and system energy consumption during the simulation process, and determines whether to release the control commands based on the prediction results. The self-evolving scheduling mechanism is used to monitor the running data in the virtual-real dynamic mapping engine for a long time, dynamically learn user living habits, time patterns, scene preferences and energy consumption characteristics, and automatically generate new device collaborative scheduling and control strategies online.

[0010] Preferably, the intelligent control and scheduling method of the scheduling platform includes the following steps: Sp1: Constructs a high-dimensional virtual-real credibility field for home devices through a virtual-real dynamic mapping engine, dynamically calculates virtual-real deviation values, and ensures high credibility alignment between the digital twin space and the physical space; Sp2: Utilizes a behavioral causal inference engine to construct a family behavioral causal graph, performs short-term forward probability propagation prediction based on current family dynamics, and combines long-term future state evolution schemes to predict long-term macro-environmental evolution and equipment fatigue characteristics. Sp3: When a control command to be executed is generated, a virtual pre-execution scheduling mechanism is used to perform simulation pre-run, conflict detection, comfort and energy consumption prediction in the digital twin space. The control command is only sent to the real equipment in the physical space for execution when the virtual simulation results meet the target. Sp4: Through a self-evolving scheduling mechanism, it dynamically learns family characteristics and habits, and continuously generates and iterates online device collaborative scheduling control strategies.

[0011] Preferably, the virtual-real dynamic mapping engine in Sp1 calculates the virtual-real deviation value by including the following steps: Sp1.1: Calculate time reliability based on the difference between the device's historical reporting cycle and the current timestamp; calculate data reliability based on the integrity of the data packet and the noise range; calculate behavior reliability based on whether the device's current action conforms to the device's preset operation logic chain; and calculate environmental reliability based on the data collaborative verification results of associated environmental sensors. Sp1.2: Interweave time credibility, data credibility, behavior credibility and environmental credibility in multiple dimensions, map them into the virtual and real credibility field of each device in the current time and space, and compare the predicted state of the twin space with the actual reported state of the physical space to calculate the virtual and real deviation value that reflects the degree of virtual and real alignment. Sp1.3: Determine whether the virtual-to-real deviation value exceeds the preset confidence threshold. If it does, determine that the physical reported data or twin model is distorted and automatically trigger the self-healing control mechanism.

[0012] Preferably, the self-healing control mechanism includes: The resampling unit is used to increase the physical data sampling frequency of the corresponding faulty device and perform multi-channel retransmission. The state correction unit is used to perform weighted offset correction on the distorted state of the abnormal device using observation data from associated sensors located in the same physical area as the abnormal device. The edge reconstruction unit is used to reinitialize the twin node of the device at the edge side by calling the underlying initial physical behavior model of the device from the local gateway when the correction fails. An abnormal rollback unit is used to control the state of the device in the twin space to roll back to the last historical state that was determined to be trustworthy during reconstruction.

[0013] Preferably, the behavioral causal inference engine in Sp2 constructs a family behavioral causal map by including the following steps: Sp2.1: Using time as the axis, extract continuous actions of family members and match the spatial location, environmental index fluctuations, and response status of surrounding devices when the actions occur as causal events; Sp2.2: Establish directed association edges between different causal events, where the direction of the association edge is determined by the chronological order, and the weight of the association edge is determined by the frequency of historical occurrence and the depth of environmental coupling, thereby generating a causal graph of family behavior. Sp2.3: Based on the causal meta-events triggered at the current moment, forward probability propagation is carried out along the directed correlation edges of the family behavior causal graph to predict abnormal high-risk state trends or high energy consumption tendencies in the future short-term period.

[0014] Preferably, the abnormal high-risk state trend or high energy consumption tendency scenario in the future time period includes: the trend of increased fall risk value due to the behavioral event of the elderly frequently getting up at night; the trend of increased kitchen danger probability due to the behavioral sequence of continuous rise in kitchen ambient temperature, gas valve opening and long-term stillness of people in the space; and the trend of energy waste due to the behavioral sequence of long-term no movement of people in the family space and the air conditioning equipment being in operation.

[0015] Preferably, the behavioral causal inference engine in Sp2 includes the following steps when performing long-term future state evolution: Sp2.4: Building upon the results of short-cycle simulations, the fluctuation trend of environmental indicators within a short-cycle period is used as the initial boundary condition. In the digital twin space, a model of long-cycle thermal inertia decay parameters of buildings and cumulative loss of equipment operation is introduced. Sp2.5: Performs accelerated time-flow simulation evolution in a digital twin space to simulate the evolution of macroscopic environmental conditions, long-term household electricity load accumulation, and equipment performance degradation trajectory over a long period of time. Sp2.6: Assess whether the evolution of the macro-environmental state triggers the long-term environmental degradation threshold, or whether the performance degradation trajectory reaches the equipment fatigue limit ahead of schedule, and feed the assessment results back to the self-evolutionary scheduling mechanism in real time as a constraint boundary.

[0016] Preferably, the virtual pre-execution scheduling mechanism in Sp3 for virtual control simulation includes the following steps: SP3.1: Intercept sets of control instructions to be executed from user-triggered, causal inference-triggered, or self-evolutionary strategies; SP3.2: Input the set of control instructions to be executed into the current digital twin space, simulate execution in the shadow physics engine, calculate the physical interference conflicts of multiple devices in a concurrent state, and simulate and predict the changes in physical environment parameters after execution; Sp3.3: Input the predicted physical environment parameters into the preset user comfort model and energy consumption assessment model, and determine whether the simulation results simultaneously meet the environmental comfort scale and energy consumption budget target; SP3.4: If the judgment result is satisfied and there is no physical interference conflict, the control interception barrier is opened, and the control instruction set to be executed is converted into a real electrical signal and sent to the real device in the physical space. If the judgment result is not satisfied or there is a physical interference conflict, the control instruction is terminated and the simulation conflict reason is fed back to the scheduling platform.

[0017] Preferably, the self-evolutionary scheduling mechanism in Sp4 dynamically learns family characteristics and habits, including the following steps: SP4.1: Continuously collect environmental feedback time series tables after the execution of control scheduling in the home physical space, user's active fine-tuning behavior of the control results, and electricity load curves over long periods; Sp4.2: Extract users' personalized environmental adjustment long-tail preferences in different seasons and at different times, and combine them with the evaluation results of the long-term future state evolution to identify the hysteresis response characteristics of the physical space environment to device control. SP4.3: Based on extracted preferences, features, and long-term constraint boundaries, dynamically update the associated weight parameters in the control strategy library, or adaptively merge the original control conditions to automatically generate an adaptive collaborative scheduling strategy to replace the fixed linkage rules.

[0018] Beneficial effects This invention provides a digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios. It has the following beneficial effects: 1. This invention breaks away from the traditional blind mode triggered by a single sensor threshold by setting up a virtual-real dynamic mapping engine. It uses a multi-source heterogeneous information fusion model to calculate the four-dimensional credibility of the reported data in terms of time, data, behavior, and environment. By projecting the high-dimensional virtual-real credibility field vector into space, it can effectively identify and amplify the malicious distortion caused by hardware failures by using nonlinear penalty effects. With the state correction and anomaly rollback mechanism, the platform system can refuse dirty data synchronization at the edge and achieve weighted hedging and self-healing of the state. This ensures high credibility alignment between the digital twin shadow space and the physical world, and effectively prevents subsequent policy mis-triggering and control oscillation caused by sensor component drift or channel jitter.

[0019] 2. This invention upgrades traditional post-feedback control to proactive predictive control by setting up a behavioral causal inference engine and a virtual pre-execution scheduling mechanism. On the one hand, the system can use a dynamic causal Bayesian network model to conduct short-cycle forward probability propagation along the graph behavioral causal chain, and detect secondary high-risk hazards caused by discrete behaviors before danger occurs. On the other hand, it can use physical information neural networks to accelerate the time flow and quantitatively predict long-cycle macro-environmental degradation and equipment fatigue trajectories several days to several months later. By introducing a spatiotemporal correlation conflict matrix and a multi-objective game optimization utility function model in a trusted execution environment, it realizes a sandbox pre-run before the physical execution of instructions, outputs continuous simulation evolution curves, forcibly intercepts strong conflict instructions that violate ergonomic red lines and have the risk of overload tripping, smooths out peak energy consumption at the Pareto optimal frontier, and ensures the absolute reliability and optimality of the control system.

