A smart office space management method based on digital twin technology

Abstract

The application relates to the technical field of digital twinning and environment control, and discloses a smart office space management method based on a digital twinning technology, which comprises the following steps: constructing a lumped parameter network model to calculate theoretical state data, calculating a state residual error vector of measured data and the theoretical data, decoupling the residual error into a transient load disturbance component and a structural parameter drift component through a double-frequency band filter, inversely projecting the transient component into an equivalent disturbance parameter injection model for advance deduction by using a model inverse dynamics principle, and identifying and correcting a model structure parameter matrix by using a gradual state component; through frequency domain decoupling and reverse analysis of the state residual error, the software-defined perception of unmeasurable implicit loads and the internal adaptive correction of model aging drift are realized, response lag of a large thermal inertia system is avoided, and the self-evolution ability of the system in the whole life cycle is improved without additional hardware investment.

Description

Technical Field

[0001] This invention relates to a smart office space management method based on digital twin technology, belonging to the field of digital twin and environmental control technology. Background Technology

[0002] In current building automation and agile office management, the flow of people in office spaces exhibits highly random and transient characteristics, causing the spatiotemporal distribution of heat and humidity loads within the space to deviate from traditional fixed timetable patterns. Current mainstream control adopts a closed-loop control logic based on physical sensor feedback, utilizing temperature and humidity sensors distributed in the environment to monitor real-time status. When the monitored values ​​deviate from the set target range, the controller drives the air conditioning terminals or fresh air equipment to adjust based on the deviation signal. This control mode relies on the explicitness of physical parameter deviations, and the control loop only activates after a substantial deviation in the environmental state. For thermally inertial central air conditioning water systems and large-scale physical spaces, there is a time lag between the terminal action and the establishment of thermal balance in the entire space. Faced with sudden gatherings of people or step-like load disturbances caused by concentrated activation of equipment, the feedback control logic falls into a catch-up error oscillation state, causing environmental parameters to fluctuate significantly around the target value, resulting in wasted energy from cold and heat sources. Existing solutions attempt to introduce visual recognition or infrared counting to directly monitor the flow density and achieve feedforward compensation. However, these solutions are limited by privacy compliance risks, high hardware deployment costs, and inaccurate counting due to visual obstruction, and are difficult to cover non-biological heat source disturbances caused by equipment heating.

[0003] Simply stacking hardware or simple spatial mapping cannot solve the lag at the control algorithm level. For example, Chinese invention patent CN118158364A discloses an office digital twin monitoring system and method. It fills in monitoring blind spots and intuitively manages equipment by mapping the simulated state of personnel. The core logic focuses on the spatial reproduction of visual information and remote issuance of commands. It does not establish an inverse dynamic relationship between thermal and humidity load and environmental response. When faced with sudden transient disturbances caused by sudden gathering of personnel, the system cannot quantify the specific thermal and humidity load values. It needs to wait for substantial changes in environmental parameters or rely on manual judgment to trigger adjustment, which cannot fundamentally eliminate the response lag of the large thermal inertia system. The physical environment system is affected by factors such as aging of the building envelope, dust accumulation on filters and deterioration of heat exchanger efficiency over a long period of time. The thermodynamic characteristic parameters drift slowly. Single feedback control or traditional feedforward control cannot distinguish whether the environmental deviation is caused by sudden changes in external transient load or evolution of internal structural parameters. The confusion of multi-source error attribution leads to the gradual failure of control strategies based on fixed models or single feedback loops throughout the entire life cycle. The system mistakenly attributes the decline in equipment capacity due to equipment aging to increased load, and issues incorrect adjustment commands, which exacerbates energy consumption and equipment wear.

[0004] Therefore, the technical problem to be solved by this invention is how to accurately identify transient load disturbances and eliminate the response lag of large inertial systems by utilizing existing basic environmental monitoring data without relying on dedicated sensing hardware, while distinguishing between external load changes and internal performance degradation adaptive capabilities. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A smart office space management method based on digital twin technology, the method comprising the following steps:

[0006] A lumped parameter network model of the physical office space is constructed. This model defines a structural parameter matrix that describes the heat transfer characteristics between nodes. The theoretical state data of the physical office space is input based on the boundary conditions.

[0007] The measured state data of the environmental monitoring device in the physical office space is collected in the first cycle, and the theoretical state data of the lumped parameter network model at the current moment is obtained simultaneously. The state residual vector between the two is calculated.

[0008] Configure a first filter and a second filter with different cutoff frequencies, and input the state residual vector into the first filter and the second filter respectively to obtain the first frequency band residual component characterizing transient load disturbance and the second frequency band residual component characterizing structural parameter drift.

[0009] The residual components of the first frequency band are input into the state observer. The equivalent perturbation parameters inside the lumped parameter network model are solved using the principle of model inverse dynamics. The equivalent perturbation parameters are then injected into the lumped parameter network model as boundary conditions to deduce the environmental state evolution trajectory within a future preset time window.

