Intelligent building environment adaptive regulation and control method based on digital twinning and AI

Abstract

The invention provides a smart building environment adaptive regulation and control method based on digital twinning and AI, and relates to the technical field of artificial intelligence, and the method comprises the following steps: extracting dynamic features based on multi-source system data after time alignment, so as to construct a dynamic feature set; inputting multi-source system data into the digital twin, performing iterative simulation on key environment parameters of a plurality of time steps in the future by taking a current building equipment state as a boundary condition, and outputting a multi-step prediction sequence; inputting into a model prediction controller, and solving to obtain an equipment regulation and control instruction sequence in a future first preset duration by taking the lowest total energy consumption in a prediction period as an optimization target and taking the condition that the key environment parameters do not exceed a comfort interval as a constraint condition; and the first equipment regulation and control instruction in the equipment regulation and control instruction sequence is issued to the field execution mechanism, so that self-adaptive regulation and control of the building environment are realized, the defect that passive response is performed after environmental parameters exceed the standard is overcome, and the quick response capability of the system to sudden environmental changes is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a smart building environment adaptive regulation method based on digital twinning and AI. BACKGROUND

[0002] The existing smart building operation analysis technology, such as the scheme disclosed in CN120123757B, has made progress in multi-source data fusion and causal diagnosis, but its core limitation is the lack of real-time prediction and forward-looking regulation capability for building environment changes. The specific performance is as follows: Through hidden variable modeling and transfer entropy analysis, the causal chain of people flow density CO2 concentration HVAC energy consumption can be effectively identified, and the surge of people flow density is located as the root cause, but this analysis is essentially a post-diagnosis mechanism. When the system identifies the root cause, the environmental parameters (such as CO2 concentration) have often exceeded the standard, and the HVAC system can only respond passively, resulting in regulation lag and sharp increase in energy consumption.

[0003] More importantly, this method cannot cope with sudden environmental changes caused by building thermal inertia, people flow impact conduction, etc. For example: During the morning rush hour, the heat load and CO2 emissions caused by the sudden increase in people flow will not be immediately reflected, but will be delayed and conducted with the building thermal inertia; The traditional control system will start regulation only when the temperature or CO2 concentration actually exceeds the standard, at which time the cooling load has already accumulated to the peak, causing a sharp increase in energy consumption and a decrease in comfort.

[0004] Therefore, it is difficult to meet the rapid response and energy efficiency optimization needs of smart buildings to sudden environmental changes. SUMMARY

[0005] In view of the above defects or deficiencies in the prior art, the present application aims to provide a smart building environment adaptive regulation method based on digital twinning and AI, comprising the following steps: Based on the time-aligned multi-source system data, dynamic features are extracted to construct a dynamic feature set; the multi-source system data includes environmental data, device operation data and people flow data; the dynamic features include thermal inertia delay features and people flow impact conduction features; Input the multi-source system data into the digital twin, and iteratively simulate the key environmental parameters at future time steps with the current building device state as the boundary condition, and output a multi-step prediction sequence; the key environmental parameters include indoor temperature and CO2 concentration; Input the multi-step prediction sequence into the model predictive controller, take the minimum overall energy consumption in the prediction period as the optimization objective, and take the key environmental parameters not exceeding the comfort interval as the constraint condition, and solve to obtain the device regulation instruction sequence in the first preset time in the future; The first device regulation instruction in the device regulation instruction sequence is sent to the field execution mechanism to realize adaptive regulation of the building environment. The digital twin is a physical information neural network constructed by fusing the dynamic feature set and a building information physical model; the building information physical model is obtained by discretizing a thermodynamic differential equation, and the loss function of the physical information neural network includes a data fitting term and a physical equation constraint term.

[0006] According to the technical scheme provided in the present application, based on the time-aligned multi-source system data, the thermal inertia delay feature is extracted, including the following steps: Based on the environmental temperature sequence and the HVAC power sequence in the target historical time period, the mutual information value under different time offsets is calculated, and the time offset that maximizes the mutual information value is taken as the thermal inertia delay time of the building; For each collection time, the thermal inertia delay feature is obtained from the real-time environmental temperature collected at the collection time and the HVAC energy consumption delayed by the thermal inertia delay time at the collection time.