[0020] 3. This invention, by setting a self-evolving scheduling mechanism, endows the control center with the vitality of self-reproduction and incremental strategy upgrades. The platform system uses a monitoring module to strongly align long-term environmental response, power load, and users' daily proactive fine-tuning behavior on the hardware clock source. It also uses a long-delay correlation function model to accurately calculate the unique building thermal inertia time delay constant of the residence, overcoming the algorithmic mismatch between control signals and physical environment feedback. By running iterative equations containing a spatiotemporal decay incremental reinforcement learning model on a continuous time axis, the dynamic reward function can sensitively capture users' implicit dissatisfaction using a time decay factor. Nonlinear penalties force the matrix parameters to perform self-incremental reconstruction and lossless rule pruning and merging, thereby automatically generating a long-tail adaptive collaborative scheduling strategy that best fits the specific family life norms online, achieving the lifelong self-evolution effect of the platform system. Attached Figure Description

[0021] Figure 1 This is a platform system architecture diagram of the present invention; Figure 2 This is a flowchart of the method of the present invention; Figure 3 This is a display of the platform's main interface for the present invention. Figure 1 ; Figure 4 This is a display of the platform's main interface for the present invention. Figure 2 ; Figure 5 This is a display of the platform's main interface for the present invention. Figure 3 ; Figure 6 This is a display of the platform's main interface for the present invention. Figure 4 . Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention. Specific Implementation Example 1: Please see Figure 1 and Figure 2 As shown, a digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios includes a virtual-real dynamic mapping engine, a behavior causal inference engine, a virtual pre-execution scheduling mechanism, and a self-evolving scheduling mechanism, among which: The virtual-real dynamic mapping engine is used to collect device status, environmental data, and multimodal behavioral data in the home physical space. By dynamically constructing a high-dimensional virtual-real credibility field composed of time credibility, data credibility, behavioral credibility, and environmental credibility for each device, it evaluates the virtual-real deviation value between the physical space and the digital twin space in real time. When the virtual-real deviation value exceeds a preset threshold, a self-healing control mechanism is triggered to achieve high-credibility dynamic alignment from the physical space to the twin space. This engine is permanently deployed in the smart gateway at the edge of the home and is responsible for transforming fragmented and heterogeneous raw data in the physical space into highly cohesive and highly reliable state objects in the digital twin space. The engine is organically composed of the following four functional modules: Multimodal data ubiquitous acquisition module: responsible for accessing IoT devices in the home (such as smart air conditioners, gas valves, wristbands, etc.) through protocols such as Matter, ZigBee, and Wi-Fi, and integrating non-invasive sensors (such as millimeter-wave radar and infrared arrays) to acquire multimodal data such as high-frequency device status, spatiotemporal environmental indicators, and human body micro-motion Doppler frequency shift; Field Space Generation Module: Responsible for allocating a dynamic matrix space in memory and establishing a four-dimensional (time, data, behavior, environment) evaluation mechanism that does not blindly trust the original reported data, namely the virtual and real credibility field; Spatial Deviation Calculation Module: Responsible for comparing the physically reported state vector with the predicted state vector of the digital twin space in real time, dynamically calculating the Euclidean distance or cosine similarity between the two, and thus outputting the virtual-real deviation value; Dynamic closed-loop self-healing module: It includes four underlying units: resampling, state correction, edge reconstruction and anomaly rollback, which are used to perform self-healing correction on the physical side or the shadow side when data is distorted.

[0024] In home-based digital twin scenarios, data is highly susceptible to uncontrollable factors such as network jitter, data drift caused by sensor dirt, or malicious tampering by hackers. If the digital twin central system blindly trusts and directly adopts the raw values ​​reported by the physical layer, these errors will be amplified by the system, causing the subsequent behavior inference engine to run on an incorrect benchmark and thus issue destructive control commands. The core function of this engine is to overcome the fatal flaw of blindly trusting reported data in traditional digital twins. By setting up a trust filter between the physical side and the twin side, dirty data and logically contradictory states are eliminated during the synchronization phase, ensuring that the digital twin space is an absolutely highly reliable mirror of the physical world, thereby providing a 100% reliable decision benchmark for subsequent behavior inference and interception scheduling.

[0025] Furthermore, the specific control flow of this engine during runtime is described in detail below: SP1: When this platform is started, the virtual-real dynamic mapping engine in the smart edge gateway establishes a resident daemon process at the bottom layer. The gateway polls at a preset frequency or asynchronously listens for various electrical signal messages reported by physical space devices through an interrupt mechanism. This step serves as the starting point for the entire control and scheduling process. Its purpose is to eliminate the spatiotemporal uncertainty in the Internet of Things environment by continuously cleaning and reconstructing the state of the physical world.

[0026] To quantitatively assess the physical reliability of reported data within the main control process, the virtual-real dynamic mapping engine cannot directly synchronize the original values. It must first perform fine-grained deconstruction of the data's spatiotemporal context attributes. Therefore, the system automatically calls down to SP1.1: when the gateway receives a specific device... exist Status data reported in real time Subsequently, the data stream immediately enters the multi-dimensional parallel computing pipeline. The virtual-real dynamic mapping engine, based on a multi-source heterogeneous information fusion model, simultaneously performs credibility index calculations across four dimensions: time, data, behavior, and environment, in four hardware threads. Time credibility Solution: The virtual-real dynamic mapping engine reads the arrival timestamp of the packet network protocol stack. And unpack the data to extract the raw hardware timestamps generated at the physical layer. The degree of expiration due to network congestion is calculated using the following formula: ; in, This is the network decay constant. The longer the delay, the more exponentially the time reliability decreases, thus avoiding the accumulation of excessive control commands.

[0027] Data credibility Solution: To remove dirty data that exceeds hardware limits or causes sudden high-frequency noise, the virtual-real dynamic mapping engine calls the device physical range pre-stored in the gateway cache. It also combines historical sliding windows to calculate the first derivative (rate of change) of the current data, and its formula is expressed as: ; in, This is an indicator function; when the data exceeds the physical range, the function value immediately returns to zero. This serves as a mutation penalty factor, thereby constraining the continuity of the data.

[0028] Behavioral credibility Solution: The virtual-real dynamic mapping engine inputs the current action command into the preset Markov state transition probability matrix. In the process, the evaluation determines whether the action conforms to the preceding and following logic chains of the hardware operation (if the device suddenly reports a "temperature adjustment" status without a "power on" signal, it is considered illogical). The calculation formula is as follows: ; Environmental credibility Solution: To prevent false judgments caused by physical drift of a single sensor, the virtual-real dynamic mapping engine searches the set of neighboring devices in the same room as the current device through the home LAN topology. (Number of neighbors is) The calculation formula is as follows: This utilizes spatially adjacent sensors to cross-verify physical laws. ; in, The environmental coupling coefficient is such that if the temperature controller reports a high temperature but the surrounding infrared and air sensors show normal readings, this component adaptively approximates the zero value through the hyperbolic tangent function.

[0029] Single-dimensional verification cannot identify advanced tampering or deep physical drift. By using the four-dimensional decomposition of the multi-source heterogeneous information fusion model mentioned above, scattered state signals are quantified into multi-dimensional confidence indicators, which can lock down the deception space of abnormal data from all aspects of time and space.

[0030] After independently obtaining the credibility scalar components of time, data, behavior, and environment, these scattered indicators cannot be directly used to assess the global deviation. Therefore, the system enters Sp1.2, which uses matrix interleaving to upgrade these indicators into a unified field: the virtual-real dynamic mapping engine performs multi-dimensional interleaving of the four scalar components calculated in parallel in Sp1.1, dynamically constructing a four-dimensional manifold space field covering the device in memory, i.e., a high-dimensional virtual-real credibility field vector. ; Subsequently, the spatial deviation calculation module retrieves the theoretical predicted state of the entity from the memory cache of the digital twin space, which was predicted in the previous moment based on the shadow physics engine (such as thermodynamic diffusion model, airflow field model, and device power consumption behavior model). To achieve confidence-based bias calibration, the virtual-real dynamic mapping engine uses the virtual-real confidence field vector to spatially project the difference between the actual reported state and the twin predicted state. Then, using the comprehensive weighted Euclidean distance formula below, it calculates the virtual-real bias value, which reflects the degree of virtual-real alignment. : ; in, For the pre-defined weight bias vectors of the four dimensions, Represents the normalized difference in the physical state space dimension.