[0010] The second frequency band residual component is input into the parameter identification module, and the least squares algorithm is used to correct the structural parameter matrix of the lumped parameter network model.

[0011] When the environmental state evolution trajectory exceeds the preset target range, the control compensation amount at the current moment is calculated based on the prediction results of the lumped parameter network model, and the control command is generated and sent to the environmental control equipment.

[0012] Preferably, the steps for constructing a lumped parameter network model of a physical office space include: discretizing the physical office space into multiple virtual thermodynamic nodes, and determining the thermal resistance and heat capacity parameters connecting each virtual thermodynamic node based on the geometric dimensions of the physical office space and the thermophysical properties of the enclosure structure; establishing a set of linear differential equations describing the heat transfer and temperature dynamic evolution between each virtual thermodynamic node based on the law of conservation of energy, and transforming the set of linear differential equations into a state-space equation form as a lumped parameter network model.

[0013] Preferably, the steps for solving the equivalent perturbation parameters inside the lumped parameter network model using the principle of model inverse dynamics include: constructing a Kalman filter or Runge-Kutta observer containing extended state variables, and inputting the first frequency band residual component as a correction term into the observer; in the iterative calculation of the observer, treating the heat source term of the lumped parameter network model as an unknown variable to be identified, and inversely approximating the virtual heat source value that can generate the current state residual vector by minimizing the cost function of the first frequency band residual component; and determining the converged virtual heat source value as the equivalent perturbation parameter, which is a value characterizing the dynamic heat and humidity load that is not directly measured in the physical office space.

[0014] Preferably, the steps for correcting the structural parameter matrix of the lumped parameter network model include: setting a structural parameter correction period, which is greater than the first period; accumulating the data sequence of the second frequency band residual components within each structural parameter correction period; constructing a sensitivity equation for the structural parameter matrix and solving for the correction amount of the structural parameter matrix using a recursive least squares algorithm with a forgetting factor; and updating the thermal resistance and thermal capacity values ​​in the structural parameter matrix using the correction amount when the magnitude of the correction amount exceeds a preset safety threshold.

[0015] Preferably, the method further includes an active verification step for the control loop, monitoring the amplitude of the state residual vector, and when the amplitude remains below a steady-state threshold for a preset period, generating a micro-perturbation control signal and simultaneously applying it to the environmental control equipment and the lumped parameter network model; and acquiring the actual dynamic response sequence of the environmental monitoring device to the micro-perturbation control signal. And the theoretical dynamic response sequence of the lumped parameter network model And calculate the consistency index of the two according to the following formula. : ,in, This is the mean of the actual dynamic response sequence. The mean of the theoretical dynamic response sequence; when the consistency index If the value falls below the preset confidence threshold, the control loop of the physical office space is determined to be faulty, the generation of control commands is stopped, and a fault indicator is output.

[0016] Preferably, the steps for extrapolating the environmental state evolution trajectory within a future preset time window include: using the measured state data at the current moment as the initial condition and the set containing equivalent perturbation parameters as the boundary condition, iteratively solving the lumped parameter network model in the virtual time domain at a step size faster than the passage of physical time; generating time series data containing temperature, humidity and air quality indicators at each future moment as the environmental state evolution trajectory.

[0017] Preferably, the step of calculating the control compensation amount at the current moment includes: constructing a target functional containing energy consumption cost and comfort penalty terms based on the deviation between the environmental state evolution trajectory and the preset target range; solving the control input sequence that minimizes the target functional under the premise of satisfying the environmental control equipment action constraints; and selecting the first element in the control input sequence as the control compensation amount at the current moment.

[0018] Preferably, the passband frequency of the first filter is set to cover the frequency band of heat load changes caused by personnel movement, and the passband frequency of the second filter is set to cover the frequency band of parameter drift caused by changes in the thermal performance of the building envelope and the attenuation of equipment efficiency.

[0019] Preferably, the method further includes a spatial heat load distribution reconstruction step based on equivalent disturbance parameters. The equivalent disturbance parameters obtained by solving are mapped to the corresponding virtual thermodynamic nodes using the topology of the lumped parameter network model. A heat load distribution heat map of the physical office space is generated based on the amplitude of the equivalent disturbance parameters of each virtual thermodynamic node. The heat load distribution heat map represents the regional personnel density or equipment heating intensity.

[0020] Preferably, the method further includes a circuit breaker protection step for abnormal operating conditions, which involves real-time monitoring of the rate of change of the equivalent disturbance parameters; when the rate of change exceeds a preset physical limit threshold, determining that the state residual vector contains non-physical interference signals; keeping the current boundary conditions of the lumped parameter network model unchanged, and temporarily blocking the injection of the equivalent disturbance parameters into the lumped parameter network model until the rate of change falls back to within the physical limit threshold.