[0007] According to the technical scheme provided in the present application, the mutual information value is obtained by the following formula: ; Wherein, represents the mutual information value; represents the time offset; T represents the environmental temperature sequence; t represents the real-time environmental temperature; p represents the HVAC power sequence; represents the temperature value t i of the i-th collection time and the delayed power value of the j-th collection time; represents the temperature edge probability; represents the delayed power edge probability; N represents the total number of data of the environmental temperature sequence T; M represents the total number of data of the HVAC power sequence p.

[0008] According to the technical scheme provided in the present application, before calculating the mutual information value under different time offsets based on the environmental temperature sequence and the HVAC power sequence in the target historical time period, the following steps are included: Obtain the mode feature vector of the current day to determine the building operation mode category to which the current day belongs; the mode feature vector includes the daily average outdoor temperature, the daily HVAC energy consumption peak value, and the people flow density distribution histogram; According to the building operation mode category to which the current day belongs, the corresponding target historical time period is obtained.

[0009] According to the technical scheme provided in the application, the mode feature vector of the current day is obtained, and the building operation mode category to which the current day belongs is determined, including the following steps: The mode feature vector of the current day is obtained, the historical mode database is called and traversed, the building operation mode category corresponding to the mode feature vector of the current day is obtained, and the building operation mode category is taken as the building operation mode category to which the current day belongs. The historical mode database includes a plurality of historical time periods and the building operation mode category corresponding to each historical time period.

[0010] According to the technical scheme provided in the application, the method further includes the following steps: While the model predictive controller is solving the device control instruction sequence, the expected operation state of a key device when the device control instruction sequence is executed is simulated synchronously based on the digital twin, and the expected operation state includes the expected load rate of a compressor and the expected rotating speed of a fan. After the device control instruction sequence in the future for a first preset time period is solved, the following steps are further included: It is determined whether the expected operation state of any key device in the first preset time period continuously exceeds the safe operation boundary thereof. The first device control instruction in the device control instruction sequence is issued to an on-site execution mechanism, including the following steps: If not, the first device control instruction in the device control instruction sequence is issued to the on-site execution mechanism.

[0011] According to the technical scheme provided in the application, after it is determined whether the expected operation state of any key device in the first preset time period continuously exceeds the safe operation boundary thereof, the following steps are further included: If yes, the affected environmental control loop is identified, and the control right of the environmental control loop is dynamically reconstructed to a reconstructed system architecture in a virtual environment based on the digital twin, the reconstructed system architecture including one or more healthy backup devices or redundant control loops; The model predictive controller is re-run with the reconstructed system architecture as an updated boundary condition, and an alternative device control instruction sequence in the future for a second preset time period is solved. The first device control instruction in the alternative device control instruction sequence is issued to the on-site execution mechanism, and a device warning and system reconstruction report are synchronously generated.

[0012] According to the technical scheme provided in the application, the control right of the environmental control loop is dynamically reconstructed to the reconstructed system architecture in the virtual environment based on the digital twin, including the following steps: generating at least two alternative reconfiguration strategies of allocating the load of the environment regulation loop to healthy backup devices or redundant control loops based on the system-based device topology relationship; for each of the alternative reconfiguration strategies, performing forward simulation in the digital twin to preview the process of executing the alternative device regulation instruction sequence and calculating a preview performance index of each of the alternative reconfiguration strategies, the preview performance index including the impact strength of the switching process on comfort, additional energy consumption cost, and device life loss coefficient; based on the preview performance index, comprehensively scoring each of the alternative reconfiguration strategies; obtaining the reconfigured system architecture according to the alternative reconfiguration strategy with the highest score.