[0031] This step, by introducing a high-dimensional virtual-real confidence field vector as the projection matrix, enables intelligent identification and decoupling filtering of "benign deviations" caused by occasional abrupt changes in physical space and "malicious distortions" caused by hardware physical failures. Specifically, when a home physical space encounters a real environmental abrupt change (a sudden fire causing a sudden increase in temperature), although there is a huge absolute difference between the actual reported state value of the physical space and the theoretical predicted state value of the digital twin space, the environmental confidence component verified by neighboring devices and the behavioral confidence component constrained by the state transition matrix both exhibit extremely high confidence. After smoothing the manifold space projection using the aforementioned comprehensive weighted deviation formula, the calculated... The virtual-to-real deviation value will still be effectively constrained within the preset security trust threshold, thereby driving the system to determine that the data belongs to a real physical mutation, choose to trust the actual reported state and update the digital twin space in real time. Conversely, if there are no abnormal fluctuations in the surrounding environment, and only a single sensor experiences a surge in data due to its own component drift or hardware short circuit, since its environmental credibility component and behavioral credibility component are close to zero, the high-dimensional virtual-to-real credibility field will produce a nonlinear penalty effect when performing matrix projection, which will directionally amplify the difference between the actual reported state value and the theoretical predicted state value, causing the final calculated virtual-to-real deviation value to instantly break through the threshold limit, thereby accurately and without error locking the hardware data distortion.

[0032] With Sp1.2 calculating the quantized virtual-real deviation value online... The control system needs an automated decision-making logic to determine whether intervention is required. Therefore, the process enters Sp1.3: the virtual-real dynamic mapping engine calculates the virtual bias value obtained from Sp1.2. The data is transmitted to the hardware comparator inside the gateway and compared with the system's preset trusted security threshold. Perform real-time comparison; if the comparison result shows... If the current virtual-physical mapping is determined to be in a highly reliable alignment state, the gateway directly releases the data pipeline and updates the digital twin space with physical reported data; otherwise, once... The comparator immediately sends a high-priority interrupt signal to the system kernel, determining that the physically reported data or the digital twin model has suffered severe distortion. It then automatically triggers a self-healing control mechanism to intercept untrusted data streams. Based on the credibility field component that caused the specific out-of-target point, it selectively invokes subsequent self-healing control mechanisms to perform spatial self-healing. This is a crucial hub for realizing an unattended, self-healing smart home control center. It can automatically intercept and warn before dirty data contaminates the digital twin shadow space, causing a complete collapse of the subsequent causal control chain, thus providing a robust and robust security barrier for the platform.

[0033] When Sp1.3 determines that the virtual-to-real deviation value exceeds the limit and triggers the self-healing control mechanism, this platform performs three-dimensional repair of the distortion state through the precise coordination of the following four units: Resampling unit: If Sp1.1 decision system time reliability A sudden drop (large network packet loss and latency, large message delay) or data reliability If an occasional spike causes a exceedance, the resampling unit is immediately activated. The gateway sends a reverse control command to the physical device through a dedicated hardware control channel, instantly increasing its physical sampling frequency to 10 times the original frequency and enabling multi-channel redundant retransmission. Through high-frequency resampling combined with median filtering, occasional channel white noise is quickly filtered out at the edge, and the time and data reliability are restored within 500 milliseconds, allowing the system to return to normal mapping. State correction unit: If the deviation still exceeds the standard after high-frequency resampling, and the judgment system has low environmental reliability. If there are deficiencies (such as the living room thermostat continuously reporting 50°C, which is manifested as physical temperature drift attenuation of the sensor), the state correction unit is activated. It automatically retrieves the observation values ​​of the auxiliary thermistors integrated inside the smart TV and ambient light sensor in the same area, calculates the weighted average state of these neighboring devices as a physical constraint, and performs weighted offset correction on the shadow state of the distorted thermostat (such as forcibly smoothing and correcting the shadow state of the distorted thermostat to 26°C) to prevent the digital twin space from undergoing abrupt changes due to the failure of a single hardware component. Edge reconstruction unit: If the behavior is reliable Zeroing out occurs when a device reports an illegal logical action that is not supported by the underlying hardware (such as a single-cooling air conditioner reporting that it is heating or a network topology link being physically offline due to a break). This causes the aforementioned state correction to lose its physical reference. The edge reconstruction unit is immediately activated, and the gateway directly cuts off the untrusted mapping stream of the node. It calls the device's factory-preset underlying initial physical behavior model and standard state space transition matrix from the read-only memory (ROM) and clones and initializes a brand-new digital twin shadow node in the edge memory to clear the logical disorder caused by the firmware error. Anomaly Rollback Unit: During the time gap (usually 1 to 3 seconds) when the edge reconstruction unit initializes and reconstructs shadow nodes, in order to prevent the behavioral causal inference engine from reading zero values ​​or floating states, the anomaly rollback unit runs synchronously. It immediately suspends the current real-time alignment stream, retrieves the historical standard state safety point that has been determined to be trustworthy from the historical logs of the local lightweight database, forces the shadow node state in the current twin space to roll back and freeze to the safety point, provides a legal transition benchmark for the system, and ensures the stability and continuity of global control scheduling.

[0034] The behavioral causal inference engine is used to construct a time-sliding causal graph of family behavior based on historical family behavior sequences, dynamic environmental changes, multi-member interaction information, and device status correlations. This causal graph is then used to predict short-term events in the family's future, while simultaneously incorporating long-term future state evolution schemes for macro-trend inference. Deployed in the core computing unit of the home gateway or a high-performance edge server, the engine's core task is to break down the technical barriers of isolated interaction between traditional smart home devices, endowing the system with a deep causal understanding of the behavior of multiple family members, multiple tasks, and across time and space. The engine's functional architecture consists of the following four core functional modules working together organically: Multimodal event feature extraction module: used to perform time window slicing on heterogeneous device status, physical environment indicators and human behavior trajectories synchronized from the virtual and real dynamic mapping engine and aligned with high reliability, to extract causal meta-events with clear physical meaning; Topology Graph Dynamic Weaving Module: Used to build and maintain an online causal graph of family behavior that slides over a time window, responsible for decoupling and establishing directed association edges and weight distributions between meta-events; Short-cycle probability propagation inference module: Built-in hidden Markov chain and dynamic Bayesian network topology, responsible for forward probability propagation along the directed correlation edges of the graph, used to make advance predictions of sudden personal safety risks (such as elderly people falling, kitchen gas leaks) and instantaneous energy consumption within minutes or hours. Long-cycle time flow evolution module: It has built-in long short-term memory physical information neural network (PINN) and building physical inertia decay model. By accelerating the time flow in virtual shadow space, it is used to deduce macro trends at the future day or month level (such as equipment fatigue decay trajectory, long-cycle environmental degradation trend).

[0035] The core function of this engine is that it can not only understand the present state of the physical world, but also use causal chains to predict the future state trends of the physical world that are high-risk and high-energy-consumption, thereby providing forward-looking control inputs and interception boundaries for subsequent virtual pre-execution scheduling mechanisms.

[0036] Furthermore, the specific process of this engine during runtime is described in detail below: Sp2: During the operation of this platform, after the virtual-real dynamic mapping engine completes the highly reliable dynamic alignment from the physical space to the twin space, the output clean data stream is triggered in real time and flows into the behavior causal inference engine. This step, as the core of global prediction, is coordinated and scheduled by the multi-core processor of the gateway, and the short-cycle behavior network and the long-cycle physical evolution sandbox are opened in parallel. Traditional smart home linkage only relies on the static threshold of a single device and lacks a deep causal understanding of multi-member interaction and behavior over time. Through the main control process of this step, the scattered device states can be transformed into accurate dual-cycle inference of human high-dimensional intentions and potential risks, providing intelligent decision-making inputs that far exceed human a priori rules for subsequent virtual pre-execution and self-evolution scheduling.

[0037] In order to establish causal relationships in the main inference process, the system cannot directly use disordered data frames. It needs to align and assemble multimodal data on the spatiotemporal axis first, and condense it into discrete physical causal events. Therefore, the behavioral causal inference engine enters Sp2.1: The engine uses a sliding time window as the axis to extract the continuous actions of family members (such as human displacement trajectories extracted by millimeter-wave radar and physiological signs extracted by wristbands). At the same time, the behavioral causal inference engine automatically retrieves the physical spatial location when the action occurs, the fluctuations of environmental indicators recorded by surrounding sensors (such as local air temperature and instantaneous gas concentration), and the response status of surrounding IoT devices (such as the status of the solenoid valve of the stove). The multimodal event feature extraction module performs cross-modal association alignment and binding of the above four, and packages and encapsulates them into an independent causal meta-event with physical contextual meaning. If the multimodal data is not encapsulated into events, the system will only receive boring disordered numbers such as "temperature 30 degrees" and "switch on", and will not be able to understand what kind of human behavior motivation caused these numbers, let alone form logical causal antecedents.