[0021] Compared with the prior art, the beneficial effects of the present invention are:

[0022] 1. This invention utilizes the virtual time advance characteristic to eliminate the response lag of the physical system. By calculating the residual vector between the measured physical state and the theoretical state of the digital twin model, the residual is inversely projected into equivalent disturbance parameters using a state observer and injected into the model for ultra-real-time inference. Taking advantage of the time difference characteristic that the virtual calculation speed of the lumped parameter network model is much higher than the thermal inertia evolution speed of the physical entity, the control compensation amount required to maintain the target state is pre-calculated and commands are issued before the actual deviation of the physical space environment parameters. Based on the model feedforward compensation logic, the traditional feedback control relies on the adjustment lag mode after the deviation occurs from the control timing, which solves the technical problem that the large thermal inertia physical environment system cannot establish thermal balance in time when facing high-frequency random load disturbances, and ensures that the control action is synchronized with the load change in real time.

[0023] 2. Based on residual inverse analysis, non-intrusive equivalent load sensing is achieved. A soft measurement mechanism is established that treats state residuals as information carriers. This abandons the conventional path of directly measuring the number of people or the heat generated by equipment by adding dedicated sensors. By solving the inverse problem of the model's state equation, the complex heat and humidity load caused by the discrete behavior of personnel flow and equipment switching in the physical space is uniformly analyzed into a single continuous equivalent disturbance parameter in the model's boundary conditions. This allows the control system to obtain the accurate load information required for driving predictive control without relying on visual acquisition devices or high-density sensor arrays. This reduces the system's hardware deployment costs and maintenance complexity, and avoids the privacy and compliance risks of directly collecting biometric information.

[0024] 3. By maintaining the fidelity of the full life cycle model through dual-channel spectrum decoupling, and utilizing the frequency domain orthogonality of transient load disturbances and gradual structural aging in the physical process, a parallel filtering channel for state residuals is constructed. The high-frequency residual components are used as fast variables to input the observer to update the disturbance parameters required for real-time control, while the low-frequency residual components are used as slow variables to input the parameter identification module to iteratively correct the structural parameter matrix representing the thermal resistance of the building envelope and the heat exchange efficiency of the equipment. The hierarchical correction mechanism eliminates the error attribution confusion that may occur when a single feedback loop is operated for a long time. This ensures that the digital twin model can automatically distinguish and adapt to random changes in the external environment and the performance degradation of the physical entity itself, so that the model can always maintain a high-precision mapping of the thermodynamic properties of the physical entity without manual recalibration. Attached Figure Description

[0025] Figure 1 This is a flowchart of the closed-loop control of the digital twin system based on dual-band residual decoupling according to the present invention.

[0026] Figure 2 This is a schematic diagram illustrating the time-domain evolution and frequency-domain decoupling effect of the state residual vector in this invention;

[0027] Figure 3 This is a diagram illustrating the core technical architecture of the digital twin management method for smart office spaces in this invention. Detailed Implementation

[0028] The following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0029] This invention discloses a smart office space management method based on digital twin technology. It includes a closed-loop control system comprising a physical sensing layer, a digital twin computing layer, and an environmental control execution layer. The system runs on an edge computing gateway or server equipped with a processor. The physical sensing layer includes temperature sensors, humidity sensors, carbon dioxide concentration monitors, and return air temperature probes installed at the air conditioning terminals, distributed throughout the office space. The environmental control equipment includes fan coil units of the central air conditioning system, damper actuators of the fresh air handling unit, and chiller / heater units. For office space management, a lumped parameter network model of the physical office space is constructed, discretizing the continuous physical space into a finite number of virtual thermodynamic nodes. Based on the geometric dimensions of the physical space and the thermophysical properties of the enclosure structure, the thermal resistance and heat capacity parameters connecting each virtual thermodynamic node are determined. Based on the law of conservation of energy, a system of linear differential equations describing heat transfer and temperature dynamics between the virtual thermodynamic nodes is established, and this system of linear differential equations is transformed into state-space equations. As a lumped parameter network model, where... This is a state vector containing the temperature of each node. This is a control input vector that includes both the air conditioning cooling capacity and the heating capacity. Let the environmental disturbance vector be... and These are the structural parameter matrices characterizing the inherent properties of the system; during system operation, in the first period... Collect measured status data of environmental monitoring devices within physical office spaces. First cycle Set a time step of 1 to 5 minutes to synchronously acquire the theoretical state data of the lumped parameter network model under given boundary conditions at the current moment. Calculate the measured state data Compared with theoretical state data The difference between them generates a state residual vector. The state residual vector is processed using a first filter and a second filter with different cutoff frequencies. The first filter is set as a high-pass filter, with its passband frequency covering the frequency band of heat load changes caused by personnel movement. The cutoff frequency... Set to be higher than the reciprocal of the thermal response time constant of the building envelope, it is used to extract the first frequency band residual component characterizing transient load disturbances. The second filter is set as a low-pass filter, with its passband frequency covering the parameter drift frequency band caused by changes in the thermal performance of the building envelope and the attenuation of equipment efficiency, and its cutoff frequency... Set to a frequency lower than that of seasonal climate change, used to extract the second-band residual components characterizing structural parameter drift. .