[0013] According to the technical scheme provided in the application, obtaining the reconfigured system architecture according to the alternative reconfiguration strategy with the highest score comprises the following steps: based on the digital twin, performing stress testing on the alternative reconfiguration strategy with the highest score under multiple preset disturbance scenarios to obtain the performance retention degree of the alternative reconfiguration strategy under each of the disturbance scenarios, the preset disturbance scenarios including sudden fluctuation of human flow load and / or step change of outdoor meteorological parameters, and the performance retention degree being the time proportion that still meets the comfort constraint under disturbance; if the performance retention degree of the alternative reconfiguration strategy under each of the disturbance scenarios meets the standard, the architecture corresponding to the alternative reconfiguration strategy with the highest score is taken as the reconfigured system architecture.

[0014] According to the technical scheme provided in the application, after obtaining the performance retention degree of the alternative reconfiguration strategy under each of the disturbance scenarios, the following steps are further included: if the performance retention degree of the alternative reconfiguration strategy under any of the preset disturbance scenarios does not meet the standard, based on the characteristics of the preset disturbance scenario that does not meet the standard, the key device and its action period that cause performance degradation are reversely deduced in the digital twin; based on the device regulation instruction sequence obtained in the previous control cycle, a dynamic compensation instruction sequence for the key device in the action period is generated, the dynamic compensation instruction sequence including a gradual adjustment curve of device output power and switching timing optimization; the dynamic compensation instruction sequence and the alternative reconfiguration strategy are fused to form an enhanced reconfiguration strategy; the enhanced reconfiguration strategy is verified in the digital twin, and after verification, the architecture corresponding to the enhanced reconfiguration strategy is determined as the reconfigured system architecture.

[0015] Compared with the prior art, the application has the following beneficial effects: I. By constructing a digital twin that fuses a physical information neural network, the system can perform high-precision iterative simulation of key parameters such as temperature and CO2 concentration at multiple time steps in the future based on the current device state and multi-source dynamic characteristics. This enables the system to go beyond post-hoc causal analysis and have multi-step prediction capabilities for environmental changes, providing decision-making basis for advanced control.

[0016] II. By extracting thermal inertia delay features and human flow impact conduction features, the system can accurately depict the dynamic response process of the building environment to sudden changes. Combined with the prediction output of the digital twin, the model predictive controller can adjust the device operating state in advance before the thermal load / CO2 concentration accumulates to the peak value, fundamentally solving the control delay problem caused by building thermal inertia and human flow impact conduction.

[0017] III. The multi-step prediction sequence output by the digital twin is input into the model predictive controller to minimize overall energy consumption during the prediction period while ensuring that environmental parameters do not exceed the comfort interval. The optimal device control instruction sequence is obtained and executed in real time, forming a complete intelligent control closed loop and achieving adaptive and fine control of the building environment.

[0018] IV. By combining physical information neural networks with model predictive control, the system maintains reliability based on physical mechanisms while also having adaptability driven by data. It can dynamically optimize device operation strategies while meeting indoor environmental comfort, avoid energy consumption peaks, and has more direct energy-saving benefits and engineering application value than pure analysis solutions.

[0019] V. For early morning peak and other human flow mutation scenarios, the system predicts environmental parameter trends in advance through the digital twin and generates optimal control instructions through the model predictive controller, effectively avoiding the drawbacks of traditional methods that respond passively after environmental parameters exceed the standard, significantly improving the system's rapid response capability to sudden environmental changes. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A step flowchart of the intelligent building environment adaptive control method based on digital twin and AI provided in the present application. DETAILED DESCRIPTION

[0021] The present application will be further described in detail below in conjunction with the drawings and examples. It should be understood that the specific examples described herein are only intended to explain the related invention, and not to limit the invention. In addition, it should be noted that only parts related to the invention are shown in the drawings for ease of description.