[0038] After continuously decoupling and extracting hundreds or thousands of causal events from the multimodal data stream, these events remain isolated in memory. To find the inherent coupling patterns, the process progresses to Sp2.2: the dynamic topology graph weaving module uses the different causal events extracted in Sp2.1 as network nodes. Based on the chronological order of occurrence, the module establishes directed edges from the nodes of earlier events to the nodes of later events. Simultaneously, the module searches the historical database, counts the co-occurrence frequency of these two events over a long historical period, and extracts the coupling depth of environmental indicators at the time of occurrence. These two are then fused and transformed into the weight of the directed edge. Finally, a time-sliding causal graph of family behavior that reflects the unique living norms of the family is woven in the gateway's dynamic random access memory (DRAM). Within a certain sliding time window, the family behavior causal graph is defined as a spatiotemporal multimodal topology network. , where nodes Directed edges representing different causal events (such as "the elderly get up" or "the gas is turned on") Representative event Pointing to events Because family environments are highly personalized and private, uniform factory rules cannot be applied to all families. By weaving causal graphs online, the implicit living habits and physical coupling of a specific family can be transformed into a digital and concrete graph topology, providing a solid structural foundation for subsequent probabilistic reasoning.

[0039] As Sp2.2 constructs a highly cohesive causal graph of family behavior in memory in real time, the system needs to use this graph to perform real-time security checks on currently occurring behaviors. Therefore, the system enters Sp2.3: when a sudden behavior in the physical space triggers the current set of causal meta-events. Subsequently, the short-cycle probability propagation inference module immediately uses this set as the original input and injects it into the family behavior causal graph. At this time, the behavior causal inference engine starts the built-in Dynamic Causal Bayesian Network (DCBN) model. Using each meta-event node as the topological reference, along the topological direction of the directed association edges, it uses the following conditional probability forward iterative propagation formula to solve for various target high-risk state events in the future short-cycle period (set to the next 15 minutes to 2 hours). probability of occurrence

[0040] ; When processing this formula, the data input not only includes the state of the currently active meta-event, but also deeply integrates the comprehensive confidence scalar of the high-dimensional virtual-real confidence field vectors of each corresponding device output in real time by the aforementioned virtual-real dynamic mapping engine. The model performs conditional probability product and summation on each directed causal chain in the hardware thread. Finally, this step outputs a technical feature data stream containing probability percentages to the system kernel. This data stream characterizes short-term trends such as increased risk of falls among the elderly, increased kitchen hazards, or long-tail energy waste. The significance of this step lies in identifying trends before hazards or waste occur. Through the conditional probability propagation of the aforementioned dynamic causal Bayesian network model, it overcomes the drawbacks of post-feedback in smart homes. More importantly, by using the confidence scalar... Introducing the formula as a multiplication operator makes the probability of causal inference highly dependent on the authenticity of hardware data: if the reliability of the input sensor data is extremely low, the final probability output of its causal transmission will automatically decay, thereby eliminating the chain control false alarms caused by dirty sensor data at the algorithm level.

[0041] Although the dynamic causal Bayesian network model successfully captured discrete behavioral risks within short periods, there are still slow-evolving continuous physical phenomena in the home environment, such as heat accumulation in walls and long-term wear and tear of electrical components. These cannot be quantitatively simulated by discrete event probabilities alone. Therefore, the behavioral causal inference engine decided to extend the process to a longer period, and the process entered Sp2.4: The long-period time-flow evolution module automatically takes over the short-period environmental indicator fluctuation trends output by Sp2.3 (such as the indoor heat diffusion rate data caused by the current high-power operation of the air conditioner) and uses it as the real-time data input and initial time starting point for the algorithm evolution. The boundary conditions are determined, and simultaneously, the module retrieves pre-trained long-cycle thermal inertia decay parameters for that specific apartment type from the gateway's non-volatile memory. (Used to quantify the heat absorption and storage characteristics constants of the wall over a long period) and the corresponding cumulative wear model of household appliances (such as the fatigue wear curve constant of the compressor due to long-term thermal alternation), these physical constants, together with real-time environmental trend data, are used as mathematical constraint boundaries and loaded into the input layer of the subsequent long-term physical evolution neural network model. The long-term evolution is not an arbitrary numerical extrapolation; it must take the current real physical state as the starting point of evolution (input data). At the same time, it is necessary to configure real physical law parameters (such as thermal inertia decay parameters and material fatigue limit constants) at the input end for rigid constraints. Only in this way can the mathematical boundaries be locked for the next step of performing physically realistic time-fast forward simulation in the software shadow space, preventing the pure data model from causing algorithm divergence or logical distortion.

[0042] After successfully locking in the physical constraints and initial boundary conditions for long-term evolution, in order to preview the continuous macroscopic evolution results of the physical world several days or even months in advance in the digital world, the process progressed to Sp2.5: the evolution module, within an independent hardware sandbox in the digital twin shadow space, introduced a time acceleration factor. (satisfy The clock flow in the shadow space is artificially accelerated. Inside the sandbox, multi-threaded processors run a Long Short-Term Memory Physical Information Neural Network (PINN) and a Weibull cumulative loss model in parallel. The network receives initial boundary data established by Sp2.4, running on a continuous time axis. The following physical information constraint equations reflecting the law of conservation of energy and mathematical loss equations reflecting the aging of electrical hardware are solved using high-speed iterative processing: ; ; in, These are indicators of the macro-environmental status over a long-term future period (such as wall moisture content and indoor cumulative heat load). For the fatigue degradation trajectory of equipment (such as air conditioning compressors), The cumulative heat flow released by indoor equipment. The heat flow that is dispersed outward through the building walls. For the long-term thermal inertia decay parameters of buildings; For the long-term operating load rate of the equipment, This is the time decay coefficient in the cumulative loss model of equipment operation. and Given the inherent fatigue dimensions and shape parameters of the equipment, the model will incorporate the theoretical cumulative heat flux released by the indoor equipment during future acceleration periods in the data processing phase. Heat flow that is dispersed outward through the building walls By performing continuous integration, macroeconomic environmental state indicators for future long-term periods can be calculated and output online. (e.g., wall moisture content, indoor cumulative heat load variation curve), simultaneously, the model will include the equipment's long-term expected operating load rate in the future. Combined with time decay coefficient Integrating is performed using the inherent fatigue scale parameters. and shape parameters An exponential mapping is performed, and the fatigue decay trajectory of the core equipment is ultimately calculated and output. This allows the simulation, within the software shadow space, of the evolution trend of the household's macro-environment over a long period (such as the next 30 days or even a quarter), the cumulative amount of household electricity load over a long period, and the performance degradation trajectory of key components within each core device. For example, the safety hazards of aging circuit breakers caused by long-term high-load electricity use, or the hidden risk of mold growth due to the thermal inertia of walls during the rainy season, cannot be manifested in a short period of minutes. Through this step, the physical information neural network is used to accelerate time in the twin space, achieving a perfect integration of data-driven and physical prior laws, thus enabling the system to predict physical evolution in the distant future.

[0043] As Sp2.5 successfully accelerated the evolution of long-term macro-environment and equipment degradation trajectories in the shadow sandbox, this long-term predictive data needs to be transformed into the driving force for upgrading the current control and scheduling strategy. To this end, the process enters Sp2.6: the behavioral causal inference engine transmits the long-term macro-environment curves and performance degradation trajectories output by Sp2.5 to the system's compliance assessment unit. The unit automatically determines whether the macro-environmental state will trigger long-term environmental degradation thresholds (such as excessive humidity causing long-term mold growth), or whether the equipment performance degradation trajectory indicates that its internal components will reach the equipment fatigue limit prematurely. The assessment unit then identifies these exceeding thresholds. The timestamps and key load parameters of the points are extracted and transformed into specific long-term physical constraint boundaries. Then, through the data bus inside the gateway, these constraint boundaries are fed back in real time and injected into the self-evolving scheduling mechanism, driving it to reconstruct the long-term weights in the multi-device collaborative control scheduling strategy online. This step realizes the proactive feedback of the future to the current control, so that the entire intelligent control scheduling platform can not only provide immediate comfort, but also start from the macro-level situation of extending the service life of home appliances, preventing long-term environmental degradation, and stabilizing monthly electricity budgets, and achieve truly self-evolving intelligent long-term refined control scheduling.