[0030] The equivalent perturbation parameters inside the lumped parameter network model are solved using the principle of model inverse dynamics. A state observer containing extended state variables is constructed, and a Kalman filter or Runge-Kutta observer is selected to include the first frequency band residual components. As a correction term input to the observer, the heat source term of the lumped parameter network model is treated as an unknown variable to be identified in the iterative calculation of the observer. This is achieved by minimizing the residual components of the first frequency band. The cost function is used to inversely approximate the virtual heat source value that can generate the current state residual vector. The converged virtual heat source value is determined as the equivalent perturbation parameter. This parameter characterizes the dynamic heat and humidity load values ​​within the physical office space that are not directly measured; the equivalent disturbance parameter The lumped parameter network model is injected as a boundary condition to predict the environmental state evolution trajectory within a preset time window. Using the current measured state data as the initial condition and a set containing equivalent disturbance parameters as the boundary condition, the lumped parameter network model is iteratively solved in the virtual time domain with a step size faster than physical time elapsed. This generates time-series data containing temperature, humidity, and air quality indicators for each future time as the environmental state evolution trajectory. When the environmental state evolution trajectory exceeds the preset target range, the control compensation amount for the current time is calculated based on the prediction results of the lumped parameter network model, and a control command is generated and issued to the environmental control equipment. When calculating the control compensation amount, a target functional including energy consumption cost and comfort penalty terms is constructed based on the deviation between the environmental state evolution trajectory and the preset target range. Under the premise of satisfying the action constraints of the environmental control equipment, the control input sequence that minimizes the target functional is solved, and the first element in the control input sequence is selected as the control compensation amount for the current time. The structural parameter matrix of the lumped parameter network model is corrected, and the structural parameter correction period is set. This correction cycle Greater than the first cycle During each structural parameter correction period, the second frequency band residual components are accumulated. Based on the data sequence, a sensitivity equation for the structural parameter matrix is ​​constructed. The correction amount of the structural parameter matrix is ​​solved using a recursive least squares algorithm with a forgetting factor. When the magnitude of the correction amount exceeds the preset safety threshold, the thermal resistance and heat capacity values ​​in the structural parameter matrix are updated using the correction amount.

[0031] This method also includes an active verification step for the control loop, monitoring the amplitude of the state residual vector. When this amplitude remains below a steady-state threshold for a preset period, a micro-perturbation control signal is generated and simultaneously applied to the environmental control equipment and the lumped parameter network model. The amplitude of the micro-perturbation control signal is less than the critical value that would cause the environmental state to exceed the target range. The actual dynamic response sequence of the environmental monitoring device to the micro-perturbation control signal is then obtained. And the theoretical dynamic response sequence of the lumped parameter network model According to the formula Calculate the consistency index between the two. ,in, This is the mean of the actual dynamic response sequence. The mean of the theoretical dynamic response sequence, when the consistency index When the pressure falls below a preset confidence threshold, the control loop of the physical office space is deemed to be faulty, the generation of control commands is stopped, and a fault indicator is output. In addition, the spatial heat load distribution is reconstructed based on equivalent disturbance parameters. Using the topology of the lumped parameter network model, the solved equivalent disturbance parameters are mapped to the corresponding virtual thermodynamic nodes. Based on the amplitude of the equivalent disturbance parameters of each virtual thermodynamic node, a heat load distribution heat map of the physical office space is generated to characterize the regional personnel density or equipment heating intensity. At the same time, the circuit breaker protection steps for abnormal operating conditions are executed, and the rate of change of the equivalent disturbance parameters is monitored in real time. When the rate of change exceeds the preset physical limit threshold, it is determined that the state residual vector contains non-physical interference signals. The current boundary conditions of the lumped parameter network model remain unchanged, and the injection of equivalent disturbance parameters into the lumped parameter network model is temporarily blocked until the rate of change falls back to within the physical limit threshold.

[0032] Example 1: In an agile office space with an area of ​​500 square meters, accommodating 80 workstations and including 3 instant meeting areas, the personnel density exhibits high-frequency random fluctuations with project progress, and the central air conditioning terminal equipment has been in operation for more than 5 years, resulting in an unknown degree of degradation in heat exchange efficiency; when a large number of people flood into the space during the morning peak hours and multiple impromptu meetings are held, causing a sudden jump in indoor heat and humidity load, the system operates on a 1-minute cycle. Simultaneous collection of temperature, humidity and carbon dioxide concentration monitoring data And run the lumped parameter network model to calculate the theoretical state data under the current boundary conditions. Then, the difference between the two is calculated in real time to generate the state residual vector. The input is then fed into a first filter and a second filter configured in parallel. The first filter extracts the first frequency band residual component that characterizes the high-frequency load variation. The system uses a state observer to... Inverse analysis to equivalent perturbation parameters This allows for the direct quantification of the additional heat load caused by a surge in people at any given moment without the need for invasive crowd counting equipment; based on the analytical results... The system is injected into the lumped parameter network model as a boundary condition and extrapolates the environmental state evolution trajectory 100 times faster in the virtual time domain. According to the extrapolation results, the temperature in the conference area will exceed the comfort limit in 15 minutes. The system solves the optimal control sequence based on the model predictive control algorithm and generates control commands to increase the opening of the chilled water valve and increase the fan speed 15 minutes in advance, which are then sent to the environmental control equipment. This model-based feedforward compensation mechanism ensures that the cooling capacity is delivered and thermal balance is established before the indoor temperature actually deteriorates, avoiding temperature overshoot and oscillation caused by thermal inertia in traditional feedback control.