[0022] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0023] Embodiment 1 As mentioned in the background, in order to solve the problems in the prior art, the present application proposes a smart building environment adaptive regulation method based on digital twinning and AI, as shown in Figure 1 The method comprises the following steps: S1, based on the time-aligned multi-source system data, extracting dynamic features to construct a dynamic feature set; the multi-source system data includes environmental data, equipment operation data and people flow data; the dynamic features include thermal inertia delay features and people flow impact conduction features; Specifically, first, the environmental data (temperature, humidity, CO2 concentration, light intensity), equipment operation data (such as HVAC system instantaneous power, lighting power, equipment current, etc., which are used to calculate energy consumption and related features) and people flow data (density, speed, residence time) are collected by the sensor network deployed in the building. Since there may be differences in sampling frequency and clock of different sensors, time alignment processing is needed. The time axis mapping method based on weighted least squares is adopted to calculate the average time difference for each type of sensor data, and the best alignment time point is found through nonlinear optimization to form an alignment matrix of unified time reference. Thermal inertia delay feature: reflects the temperature response lag phenomenon caused by the heat storage capacity of building envelopes, furniture, etc. By analyzing the mutual information peak value of historical temperature sequence and HVAC power sequence, the building-specific thermal inertia delay time is determined. People flow impact conduction feature: quantifies the conduction effect of people flow mutation on environmental parameters. By calculating the cross-correlation between people flow density change rate and CO2 concentration rise rate, the conduction strength and delay time are extracted. Building environment is a complex system with multiple time scales coupled. Thermal inertia causes temperature regulation to have a delay of minutes to hours, while the impact of people flow shock shows spatial conduction characteristics. By extracting these dynamic features, the transient response behavior of the system can be more accurately described.

[0024] S2, inputting the multi-source system data into the digital twin, taking the current building equipment state as the boundary condition, iteratively simulating the key environmental parameters at multiple time steps in the future, and outputting a multi-step prediction sequence; the key environmental parameters include indoor temperature and CO2 concentration; Specifically, the digital twin adopts a physics-informed neural network architecture, which is the core of deep fusion of building information-physical model and data-driven method. The building information-physical model is obtained by discretizing the thermodynamic differential equation, which describes the physical processes of building envelope heat transfer, air flow, pollutant diffusion, etc. The loss function of the physics-informed neural network contains a data fitting term (minimizing the difference between the predicted value and the measured value) and a physical equation constraint term (ensuring that the predicted result conforms to the physical law). The physics-informed neural network introduces the physical equation as a regularization term into the loss function, ensuring that even in the sparse data area, it can produce physically reasonable prediction results. This fusion method not only maintains the flexibility of data-driven models, but also inherits the generalization ability of physical models.

[0025] Specific operation: With the current building equipment state (such as air conditioner set temperature, fan speed, fresh air valve opening) as the boundary condition, input the multi-source system data and dynamic feature set into the digital twin, and generate a prediction sequence of key environmental parameters (such as indoor temperature and CO2 concentration) in multiple time steps (such as 2 hours in the future, step 15 minutes) in the future through iterative simulation.

[0026] S3, input the multi-step prediction sequence into the model predictive controller to predict the equipment control instruction sequence in the first preset time in the future, with the minimum overall energy consumption in the prediction period as the optimization target, and the key environmental parameters not exceeding the comfort interval as the constraint condition; Specifically, the model predictive controller takes the minimum overall energy consumption in the prediction period as the optimization target, while constraining the key environmental parameters not to exceed the comfort interval (such as temperature 22-26℃, CO2 concentration <1000ppm). The optimization problem can be expressed as a quadratic programming problem with constraints, which is solved by the interior point method or the efficient set method. The specific process is as follows: the controller solves the optimal equipment control instruction sequence in the first preset time (such as 1 hour in the future) based on the current state and future prediction at each decision-making time, including air conditioner set temperature adjustment, fresh air volume control instruction, lighting dimming level, etc. The model predictive control adopts a rolling optimization strategy, which re-solves the finite-time optimization problem at each control period and only implements the first control instruction. This strategy can compensate for model errors and external disturbances in a timely manner, achieving closed-loop optimization control.