[0044] Furthermore, to intuitively demonstrate the specific technical features and significant technological advancements of the short-period forward probability propagation prediction in Sp2.3, this behavioral causal inference engine incorporates the following three typical high-dimensional causal chain determination and output processes during actual operation: Scenario 1: Prediction of elderly people's risk of falling at night: After Sp2.1 extracts the meta-event consisting of "Time: 2:00 AM", "Location: Bedside in the bedroom", and "Action: Upright displacement of the human body", Sp2.3 propagates along the directed edges of the causal graph. If the graph shows that the member has experienced three consecutive causal chains of getting up frequently in the historical sliding window, the system will automatically calculate the probability of the "fall risk value" increasing due to fatigue on the movement line in the next 10 minutes. When the probability exceeds 75%, the system outputs a risk warning in advance and triggers the subsequent scheduling mechanism to turn on the night light slightly in advance on the movement line to perform active defense. Scenario 2: Prediction of Secondary High-Risk Safety Risks in the Kitchen: When Sp2.1 identifies the meta-events of "gas valve opening" and "continuous rise in kitchen ambient temperature", if the multimodal perception module detects that the human body in the space is stationary or has no displacement characteristics for a long time through the Doppler frequency shift of millimeter-wave radar (determined to be an elderly person sleeping or forgetting to leave home), Sp2.3 extrapolates along the causal chain to the future and predicts that the "probability of kitchen danger (dry burning, fire)" will show an exponential step trend within the next 30 minutes. The system then outputs a high-risk interception signal in advance and forcibly shuts off the gas solenoid valve. Scenario 3: Long-term long-tail energy waste prediction: When Sp2.1 extracts a meta-event where no one moves within the home space for an extended period, and the air conditioning unit is still operating at high power, Sp2.3 uses the probability forward propagation of the time axis related edges in the causal graph to deduce the energy waste trend that this behavior sequence will lead to in the next few hours. This prediction will directly trigger the virtual pre-execution scheduling mechanism, which will virtually calculate the energy consumption comparison after raising the air conditioning temperature in the twin space. After confirming that the energy-saving target is met, the real control will be issued to switch the air conditioning to a low-energy sleep state.

[0045] The virtual pre-execution scheduling mechanism intercepts control commands and projects them into a digital twin space for virtual control simulation before they are sent to the physical space. It performs full prediction of multi-device conflicts, user comfort, and system energy consumption during the simulation process, and determines whether to release the control commands based on the prediction results. Deployed in the Trusted Execution Environment (TEE) of the smart gateway or in a local high-performance edge computing isolation chamber, the core task of this mechanism is to break the traditional blind execution mode of smart homes that immediately sends commands to physical execution upon receipt. It establishes a digital security buffer between control commands and physical hardware. The mechanism's functional architecture consists of the following four core functional modules working together organically: Command interception and flow configuration module: responsible for intercepting various "control commands to be executed" issued by users, accessed by third-party platforms, or generated by the local intelligent causal inference engine in real time, suspending them and transferring them into the virtual sandbox pipeline; Digital Twin Spatiotemporal Sandbox Exercise Module: An independent simulation parallel space is created in memory to simulate the transient and steady-state responses of physical layer hardware after receiving instructions, enabling advanced exercise of control effects; Spatiotemporal boundary conflict detection module: Built-in hardware-level spatiotemporal conflict matrix, responsible for comparing whether the concurrent commands of multiple devices will trigger hidden conflicts or safety red lines when they intersect in physical space (such as simultaneous heating and cooling, or overload circuit breaking caused by concurrent high-power devices). Multi-objective collaborative optimization game module: It has a built-in utility function model and is responsible for comprehensively and quantitatively evaluating the simulation output results, and finding the optimal control balance point among "human comfort, system energy consumption, equipment life and timeliness".

[0046] The core function of this mechanism is to ensure that every control command issued to the physical space has undergone a complete safety, energy saving and comfort compliance audit through the closed-loop logic of "simulation first, execution later", which fundamentally eliminates hardware conflicts and long-tail energy waste in the collaborative control of multiple devices.

[0047] Furthermore, the specific process of this mechanism during operation is described in detail below: SP3: During the operation of this platform, whenever the system generates a control command to be executed (regardless of whether the command originates from the user's mobile APP, voice panel, or autonomous defense strategy output by the behavioral causal inference engine), the virtual pre-execution scheduling mechanism in the edge gateway is immediately activated. This step, as the main gate before physical distribution, is managed and executed by the gateway's secure isolation chamber (TEE), enabling multi-threaded digital space sandbox calculations and multi-objective optimization rating. If the control command is not audited in a virtual pre-execution manner, during concurrent operation by multiple members or intelligent linkage of multiple devices, the phenomenon of "energy consumption offsetting" may easily occur, such as the air conditioner cooling while the fresh air system draws in outdoor hot air, or the risk of the household circuit breaker tripping due to the simultaneous operation of high-power appliances. Through this step, the traditional trial-and-error control in the physical world is upgraded to error-free predictive control in the digital world.

[0048] To obtain a clean and independent simulation benchmark in the main audit process, control signals cannot directly reach the physical layer bus. They must be suspended first, and a simulation sandbox is customized for it in memory. Therefore, when the system enters SP3.1, the gateway's communication driver layer forcibly modifies its status bit to "PENDING" the moment it receives the control command, cutting off its data delivery to the physical bus (such as Matter / ZigBee bus). Immediately afterwards, the configuration module opens an isolated simulation sandbox space in the gateway's high-speed dynamic random access memory (DRAM) and pulls the latest high-reliability digital twin space shadow node status from the aforementioned virtual-real dynamic mapping engine in real time as the initial physical background of the sandbox, completing the initialization of the exercise environment. Dynamically intercepting and configuring the isolated sandbox is to ensure that the simulation exercise is entirely theoretical in the software world. Any calculation waveform or trial and error adjustment of any command will not disturb the physical devices in the physical world, thereby achieving advanced reasoning under the premise of absolutely ensuring the safety of home operation.

[0049] After the sandbox space is initialized and a highly reliable image is successfully loaded as the background, the primary technical task is to eliminate rigid logical conflicts that violate physics or hardware logic in multi-device concurrent control. To this end, the system enters SP3.2: the conflict detection module extracts multiple control instructions intercepted within the sandbox, converts them into vectors to be verified, and inputs them into the built-in spatiotemporal correlation conflict matrix model. Let the set of currently intercepted control instructions be... The system constructs a spatiotemporal correlation conflict matrix. The rows and columns of the matrix represent different device operation types. The gateway hardware processor runs a conflict resolution program for devices operating in the same room (spatial area Ω) within a preset overlapping sliding time window. The instruction pairs to be executed strictly follow the matrix element cross-calculation formula below: ; in, For the preset exclusive physical matrix of equipment functions (such as exclusive heating and cooling). Spatial intersection indicator function The expected execution time of the two instructions is given. If the time and space overlap and the functions are mutually exclusive, the matrix element values ​​will converge to 1, directly triggering a hard interception.

[0050] The model performs a spatial intersection indicator function on the input instructions. Boolean logic determination, and extraction of exclusive physical matrix. The mutual exclusion constants are combined with the time distance to perform exponential decay processing, and the resulting conflict matrix is ​​calculated. Any off-diagonal element in If the value exceeds the preset hard conflict threshold, the module immediately outputs a strong conflict interception signal, directly terminating the subsequent processing of the instruction stream, classifying it as a "high-risk conflict instruction," and notifying the self-evolutionary scheduling mechanism for online strategy rewriting. This is the key to achieving the inherent safety of the control system. By using the spatiotemporal correlation conflict matrix for quantitative solution, it is possible to identify strong spatial conflict risks online with extremely high computational efficiency (matrix multiplication), such as logical contradictions like "simultaneous operation of underfloor heating and air conditioning cooling in the same area," or "the instantaneous overload of electricity caused by the concurrent operation of a high-power oven and an instantaneous water heater," thus triggering the first line of defense on the digital side.

[0051] After successfully passing the rigorous spatiotemporal conflict detection and confirming that there are no safety red lines in the multi-device collaborative instructions, the instruction flow then enters a deeper stage of refined auditing of economy and comfort. The process enters SP3.3: For the set of control instructions that have passed the SP3.2 safety test, the multi-objective collaborative optimization game module uses them as input variables and calls the multi-objective game optimization utility function model in the sandbox space. The processor core synchronously pulls the long-term macro-environment state evolution curve output from the aforementioned behavioral causal inference engine. With the performance degradation trajectory of core equipment The model iteratively fits each possible combination of control deviations over a continuous-time sliding window, and the comprehensive utility function score is calculated using high-speed iteration. Its comfort utility Energy consumption penalty and penalties for component wear Weighted compromise composition: ; in, and These are the long-term environmental indicator evolution values ​​and equipment fatigue characteristic values, respectively, derived from the long-term physical neural network output of the aforementioned behavioral causal inference engine. The algorithm aims to solve for the dynamic weight bias injected into the self-evolving scheduling mechanism on the Pareto front, which makes the dynamic weight bias dynamically injected into the self-evolving scheduling mechanism. By maximizing the optimal control bias, the model transforms the contributions of simulated indoor temperature and humidity, and PM2.5 to the human thermal comfort index (PMV) into positive comfort utility at the algorithm layer. Simultaneously, the expected electrical power and gas consumption curves generated by each device during virtual response in the sandbox are converted into negative energy consumption penalties. And combine the fatigue characteristics of the equipment to penalize the wear and tear of parts. In the high-speed iterative solution of the formula, the shadow physics engine within the sandbox continuously simulates and records the continuous feedback of the instruction set on the physical environment over the next two hours, with a step size of 10 seconds. This results in the rendering and output of a "transient simulation evolution curve of indoor temperature / humidity over the next two hours" and a "transient simulation evolution curve of cumulative electrical power / energy consumption" in memory. Finally, this step calculates the discrete comprehensive utility score. The simulation evolution curves of these accompanying sub-indicators are packaged and output as a combined data stream.