[0033] During the long-term operation of the system, the second filter continuously monitors the low-frequency components in the state residual vector. When the second frequency band residual component is detected... When a long-term unidirectional drift trend is observed, the system determines that the thermal resistance of the building envelope or the heat exchange efficiency of the equipment has undergone structural changes, and uses a recursive least squares algorithm to process the accumulated [data / effects]. The data sequence identifies new structural parameter matrices and updates the lumped parameter network model, thereby automatically calibrating model biases caused by equipment aging. During nighttime low-load steady-state periods, when the state residual is consistently below the steady-state threshold, the system automatically generates a micro-disturbance signal with an amplitude of 2% of the damper opening and injects it into the system. The dynamic response consistency index is compared between physical monitoring data and model prediction data. To confirm the integrity of the physical connection between the sensor and the actuator and to eliminate the risk of silent failure, this embodiment uses frequency domain decoupling and inverse dynamic analysis mechanism to simultaneously resolve the contradiction between transient load response lag and gradual model aging inaccuracy within a single control architecture.

[0034] Example 2: This example describes a hardware-in-the-loop simulation experiment. The test platform consists of a constant-temperature environment chamber with adjustable anisotropic thermal conductivity, a heat source simulator containing a programmable heating resistor array, and an embedded control unit running the algorithm of this invention. The thermal resistance of the enclosure structure of the environment chamber is... Design value heat capacity Design value The data acquisition system uses a precision of The industrial-grade thermocouple array, with a sampling frequency set to For the first cycle The setting is based on the Nyquist sampling theorem and the system's thermal inertia time constant. Engineering trade-offs, setting it as This value ensures the capture of heat load fluctuations caused by personnel movement, while suppressing the interference of high-frequency measurement noise introduced by oversampling on differential operations.

[0035] During the experiment, the target temperature of the environmental chamber was set as follows: The standard deviation is superimposed on the raw temperature data collected by the sensor. Gaussian white noise was used to simulate electromagnetic interference and measurement errors in real industrial environments. The experiment was divided into two stages: the first stage simulated transient load disturbances. At that time, an amplitude of [value] is introduced through the heating resistor array. The step heat load; the second stage simulates the gradual structural parameter drift, in to During the process, the thermal conductivity of the environmental chamber walls is adjusted to reduce thermal resistance. linear decay to initial value A control group was established, employing a PID feedback control strategy without residual frequency domain decoupling, the parameters of which were tuned using the Ziegler-Nichols method; That is, 5 minutes after the step disturbance occurs, although the original measured data contains noise, the first frequency band residual component extracted by the first filter... It still reflects transient impacts, and the state observer uses this to inversely calculate the equivalent disturbance parameters. convergence to , and the actual applied With loads close together, feedforward compensation based on this parameter suppresses the temperature deviation to 0.81. The control group, due to its delayed response, experienced a maximum temperature deviation of [missing value]. In the later stages of the experiment At that time, due to thermal resistance The continuous decline occurred, and the control group showed The steady-state static error; this method extracts the non-zero mean second-band residual component through a second filter. The driving parameter identification module identifies the thermal resistance parameters inside the model. Revised to To eliminate model bias, the temperature control error was maintained at 0.05. When micro-perturbation signals are injected, the calculated consistency index Stay The above experimental data records are shown in Table 1.

[0036] Table 1: Comparison of Control Performance under Dynamic Load Disturbance and Parameter Drift

[0037]

[0038] Experimental results show that by frequency domain decoupling and inverse analysis of the state residual vector, this method can separate and quantify external transient loads and internal parameter drift in noisy environments. The technical effect is reflected in the suppression of step disturbances and the maintenance of convergence of the control target during the nonlinear decay process of changes in physical entity characteristics.

[0039] Example 3: This example combines Figures 1 to 3 This describes a smart office space management method based on digital twin technology, such as... Figure 1 As shown, the system collects measured status data through environmental monitoring devices. Simultaneously, theoretical state data are calculated using a lumped parameter network model. These two data streams converge into the state residual calculation module to generate the state residual vector E, which then enters the dual-band filter for frequency domain decoupling. This process splits the signal into two paths. One path is the high-frequency component characterizing transient load disturbance, i.e., the first frequency band residual component. This component is input to the state observer to reverse analyze the equivalent disturbance parameters, which are then used for advance extrapolation and command generation. By injecting disturbance parameters to predict the trajectory, the environmental control equipment is guided to execute control commands and achieve hot and cold load compensation. This path also includes the active verification step of the control loop shown in the dashed box to perform micro-disturbance signal consistency analysis. The other decoupling signal is the low-frequency component characterizing structural parameter drift, i.e., the second frequency band residual component. This component is input to the parameter identification module to correct the structural parameter matrix, and the updated parameter matrix is ​​fed back to the lumped parameter network model to complete the adaptive closed-loop control of the system.