[0027] S4, the first equipment control instruction in the equipment control instruction sequence is issued to the field execution mechanism to realize adaptive regulation of the building environment. Among them, the digital twin is a physics-informed neural network constructed by fusing the dynamic feature set with the building information-physical model; the building information-physical model is obtained by discretizing the thermodynamic differential equation, and the loss function of the physics-informed neural network contains a data fitting term and a physical equation constraint term.

[0028] Specifically, the solved device regulation instruction sequence is issued to the field execution mechanism through a building automation system protocol (such as BACnet or Modbus). Only the first instruction in the sequence is implemented, and then the prediction and optimization are performed again in the next control cycle.

[0029] In a preferred embodiment, based on the time-aligned multi-source system data, a thermal inertia delay feature is extracted, including the following steps: Based on the environmental temperature sequence and the HVAC power sequence in the target historical time period, the mutual information values at different time offsets are calculated, and the time offset that maximizes the mutual information value is taken as the thermal inertia delay time of the building; For each collection time, the real-time environmental temperature collected at the collection time and the HVAC energy consumption delayed by the thermal inertia delay time of the collection time are used to obtain the thermal inertia delay feature.

[0030] Specifically, the data preparation stage: collect the environmental temperature sequence and the HVAC power sequence in the target historical time period (such as the past 30 days). The data needs to be time-aligned and abnormal value processed to ensure data quality. The mutual information calculation stage: calculate the mutual information value of the environmental temperature sequence and the delayed HVAC power sequence at different time offsets (such as 0-120 minutes, step 5 minutes). The mutual information measures the statistical dependence between two variables. Find the time offset τ1 that maximizes the mutual information value, which is the thermal inertia delay time of the building. This process is achieved by scanning the mutual information at different τ values to identify the peak position. For each collection time, the real-time environmental temperature at that time is extracted, and the HVAC energy consumption value after a delay of τ1 time is found, and these two values are combined to form the thermal inertia delay feature.

[0031] Further, the mutual information value is obtained by the following formula: ; Wherein, represents the mutual information value; represents the time offset; T represents the environmental temperature sequence; t represents the real-time environmental temperature; p represents the HVAC power sequence; represents the temperature value t i at the i th collection time and the delayed power value at the j th collection time; represents the temperature marginal probability; represents the delayed power marginal probability; N represents the total number of data of the environmental temperature sequence T; M represents the total number of data of the HVAC power sequence p.