[0052] The control and scheduling of smart homes is a typical dynamic multi-objective, multi-agent game process: excessive pursuit of extreme comfort (such as fixing the temperature at an absolute 24°C) will lead to soaring energy consumption and frequent start-stop of home appliances (shortened lifespan). Through this utility function model, physical quantities that were originally not directly comparable (temperature, electricity consumption, component wear) can be uniformly transformed into scalarized utility scores. At the same time, the simulation evolution curves accompanying the output can provide dynamic trend evidence on a continuous time axis for subsequent decision-makers, avoiding the technical defects of a single static score that cannot reflect instantaneous energy consumption exceeding the limit or instantaneous temperature overload. Thus, an optimal control path that balances comfort, energy saving, and long lifespan can be deduced in advance in the digital twin space.

[0053] With SP3.3 calculating the overall utility score of this instruction set online... The virtual pre-execution scheduling mechanism requires a strict set of compliance thresholds to make the final decision on whether to allow passage. Therefore, the system enters SP3.4: the mechanism uses the comprehensive utility score output from SP3.3. The simulation evolution curves of various sub-indicators (i.e., predicted curves of temperature / humidity / energy consumption changing over time) are sent to the compliance decision-maker inside the isolation chamber. The decision-maker compares these curves with the comprehensive target threshold set by the platform. (And the indoor comfort red line and maximum safe electricity limit stipulated by national / industry standards) are compared in real time. If the comparison results show that any indicator predicted by the simulation does not meet the target requirements (for example, the total utility score is lower than expected, or the energy consumption curve prediction will trigger the monthly household electricity consumption tiered pricing penalty), the decision-maker issues a "not pass" instruction, sends the control flow back and feeds it back to the self-evolving scheduling mechanism, which then re-corrects and rewrites the instruction parameters online according to the long-cycle constraint boundary (such as automatically raising the temperature of a certain air conditioner instruction by 0.5°C). Only when the simulation results of the virtual pre-execution fully meet, or even exceed, the target requirements ( Only when the decision-maker issues a release command and releases the data pipeline does it. This step establishes a fully automated digital compliance quality inspection red line, ensuring that all costs of execution failure (such as excessive electricity bills, discomfort for the elderly, and conflicts between multiple devices) remain in the virtual shadow world for trial and error and digestion. The control commands flowing to the real physical world must be perfect solutions after rigorous mathematical proof by the system, ensuring the absolute reliability and optimality of the control system.

[0054] Once the SP3.4 compliance decision-maker issues a pass signal, proving that the command stream has passed all safety, comfort, and energy efficiency audits, the mechanism, upon receiving the pass signal, immediately releases the "PENDING" state of the control command to be executed, sets the command's legality flag, and activates the gateway's communication bus driver module. This transforms the optimized control parameters at the software level into corresponding hardware electrical signal messages (such as IoT control frames conforming to the Matter standard). These messages are then precisely distributed and delivered to various real IoT devices in the physical space via the home's local area network and wireless radio frequency channels (such as Wi-Fi, Thread, and ZigBee). The underlying actuator hardware of these devices (such as air conditioner compressor relays, fresh air inverter motors, and gas valve stepper motors) receives the electrical signals and immediately responds, completing the final physical execution. By securely transforming digital commands that have passed rigorous mathematical audits into real hardware physical actions, the mapping from "highly reliable digital space drills" to "precise closed-loop control in physical space" is completed, thus providing users with a deterministic, self-evolving smart home living experience with strong safety and energy efficiency guarantees.

[0055] The self-evolving scheduling mechanism is used to monitor the runtime data in the virtual-real dynamic mapping engine over a long period of time. It dynamically learns user habits, time patterns, scene preferences, and energy consumption characteristics, and automatically generates new device collaborative scheduling and control strategies online. It is deployed in a distributed architecture: its lightweight policy executor and incremental data collection module are deployed in the smart edge gateway in the home to ensure the timeliness and privacy of control, while its deep policy derivation, policy generalization, and large-scale matrix training modules are deployed in a local high-computing-power private cloud or a trusted cloud server. The core task of this mechanism is to break the limitations of traditional smart homes that rely on "fixed linkage rules (IF-THEN)," which leads to rigid strategies and an inability to adapt to changes in user habits. In terms of functional architecture, this mechanism is organically composed of the following four functional modules: Multi-dimensional environment feedback stream monitoring module: responsible for long-term, uninterrupted collection of multi-modal environment response sequences synchronized from the virtual-real dynamic mapping engine, user's active fine-tuning behavior of control results (such as frequent manual overriding of automatic temperature adjustment by users), and energy consumption curve data over long periods. Personalized long-tail feature extraction module: responsible for filtering noise from massive historical feedback, mining the family's unique, complex long-tail preferences that drift with seasons and time, and quantitatively identifying the long-term hysteresis response characteristics of the family's physical environment to device control. Collaborative strategy online derivation and evolution module: It has a built-in reinforcement learning network and rule aggregation engine, which is responsible for dynamically reconstructing and updating the parameters in the control library based on the extracted preference features and the long-term physical constraint boundary inherited from the causal inference engine. Control rule pruning and merging module: responsible for transforming discrete control parameters into highly aggregated adaptive collaborative scheduling strategies, realizing lossless replacement and generalized iteration of the original fixed linkage rules.

[0056] The core function of this mechanism is to endow the system with the vitality of self-reproduction and autonomous evolution. By long-term monitoring of the feedback from the three-way game between the family's "environment-people-device", the control strategy is automatically optimized online and rules emerge at the software level, so that the system becomes smarter and smarter with use and more and more in line with the personalized lifestyle of a specific family.

[0057] Furthermore, the specific process of this mechanism during operation is described in detail below: SP4: During the long-term operation of this platform, whenever the virtual pre-execution scheduling mechanism delivers and executes control commands to the physical layer's real hardware, the self-evolution scheduling mechanism in the edge gateway asynchronously starts the main control process in the background. This process is managed by the gateway's multi-threaded incremental learning engine. Without affecting the real-time performance of the front-end control, it constantly listens for secondary feedback occurring in the physical space and is responsible for transforming the "experience" of the home's historical operation into the "intelligence" of the system's scheduling rules. Traditional smart homes use unchanging linkage rules, which cannot cope with the drift of living habits caused by changes in occupancy rates, aging of electrical components, seasonal changes, and the aging of users. Through this main control process, the system can have dynamic self-correction capabilities, get rid of dependence on manually configured rules, and achieve lifelong self-iteration of control strategies.

[0058] To provide accurate and reliable data fuel for strategy self-evolution, the system cannot rely solely on single snapshot data. A continuous feedback pipeline must be established on a unified timeline. Therefore, the system enters SP4.1: the multi-dimensional environmental feedback stream monitoring module continuously collects multi-dimensional time-series data after the execution of control and scheduling in the home physical space via the data bus. The input data stream mainly includes: A time series list of physical environment feedback (including indoor and outdoor temperature and humidity fluctuations and carbon dioxide concentration curves in 1-minute increments); a dataset of user active fine-tuning behaviors in response to control results (precisely recording the timestamps of user actions such as manually pressing the remote control to adjust the temperature and opening the curtains). The module aligns the heterogeneous multi-time series data with a unified hardware clock source, including the overall household electricity load curve and single device power consumption trajectory reported by the power metering chip over a long period, and generates a standardized "control-environment-behavior" multi-dimensional incremental log in the local database. The user's active fine-tuning behavior (such as the system automatically turning the air conditioner to 26°C, and the user manually adjusting it to 24°C within 3 minutes) is the most core and most authentic negative feedback evidence of human "dissatisfaction" with the current automation strategy. By strongly aligning and binding the environmental response and fine-tuning behavior in time, it is to provide highly accurate causal penalty labels for subsequent reinforcement learning algorithms and to identify the technical direction for strategy optimization.