[0040] like Figure 2 As shown, the horizontal axis represents time, and the vertical axis represents the residual value, with units of _____. In the figure, the solid line represents the original state residual vector E, which exhibits a peak in the early stage of disturbance, and the dashed line represents the high-frequency components separated by filtering. The waveform closely follows the rapidly changing portion of the total residual and quickly falls back after the disturbance ends. The dotted line represents the low-frequency component. The changes were gradual and exhibited a non-zero offset trend over a 95-minute time span; for example... Figure 3 As shown, the system is mainly supported by six branches: in terms of physical sensing construction, it covers multi-dimensional sensor sampling and the establishment of virtual thermodynamic nodes; in terms of digital model foundation, it relies on lumped parameter network model and state space equation Ax+Bu; in terms of core decoupling algorithm, it utilizes state residual vector E, dual-band filter and frequency domain orthogonality; and in terms of inverse predictive control, it includes state observer Kalman and equivalent disturbance parameters. The analysis and advanced inference feedforward compensation are used to achieve adaptive model updates and aging / efficiency decay correction through least squares algorithm identification in terms of structural parameter evolution. In terms of safety and reliability, it integrates abnormal working condition circuit breaker protection, consistency index R calculation and active verification mechanism based on micro-disturbance.

[0041] Example 4: Addressing the issues of model parameter adaptation and structural drift during long-term operation, this example provides an online self-calibration procedure for model parameters based on dual-time-domain residual spectrum decoupling. This eliminates ambiguity regarding the implementation details of the model update mechanism and verifies its effectiveness in dealing with slow-variable disturbances such as equipment aging. In a typical long-term office space operation scenario, the thermal resistance of the building envelope... Heat exchange efficiency of air conditioning terminals The system exhibits a slow, nonlinear decay characteristic over time. To achieve accurate identification and correction of these structural parameters, the system establishes a state residual vector... The frequency domain analysis logic uses two parallel digital filters: the first filter is a high-pass filter with a cutoff frequency of... Set as Used to extract high-frequency components characterizing transient load disturbances. The second filter is a low-pass filter with a cutoff frequency of [missing information]. Set as Used to extract low-frequency trend terms characterizing structural parameter drift. In the model's inverse dynamics solution and adaptive correction of structural parameters, the state observer adopts an augmented state vector form containing thermal and moisture load disturbance terms, and the process noise covariance matrix... Covariance matrix of observation noise The ratio is set at to The interval is used to iterate in real time using the Kalman gain matrix, so that the transient components of the state estimation residual are within the range. to Each sampling period converges to zero, locking in a unique corresponding equivalent perturbation parameter value; in the parameter identification stage, the recursive least squares algorithm is used to determine the forgetting factor. Set as to Tracking time-varying characteristics, extracting the structure drift component, and the cutoff frequency of the second filter. Let the physical space principal thermal time constant be... thermal resistance With heat capacity reciprocal of product Below a factor of 1, the cutoff frequency of the first filter is used to extract the transient load component. Set as reciprocal More than twice, to construct a frequency domain stopband isolation zone to block high-frequency random load noise from penetrating into the low-frequency structural parameter matrix values.

[0042] When the system is in steady state and there is no human disturbance, that is... When the amplitude is lower than the preset noise threshold, the parameter identification process is initiated, and the system cumulative length is... of Data sequence, in which The value is the number of sampling points covering the complete day-night cycle, constructing a parameter matrix of the structure to be identified. and The sensitivity equation is solved using a recursive least squares algorithm with a forgetting factor to achieve the objective function. Minimize the parameter adjustment amount and ,in The forgetting factor has a value of to The calculated correction amount and It is not used directly to update the model, but rather, after a physical rationality constraint verification process, the system pre-defines the physical feasibility domain of the thermal resistance of the building envelope and the heat exchange efficiency of the equipment. and If the calculated new parameter value exceeds the feasible region, the correction amount is truncated to the boundary value and an anomaly alarm is triggered. The correction amount that passes the verification is used to update the corresponding parameters in the lumped parameter network model, completing one self-evolution of the model. This process is executed periodically to ensure that the digital twin model maintains its mapping to the physical entity throughout its entire life cycle.