[0032] Specifically, data discretization: discretize the continuous ambient temperature sequence T and HVAC power sequence P into a finite number of intervals. Temperature can be discretized at 0.5 °C intervals, and energy consumption can be discretized at percentage of rated power. The degree of discretization needs to be balanced between computational accuracy and computational complexity. Probability distribution estimation: marginal probability distribution p(t): the frequency of statistical temperature values falling in each discrete interval; marginal probability distribution p(pτ): the frequency of statistical energy consumption values after delay τ falling in each discrete interval; joint probability distribution p(t, pτ): the frequency of statistical temperature t and energy consumption pτ after delay τ appearing in the corresponding interval at the same time; mutual information calculation: for each time offset τ, calculate the mutual information value according to the above probability distribution. When calculating, pay attention to the processing of zero probability, and usually use Laplace smoothing technique to avoid numerical problems. Maximum mutual information determination: among all candidate τ values, select the τ value that makes I(τ) maximum as the thermal inertia delay time. It needs to be further explained that the cumulative concept of energy consumption data, total energy consumption, etc. is calculated by integration or accumulation based on device operation data (such as power data); while performing dynamic feature extraction, its instantaneous or sequence data (such as power sequence) is used; Exemplarily, this example demonstrates the complete operation process of an intelligent office building environment adaptive regulation system. The system is deployed in a 50,000 square meter office building, focusing on solving the problem of environmental quality control and energy consumption optimization during the morning peak period. System architecture and data foundation: The system builds a complete sensing network and computing platform. The sensing network includes 86 temperature and humidity sensors, 45 CO2 concentration sensors, 12 HVAC system energy consumption monitoring ammeters, and 28 intelligent human flow statistical cameras based on computer vision. The computing platform consists of edge computing nodes and cloud digital twin platforms, with edge nodes responsible for real-time data preprocessing and control instruction execution, and cloud platforms running physical information neural networks and model predictive control algorithms. Morning peak preparation phase: In the preparation phase from 7:00 to 7:30 in the morning, the system first performs time alignment processing of multi-source data. Sensor data is unified to a standard time reference through a time axis mapping function, and abnormal data points such as zero drift of a temperature sensor are automatically identified and repaired. Then the system performs dynamic feature extraction. Through mutual information analysis, the thermal inertia delay feature is calculated, and the optimal delay time of the building is determined to be 45 minutes. The system combines the current temperature of 23.5°C with the HVAC power data 45 minutes ago to form the thermal inertia delay feature. At the same time, based on historical data analysis of the morning peak human flow pattern, the human flow impact conduction feature is extracted to predict that there will be a core area human flow peak from 8:15 to 8:45. The digital twin body makes multi-step predictions based on the current building state and external conditions. Input the current indoor temperature of 23.5°C, CO2 concentration of 450 ppm, HVAC energy-saving operation mode, and weather forecast of today's maximum temperature of 32°C, and output the predicted sequence of future environmental parameters. The prediction shows that from 9:00, the CO2 concentration will exceed the upper limit of 1000 ppm for comfort, and will reach 1250 ppm at 9:15. Model predictive control optimization: The model predictive controller takes the minimum total energy consumption in the next 2 hours as the optimization goal, while constraining the temperature not to exceed 26°C and the CO2 concentration not to exceed 1000 ppm. By solving the constrained optimization problem, the optimal device regulation instruction sequence is generated: set the air conditioning temperature to 24.5°C, the fresh air valve opening to 65%, and the fan speed to 75% at 8:00; adjust to 24.0°C, 80% opening, and 85% speed at 8:30; further adjust to 23.5°C, 95% opening, and 95% speed at 9:00. Peak period regulation execution: During actual execution, the system shows good forward-looking regulation capability. At 7:45, the system adjusts the core area air conditioning set temperature to 24.5°C in advance; at 8:15, when the human flow density reaches 0.3 people per square meter, the fresh air volume is increased to 80% in advance; at 8:40, when the CO2 concentration rises to 850 ppm, due to the advance regulation measures, the concentration has always been below the 1000 ppm threshold. The system also shows flexibility in adaptation.When the actual flow is 15 minutes later than the prediction to reach the peak, the digital twin updates the prediction model in real time, and the model predictive controller re-optimizes the subsequent instructions, appropriately reducing the fan speed after 9:00, while ensuring comfort and optimizing energy consumption. During the entire operation process, the system continuously monitors the equipment running state. When the digital twin detects that the compressor load rate is continuously higher than the safety threshold of 85%, it automatically generates an alternative strategy to transfer part of the load to the standby chiller unit to ensure that all equipment is running within the safety range. The special scene coping ability is also relatively improved, and the system also performs well in the midday meeting scene. At 12:00 noon, the system detects through the meeting reservation system that there will be a large meeting in the afternoon, and predicts that the flow density will increase from 0.1 to 0.8 person / m2 during the meeting based on historical data. The system starts progressive pre-cooling at 13:30 to avoid discomfort caused by sudden temperature drop, and the CO2 concentration is stably maintained below 950ppm during the meeting. When the outdoor temperature suddenly rises to 36°C, the digital twin combines weather forecast information to increase the refrigeration capacity 1 hour in advance, and optimizes the equipment start-stop timing through thermal inertia characteristics calculation, effectively avoiding the occurrence of power demand peak.