[0059] As Sp4.1 accumulates rich aligned time-series logs in its underlying database, the system needs to dig beyond this surface data to uncover the family's implicit, personalized long-tail lifestyle habits and the house's unique thermodynamic delays. To this end, the process moves to Sp4.2: the feature extraction module pulls the standardized incremental logs generated by Sp4.1. First, through frequency statistics and clustering algorithms, it extracts the user's personalized environmental regulation long-tail preferences in different seasons and times (for example, the model identifies that the family tends to maintain extremely low humidity even when the indoor temperature is not high during the rainy season). Next, to quantify the thermal inertia delay caused by the building environment on equipment control, the module activates the built-in long-lag correlation function model. The module feeds back the actual indoor environment data from continuous historical time periods as a time series. and the corresponding historical equipment control power action sequence As the raw data input, the cross-correlation matrix is ​​integrally calculated in parallel on the gateway's multi-core processor. The maximum value of the cross-correlation coefficient is found on the sliding time axis using the following formula, thereby calculating the response time delay constant of the specific residential environment to the control action.

[0060] ; By iteratively solving the formula, the model can accurately identify the actual physical lag time, such as "from the time the air conditioner is turned on until the indoor temperature reaches a steady state" or "from the time the underfloor heating is turned on until the wall reaches thermal equilibrium." Finally, this step extracts the personalized environmental long-tail preference matrix and the long-lag response time delay constant. The system packages and outputs a highly refined "family characteristic configuration file" to memory, providing a mathematical benchmark for the precise evolution of the next step, the strategy library. Traditional smart home control often blindly pursues "instant feedback," ignoring the inherent large thermal inertia of the building structure. If the system cannot quantitatively extract the response delay... The reinforcement learning algorithm may misjudge if the indoor environment does not change immediately after the control action is issued, leading to over-adjustment of the control strategy, frequent start-stop or strategy oscillation. By extracting the mathematical information in this step, the physical time delay is explicitly injected into the reinforcement learning state machine, which can fundamentally ensure the convergence and stability of the self-evolution process.

[0061] After successfully extracting the long-tail preference profile and physical time-delay constraints of this specific household, the core closed-loop task of the entire smart hub becomes how to transform these implicit features into the driving force for upgrading the control strategy library. To this end, the process enters SP4.3: the online strategy derivation and evolution module pulls the home feature profile output by SP4.2 and connects in real time to the long-cycle constraint boundaries (such as equipment fatigue limit points, monthly energy consumption exceedance risk inflection points) output by the aforementioned behavioral causal inference engine. The module starts the built-in multimodal spatiotemporal decay incremental reinforcement learning model to convert the current spatial state (Composed of multi-dimensional environmental indicators and user activity patterns) and the proposed coordinated scheduling actions of the equipment. As input to the state-action space, the gateway processor first executes the action reward function in a hardware thread. Online solution: ; The formula is input to the current human thermal comfort deviation. Real-time energy consumption of the system And temperature differences caused by users manually fine-tuning the settings due to dissatisfaction with the current environment. ,in, To account for the time difference between the automatic control being issued and the user's manual fine-tuning, Using the time decay factor, the immediate reward of the current action is calculated. Then, the module immediately injects it back into the state-action value function matrix, and through the following spatiotemporally decaying nonlinear Q-learning iterative equation, it assigns the associated weight parameters of each device linkage link in the control strategy library. Perform online incremental updates: ; in, For incremental learning rate, As a discount factor for future returns, with The incremental refresh of matrix parameters means that when the score of a new device's collaborative control action chain continuously surpasses the historical fixed linkage rules, the rule pruning and merging module is activated. It automatically and adaptively merges the original discrete control conditions and finally outputs an adaptive collaborative scheduling strategy with personalized long-tail characteristics to the scheduling platform to replace the original fixed linkage rules.

[0062] This step is the final step in realizing the self-evolution of the strategy in this invention, and the dynamic reward function... The exponential decay mechanism introduced in China to address users' fine-tuning behavior within a short period of time. It possesses extremely high technical ingenuity: if the user responds within a very short time after the automation strategy is executed... When the value is extremely small, reverse fine-tuning is performed, and the penalty term will increase exponentially. The negative penalty directly forces the Q-learning formula to overturn the parameter, forcing the system to evolve in a direction that better matches the user's true intentions. This changes the technical pain point of traditional smart homes requiring tedious manual programming and modification of linkage scenarios, and realizes seamless, lossless, and efficient online self-evolution of collaborative scheduling and control strategies.

[0063] Furthermore, the control flow of this platform system forms a highly reliable closed-loop process of "physical sensing - digital calculation - interception optimization - feedback evolution": The control flow is activated by the virtual-real dynamic mapping engine. The system continuously captures the raw heterogeneous data stream reported by the multi-source perception matrix in the physical space. The control center immediately suspends it and decomposes the data stream into four confidence scalar components in real time at the edge: time, data, behavior, and environment. By constructing a high-dimensional virtual-real confidence field vector, the system performs matrix projection on the actual reported state and the theoretical predicted state of the twin space. If the calculated virtual-real deviation value exceeds the security trust threshold, the control flow will be directed to the state correction and anomaly rollback unit to remove maliciously distorted dirty data caused by component hardware drift, thereby completing a highly reliable clean alignment and injecting the clean state into the twin shadow space.

[0064] The aligned high-reliability data stream then triggers the behavioral causal inference engine. At this point, the control flow evolves in parallel on two tracks: On the short-cycle track, the system injects the extracted multimodal causal events into the time-sliding family behavioral causal graph in memory, and starts the Dynamic Causal Bayesian Network (DCBN) model to perform conditional probability forward iterative propagation along the directed correlation edges, predicting the probability of personal safety or secondary high-risk equipment hazards in the next few minutes to hours; On the long-cycle track, the system takes the current environmental trend as the initial boundary condition, introduces a time acceleration factor to speed up the clock in the software independent sandbox, and uses the Physical Information Neural Network (PINN) to continuously integrate and evolve the energy conservation equation and the fatigue degradation trajectory of home appliances, predicting the macro-environmental degradation trend in the next few days or even months, and feeding these trends back to the strategy generation end as long-cycle physical constraint boundaries.

[0065] When the system generates an autonomous control strategy due to proactive defense or receives native control commands from family members, the control flow is injected into the Trusted Execution Environment (TEE) of the virtual pre-execution scheduling mechanism. All commands to be executed are immediately suspended and transformed into vectors to be verified and input into the spatiotemporal correlation conflict matrix model. Through cross-calculation of matrix elements, strong conflict commands that violate ergonomic red lines or pose a risk of concurrent overload tripping are forcibly intercepted. The command flow that passes the initial review is further input into the multi-objective game optimization utility function model. The sandbox physics engine performs high-speed simulation fitting with a fixed step size, and renders the transient simulation evolution curves of environmental indicators and cumulative power consumption in memory. The control parameters that maximize the global comprehensive utility are solved on the Pareto front. After the compliance decision device performs trajectory auditing on the evolution curve and passes it, the suspension state is officially lifted. The control flow is transformed into Matter standard IoT hardware electrical signal messages and distributed to the actuators of each physical device, realizing precise physical control to eliminate hidden dangers in advance and smooth out peak power consumption.

[0066] After the physical execution of the command, the control flow eventually flows into the self-evolutionary scheduling mechanism to initiate lifelong learning in the background. The multi-dimensional environmental feedback flow monitoring module generates incremental logs by aligning the actual environmental response, power load curve, and user fine-tuning behavior collected over a long period of time with the hardware clock source strength. The feature extraction module performs sliding cross-correlation integral calculations through a long-delay correlation function model to accurately identify the building's unique environmental long-delay response time delay constant. Subsequently, the self-evolutionary evolution module injects this time delay constant and the aforementioned long-period constraint boundary into the reinforcement learning state machine, running iterative equations containing a spatiotemporal decay incremental reinforcement learning model (Q-learning) on ​​a continuous time axis. The dynamic reward function sensitively captures the time difference between user fine-tuning and the issuance of automated strategies through a time decay factor, and uses nonlinear penalties to drive incremental updates of the state-action value function matrix, thereby adaptively pruning and merging the original linkage rules. The control flow eventually automatically emerges online to generate a new adaptive collaborative scheduling strategy with the most personalized long-tail rules for the family, completing the self-reproduction and intelligent evolution of the control center.