[0043] Example 5: This example provides a standardized offline calibration and data filling procedure. After the system is first deployed or undergoes major physical modifications, a structural parameter calibration experiment based on controlled thermal excitation is performed. This experiment requires that the test space be sealed during non-working periods and all environmental control equipment except for the test heat source be turned off. A set of standard electric heat source arrays with adjustable power is used to apply a pseudo-random binary sequence thermal power signal with a specific frequency and amplitude as the excitation input within the space. Simultaneously, a high-precision temperature sensor array is used to synchronously collect temperature response data at each virtual node location. The data collection duration covers at least three times the system's main time constant. The cycle is adjusted to ensure the full activation of the system's dynamic characteristics, and the subspace identification algorithm is used to process the input and output data pairs. The state space matrix of the lumped parameter network model can be directly estimated. and The initial values ​​are determined by the calibration process, which does not rely on prior thermal design parameters but is based on field measured data.

[0044] This embodiment further provides a pre-deployment calibration procedure. Before the system is officially put into operation, a 48-hour baseline calibration run is performed. During this period, the system operates in open-loop observation mode, without outputting control commands, and only calculates the equivalent disturbance parameters in real time through the state observer. Statistics during this period The probability density distribution, according to Confidence intervals determine steady-state thresholds Amplitude limits and steady-state thresholds for micro-perturbation signals The calibration procedure is set to three times the standard deviation of the background thermal noise, and the micro-perturbation amplitude is set to the minimum control quantity that can cause a temperature change greater than the sensor resolution but less than the human perception threshold. This calibration procedure ensures that the system can adapt to the optimal fault detection sensitivity in different application environments.

[0045] Example 6: This example provides a standardized pre-deployment calibration and model building procedure. This procedure must be executed after the initial deployment of the system or after major physical modifications. The procedure requires that the test space be sealed off during non-working hours and all environmental control equipment except for the test heat source be turned off. A set of standard adjustable-power electric heat source arrays is used to apply a pseudo-random binary sequence heat power signal of a specific frequency and amplitude as the excitation input within the space. Simultaneously, a high-precision temperature sensor array is used to synchronously collect temperature response data at each virtual node location. The data collection duration must cover at least three times the system's main time constant. The cycle is adjusted to ensure the full activation of the system's dynamic characteristics, and the subspace identification algorithm is used to process the input and output data pairs. The state space matrix of the lumped parameter network model can be directly estimated. and The initial values ​​are determined by a calibration process that does not rely on prior thermal design parameters but is built entirely based on field measured data, thus providing an accurate physical reference for subsequent digital twin operation.

[0046] To address the control parameter adaptation issues caused by differences in the thermal inertia of different building envelopes, this embodiment further provides a pre-deployment calibration process. Before the system is officially put into operation, a 48-hour baseline calibration run is performed. During this period, the system operates in open-loop observation mode, without outputting control commands, and only calculates equivalent disturbance parameters in real time through the state observer. Statistics during this period The probability density distribution, according to Confidence intervals determine steady-state thresholds Amplitude limits and steady-state thresholds for micro-perturbation signals The perturbation amplitude is set to three times the standard deviation of the background thermal noise, while the perturbation amplitude is set to the minimum control quantity that can cause a temperature change greater than the sensor resolution but less than the human perception threshold. The control loop actively verifies the signal, and the perturbation control signal injection follows dual constraints of signal-to-noise ratio and comfort, with the amplitude limited to the standard deviation of the environmental monitoring sensor's background noise. of to The amplitude absolute value remains at the human body temperature perception threshold. to of Within a certain range, the system's dynamic response is stimulated without causing changes in the thermal sensation of those present; simultaneous data collection of disturbances injected into the physical actuator and digital twin model is also performed. For step response data sequences within a time constant, calculate the normalized cross-correlation coefficient between them. The cross-correlation coefficient is consistently lower than... When a preset threshold is set, if an actuator mechanical jamming or sensor zero-point drift fault is detected, the current control output is frozen and a fault maintenance log is generated. This completes non-intrusive online loop health diagnosis, ensuring that the system can automatically adapt to the optimal fault detection sensitivity and control stability in different application environments.

[0047] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A smart office space management method based on digital twin technology, characterized in that, The method includes the following steps: A lumped parameter network model of the physical office space is constructed. This lumped parameter network model defines a structural parameter matrix that describes the heat transfer characteristics between nodes. The theoretical state data of the physical office space is input based on the boundary conditions. The measured state data of the environmental monitoring device in the physical office space is collected in the first cycle, and the theoretical state data of the lumped parameter network model at the current moment is obtained simultaneously. The state residual vector between the two is calculated. Configure a first filter and a second filter with different cutoff frequencies, and input the state residual vector into the first filter and the second filter respectively to obtain the first frequency band residual component characterizing transient load disturbance and the second frequency band residual component characterizing structural parameter drift; The first frequency band residual component is input into the state observer, and the equivalent perturbation parameters inside the lumped parameter network model are solved using the principle of model inverse dynamics. The steps of solving the equivalent perturbation parameters inside the lumped parameter network model using the principle of model inverse dynamics include: constructing a Kalman filter or Runge-Kutta observer containing extended state variables, and inputting the first frequency band residual component as a correction term into the observer; in the iterative calculation of the observer, the heat source term of the lumped parameter network model is regarded as an unknown variable to be identified, and the virtual heat source value that can generate the current state residual vector is inversely approximated by minimizing the cost function of the first frequency band residual component; the converged virtual heat source value is determined as the equivalent perturbation parameter, which is a value representing the dynamic heat and humidity load that is not directly measured in the physical office space; and the equivalent perturbation parameter is injected into the lumped parameter network model as a boundary condition to deduce the environmental state evolution trajectory within a future preset time window. The second frequency band residual component is input into the parameter identification module, and the least squares algorithm is used to correct the structural parameter matrix of the lumped parameter network model. When the environmental state evolution trajectory exceeds the preset target range, the control compensation amount at the current moment is calculated based on the prediction results of the lumped parameter network model, and the control command is generated and sent to the environmental control equipment.