[0033] In a preferred embodiment, before calculating the mutual information value at different time offsets based on the sequence of ambient temperature and the sequence of HVAC power in the target historical time period, the following steps are included: Obtain the mode feature vector of the current day to determine the building operation mode category to which the current day belongs; the mode feature vector includes daily average outdoor temperature, daily HVAC energy consumption peak, and flow density distribution histogram; According to the building operation mode category to which the current day belongs, the corresponding target historical time period is obtained.

[0034] Specifically, the mode feature vector: a multi-dimensional feature group quantifying the single-day operation state of the building, including: daily average outdoor temperature: the statistical mean of the whole-day temperature, reflecting the external thermal disturbance intensity. Daily HVAC energy consumption peak: the maximum instantaneous power of the HVAC system in the day, indicating the system load limit. Flow density distribution histogram: after dividing 24 hours into equal intervals, the frequency distribution of the average flow quantity in each interval, describing the time sequence rule of internal heat source and CO2 emission. Building operation mode category: a typical operation scene label generated based on historical data clustering, used to distinguish the behavior mode of the building under different external environment and use intensity, such as "summer weekday", "winter weekend", etc. Target historical time period: dynamically selected historical data range according to the current mode category, aiming to ensure the consistency of the data used for thermal inertia calculation with the current conditions.

[0035] The specific implementation is described below: Obtain the mode feature vector of the current day: The system automatically collects and calculates the mode feature vector from multiple data sources every day. Specifically: Daily average outdoor temperature: Collect temperature data throughout the day through a network of temperature sensors outside the building (such as PT100 platinum resistance or digital temperature sensors), with a sampling frequency of every 5 minutes, and calculate the arithmetic mean of all sampling points for the day. This value reflects the overall thermal load background.

[0036] Daily HVAC energy consumption peak: Read the power time series data of the HVAC system (unit: kW) from the smart meter or energy management system, identify the maximum value of energy consumption and its occurrence time for the day. This peak value represents the energy demand of the building under extreme load.

[0037] People flow density distribution histogram: Count the number of people through infrared counters or video analysis systems installed at entrances, and generate a 24-dimensional histogram vector with 1 hour as the time slot, with each element representing the average number of people in that period. This histogram reveals the time regularity of building usage intensity.

[0038] Data fusion: After time alignment (e.g., unified to UTC timestamp), the above data is packaged into a mode feature vector V = [Tavg, Pmax, H1, H2,..., H24], where Tavg is the daily average outdoor temperature, Pmax is the HVAC energy consumption peak, and Hi is the people flow density in the ith hour. Determine the current building operation mode category: The system pre-stores a historical mode database, which is generated by clustering analysis (such as K-means algorithm) on historical data labeling, containing three typical modes: High-efficiency operation mode: Corresponding to weekdays in mild seasons, characterized by moderate outdoor temperature (such as 15-25°C), balanced energy consumption peak, and double-peak people flow distribution (morning and evening peak).

[0039] Extreme load mode: Corresponding to cold / hot days, characterized by low / high temperature (such as <5°C or >30°C), high energy consumption peak, and concentrated people flow (such as midday peak).

[0040] Low load mode: Corresponding to holidays or night, characterized by comfortable temperature, low energy consumption peak, and flat people flow distribution.

[0041] Matching logic: Calculate the Euclidean distance between the current day feature vector and the center vector of each category in the database, and select the category with the smallest distance as the current mode. For example, if the current vector has the smallest distance to the high-efficiency operation mode center, it is classified as this mode.

[0042] Obtain the target historical time period: Implementation: Each building operation mode category is associated with a preset time window strategy: High load mode: select consecutive data with same weather condition within past 7 days, avoid seasonal bias.

[0043] High load mode: select consecutive data with same weather condition within past 7 days, avoid seasonal bias.

[0044] Low load mode: select all weekend or holiday data within past 14 days.

[0045] Dynamic query: system extracts environmental temperature sequence and HVAC power sequence of corresponding time period from time series database as input of calculating thermal inertia delay feature according to determined mode category.