[0067] Please see Figure 3 and Figure 6 As shown, it provides a screenshot of the system's main interface. Figure 3 The interface display includes an environmental data panel. Figure 4 The interface display includes a device status dashboard. Figure 5 The interface display includes a "reliability monitoring" section to assess whether the image is real or virtual. Figure 6 The interface displays a causal graph inference link, which is used to aggregate and clean up the indoor multimodal physical indicators of the home space in real time. It dynamically displays the online / discrete controlled status of various home appliances and sensing nodes, comprehensively audits the four-dimensional credibility field vector and virtual-real mapping deviation value of the core sensors, and proactively analyzes the dynamic correlation weight and historical co-occurrence frequency of causal events such as "door and window opening - human movement - light / home appliance response". In this way, it provides a highly reliable digital twin shadow background and causal inference benchmark for the entire self-evolving control and scheduling platform at the edge.

[0068] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0069] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios, characterized by: The scheduling platform includes a virtual-real dynamic mapping engine, a behavior causal inference engine, a virtual pre-execution scheduling mechanism, and a self-evolving scheduling mechanism, wherein: The virtual-real dynamic mapping engine is used to collect device status, environmental data and multimodal behavior data in the home physical space. By dynamically constructing a high-dimensional virtual-real credibility field composed of time credibility, data credibility, behavior credibility and environmental credibility for each device, it evaluates the virtual-real deviation value between the physical space and the digital twin space in real time, and triggers a self-healing control mechanism when the virtual-real deviation value exceeds a preset threshold, so as to achieve high-credibility dynamic alignment from the physical space to the twin space. The behavioral causal inference engine is used to construct a time-sliding family behavioral causal graph based on family historical behavioral sequences, dynamic environmental changes, multi-member interaction information, and device status correlation. It also uses the causal graph to predict the causal inference of family future short-term events, and combines it with long-term future state evolution schemes to infer macro trends. The virtual pre-execution scheduling mechanism is used to intercept control commands and project them into the digital twin space to perform virtual control simulation before the control commands are sent to the physical space. It makes full predictions on multi-device conflicts, user comfort, and system energy consumption during the simulation process, and determines whether to release the control commands based on the prediction results. The self-evolving scheduling mechanism is used to monitor the running data in the virtual-real dynamic mapping engine for a long time, dynamically learn user living habits, time patterns, scene preferences and energy consumption characteristics, and automatically generate new device collaborative scheduling and control strategies online.

2. The digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios according to claim 1, characterized in that: The intelligent control and scheduling method of the scheduling platform includes the following steps: Sp1: Constructs a high-dimensional virtual-real credibility field for home devices through a virtual-real dynamic mapping engine, dynamically calculates virtual-real deviation values, and ensures high credibility alignment between the digital twin space and the physical space; Sp2: Utilizes a behavioral causal inference engine to construct a family behavioral causal graph, performs short-term forward probability propagation prediction based on current family dynamics, and combines long-term future state evolution schemes to predict long-term macro-environmental evolution and equipment fatigue characteristics. Sp3: When a control command to be executed is generated, a virtual pre-execution scheduling mechanism is used to perform simulation pre-run, conflict detection, comfort and energy consumption prediction in the digital twin space. The control command is only sent to the real equipment in the physical space for execution when the virtual simulation results meet the target. Sp4: Through a self-evolving scheduling mechanism, it dynamically learns family characteristics and habits, and continuously generates and iterates online device collaborative scheduling control strategies.

3. The digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios according to claim 2, characterized in that: The virtual-real dynamic mapping engine in Sp1 calculates the virtual-real deviation value through the following steps: Sp1.1: Calculate time reliability based on the difference between the device's historical reporting cycle and the current timestamp; calculate data reliability based on the integrity of the data packet and the noise range; calculate behavior reliability based on whether the device's current action conforms to the device's preset operation logic chain; and calculate environmental reliability based on the data collaborative verification results of associated environmental sensors. Sp1.2: Interweave time credibility, data credibility, behavior credibility and environmental credibility in multiple dimensions, map them into the virtual and real credibility field of each device in the current time and space, and compare the predicted state of the twin space with the actual reported state of the physical space to calculate the virtual and real deviation value that reflects the degree of virtual and real alignment. Sp1.3: Determine whether the virtual-to-real deviation value exceeds the preset confidence threshold. If it does, determine that the physical reported data or twin model is distorted and automatically trigger the self-healing control mechanism.

4. The digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios according to claim 3, characterized in that: The self-healing control mechanism includes: The resampling unit is used to increase the physical data sampling frequency of the corresponding faulty device and perform multi-channel retransmission. The state correction unit is used to perform weighted offset correction on the distorted state of the abnormal device using observation data from associated sensors located in the same physical area as the abnormal device. The edge reconstruction unit is used to reinitialize the twin node of the device at the edge side by calling the underlying initial physical behavior model of the device from the local gateway when the correction fails. An abnormal rollback unit is used to control the state of the device in the twin space to roll back to the last historical state that was determined to be trustworthy during reconstruction.

5. The digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios according to claim 2, characterized in that: The Sp2 behavioral causal inference engine constructs a family behavioral causal graph, including the following steps: Sp2.1: Using time as the axis, extract continuous actions of family members and match the spatial location, environmental index fluctuations, and response status of surrounding devices when the actions occur as causal events; Sp2.2: Establish directed association edges between different causal events, where the direction of the association edge is determined by the chronological order, and the weight of the association edge is determined by the frequency of historical occurrence and the depth of environmental coupling, thereby generating a causal graph of family behavior. Sp2.3: Based on the causal meta-events triggered at the current moment, forward probability propagation is carried out along the directed correlation edges of the family behavior causal graph to predict abnormal high-risk state trends or high energy consumption tendencies in the future short-term period.

6. The digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios according to claim 5, characterized in that: The abnormally high-risk state trends or high energy consumption tendencies in the future time period include: the trend of increased fall risk due to the behavioral events of the elderly frequently getting up at night; the trend of increased kitchen hazard probability due to the behavioral sequence of continuously rising kitchen ambient temperature, gas valve opening and long-term stillness of people in the space; and the trend of energy waste due to the behavioral sequence of long-term no-person movement in the family space and air conditioning equipment being in operation.

7. The digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios according to claim 2, characterized in that: The Sp2 behavioral causal inference engine includes the following steps when performing long-term future state evolution: Sp2.4: Building upon the results of short-cycle simulations, the fluctuation trend of environmental indicators within a short-cycle period is used as the initial boundary condition. In the digital twin space, a model of long-cycle thermal inertia decay parameters of buildings and cumulative loss of equipment operation is introduced. Sp2.5: Performs accelerated time-flow simulation evolution in a digital twin space to simulate the evolution of macroscopic environmental conditions, long-term household electricity load accumulation, and equipment performance degradation trajectory over a long period of time. Sp2.6: Assess whether the evolution of the macro-environmental state triggers the long-term environmental degradation threshold, or whether the performance degradation trajectory reaches the equipment fatigue limit ahead of schedule, and feed the assessment results back to the self-evolutionary scheduling mechanism in real time as a constraint boundary.

8. The digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios according to claim 2, characterized in that: The virtual pre-execution scheduling mechanism in Sp3 performs virtual control simulation by including the following steps: SP3.1: Intercept sets of control instructions to be executed from user-triggered, causal inference-triggered, or self-evolutionary strategies; SP3.2: Input the set of control instructions to be executed into the current digital twin space, simulate execution in the shadow physics engine, calculate the physical interference conflicts of multiple devices in a concurrent state, and simulate and predict the changes in physical environment parameters after execution; Sp3.3: Input the predicted physical environment parameters into the preset user comfort model and energy consumption assessment model, and determine whether the simulation results simultaneously meet the environmental comfort scale and energy consumption budget target; SP3.4: If the judgment result is satisfied and there is no physical interference conflict, the control interception barrier is opened, and the control instruction set to be executed is converted into a real electrical signal and sent to the real device in the physical space. If the judgment result is not satisfied or there is a physical interference conflict, the control instruction is terminated and the simulation conflict reason is fed back to the scheduling platform.

9. The digital twin virtual-real mapping and intelligent control scheduling platform for home scenarios according to claim 2, characterized in that: The self-evolutionary scheduling mechanism in Sp4 dynamically learns family characteristics and habits, including the following steps: SP4.1: Continuously collect environmental feedback time series tables after the execution of control scheduling in the home physical space, user's active fine-tuning behavior of the control results, and electricity load curves over long periods; Sp4.2: Extract users' personalized environmental adjustment long-tail preferences in different seasons and at different times, and combine them with the evaluation results of the long-term future state evolution to identify the hysteresis response characteristics of the physical space environment to device control. SP4.3: Based on extracted preferences, features, and long-term constraint boundaries, dynamically update the associated weight parameters in the control strategy library, or adaptively merge the original control conditions to automatically generate an adaptive collaborative scheduling strategy to replace the fixed linkage rules.

Images