2. The smart office space management method based on digital twin technology according to claim 1, characterized in that, The steps for constructing a lumped parameter network model of a physical office space include: discretizing the physical office space into multiple virtual thermodynamic nodes, and determining the thermal resistance and heat capacity parameters connecting each virtual thermodynamic node based on the geometric dimensions of the physical office space and the thermophysical properties of the enclosure structure; establishing a set of linear differential equations describing the heat transfer and temperature dynamic evolution between each virtual thermodynamic node based on the law of conservation of energy, and transforming the set of linear differential equations into a state-space equation form as the lumped parameter network model.

3. The smart office space management method based on digital twin technology according to claim 1, characterized in that, The steps for correcting the structural parameter matrix of the lumped parameter network model include: setting a structural parameter correction period, which is longer than the first period; accumulating the data sequence of the second frequency band residual components within each structural parameter correction period; constructing a sensitivity equation for the structural parameter matrix and solving for the correction amount of the structural parameter matrix using a recursive least squares algorithm with a forgetting factor; and updating the thermal resistance and thermal capacity values ​​in the structural parameter matrix using the correction amount when the magnitude of the correction amount exceeds a preset safety threshold.

4. The smart office space management method based on digital twin technology according to claim 1, characterized in that, The method also includes an active verification step for the control loop, monitoring the amplitude of the state residual vector, and when the amplitude remains below a steady-state threshold for a preset period, generating a micro-perturbation control signal and simultaneously applying it to the environmental control equipment and the lumped parameter network model; and acquiring the actual dynamic response sequence of the environmental monitoring device to the micro-perturbation control signal. And the theoretical dynamic response sequence of the lumped parameter network model And calculate the consistency index of the two according to the following formula. : ,in, This is the mean of the actual dynamic response sequence. The mean of the theoretical dynamic response sequence; when the consistency index If the value falls below the preset confidence threshold, the control loop of the physical office space is determined to be faulty, the generation of control commands is stopped, and a fault indicator is output.

5. A smart office space management method based on digital twin technology according to claim 1, characterized in that, The steps for extrapolating the environmental state evolution trajectory within a future preset time window include: using the measured state data at the current moment as the initial condition and the set containing equivalent perturbation parameters as the boundary condition, iteratively solving the lumped parameter network model in the virtual time domain with a step size faster than the passage of physical time; generating time series data containing temperature, humidity and air quality indicators at each future moment as the environmental state evolution trajectory.

6. The smart office space management method based on digital twin technology according to claim 1, characterized in that, The steps for calculating the control compensation amount at the current moment include: constructing a target functional containing energy consumption cost and comfort penalty terms based on the deviation between the environmental state evolution trajectory and the preset target range; solving the control input sequence that minimizes the target functional under the premise of satisfying the environmental control equipment action constraints; and selecting the first element in the control input sequence as the control compensation amount at the current moment.

7. A smart office space management method based on digital twin technology according to claim 1, characterized in that, The passband frequency of the first filter is set to cover the frequency band of heat load changes caused by personnel movement, and the passband frequency of the second filter is set to cover the frequency band of parameter drift caused by changes in the thermal performance of the building envelope and the attenuation of equipment efficiency.

8. A smart office space management method based on digital twin technology according to claim 1, characterized in that, The method also includes a spatial heat load distribution reconstruction step based on equivalent disturbance parameters. Using the topology of the lumped parameter network model, the solved equivalent disturbance parameters are mapped to the corresponding virtual thermodynamic nodes. Based on the amplitude of the equivalent disturbance parameters of each virtual thermodynamic node, a heat load distribution heat map of the physical office space is generated. The heat load distribution heat map represents the regional personnel density or equipment heating intensity.

9. A smart office space management method based on digital twin technology according to claim 1, characterized in that, The method also includes a circuit breaker protection step for abnormal operating conditions, real-time monitoring of the rate of change of equivalent disturbance parameters; when the rate of change exceeds the preset physical limit threshold, it is determined that the state residual vector contains non-physical interference signals; the current boundary conditions of the lumped parameter network model remain unchanged, and the injection of equivalent disturbance parameters into the lumped parameter network model is temporarily blocked until the rate of change falls back to within the physical limit threshold.

Images