[0046] In a preferred embodiment, the obtaining of the mode feature vector of the current day and the determination of the building operation mode category to which the current day belongs comprises the following steps: obtaining the mode feature vector of the current day, calling and traversing a historical mode database to obtain a building operation mode category corresponding to the mode feature vector of the current day, and taking the building operation mode category as the building operation mode category to which the current day belongs. The historical mode database comprises a plurality of historical time periods and a building operation mode category corresponding to each of the historical time periods.

[0047] Specifically, the mode feature vector of the current day is obtained: a real-time data acquisition system automatically starts feature calculation at 00:00 every day. The daily average outdoor temperature is obtained from a weather station API interface, and the sampling value is averaged after Kalman filter denoising.

[0048] The HVAC energy consumption peak value is extracted from the Modbus / TCP protocol meter reading, and the scanning frequency is 1 time / minute.

[0049] The people flow density histogram is generated by the access control system log, and the in-and-out number is aggregated by hour.

[0050] Vector normalization: to prevent dimensional differences, each feature value is Min-Max normalized, for example, temperature is mapped to the [0, 1] interval.

[0051] Call and traverse the historical mode database: Database structure: the historical mode database is a relational table (such as MySQL), which contains fields: date (date), mode_class (mode category), temp_avg (average temperature), power_max (energy consumption peak value), histogram (histogram JSON).

[0052] The data source is the historical record of the past 3 years, and the mode label is updated once a month by clustering algorithm.

[0053] Traverse matching: the system executes a SQL query to obtain all historical records, and calculates the similarity between the current feature vector and the historical vector line by line. The similarity measure adopts weighted cosine similarity: S = w1-cos(Tavg, Tavg') + w2-cos(Pmax, Pmax') + w3-cos(H, H'); wherein the weights w1=0.4, w2=0.3, w3=0.3 are assigned according to the importance of the features, and cos is the cosine similarity function. The historical record with the highest similarity is selected, and the value of its mode_class field is taken as the current category. The matched category is output to the thermal inertia calculation module to trigger the query of the corresponding target historical time period.

[0054] In a preferred embodiment, the method further comprises the following steps: While the model predictive controller is solving the device control instruction sequence, based on the digital twin, a synchronous simulation of the expected operating state of the key device when executing the device control instruction sequence is performed, and the expected operating state includes the expected load rate of the compressor and the expected speed of the fan; After the solving of the device control instruction sequence for the future first preset time length, the following steps are further included: It is determined whether the expected operating state of any of the key devices in the first preset time length continuously exceeds the safe operating boundary thereof; The first device control instruction in the device control instruction sequence is issued to the field execution mechanism, including the following steps: If not, the first device control instruction in the device control instruction sequence is issued to the field execution mechanism.

[0055] Specifically, the synchronous simulation of the expected operating state of the key device: in the parallel thread of the model predictive controller (MPC) solving the control instruction sequence, the digital twin receives the sequence as input and performs device-level simulation: the expected load rate of the compressor: based on the refrigerant flow model and the start-stop signal in the instruction, the load rate Lc = Qactual / Qrated x 100% is calculated every 5-minute time step, wherein Qactual is the actual refrigeration capacity and Qrated is the rated refrigeration capacity. The expected speed of the fan: through the fan law model N = Qactual / (Pv x D2 x 0.6 x 9.81), the speed N is calculated every 5-minute time step, wherein Pv is the specific volume of the refrigerant, D is the diameter of the fan, and 0.6 and 9.81 are empirical coefficients. (P is input power) to derive the speed curve, and cross-verify with the frequency converter set value in the instruction. Generate the device state sequence completely aligned with the regulation instruction sequence timeline, including load rate time sequence {Lc(t)} and speed time sequence {N(t)}. Determine whether the device state continuously exceeds the safe operation boundary: boundary definition: the safe operation boundary is preset in the device knowledge base: the compressor load rate safe range is [10%, 90%], and >90% for 3 consecutive time steps (15 minutes) is considered "continuous exceedance". The fan speed safe range is [0, 1500 RPM], and an instantaneous value exceeding 1500 RPM triggers an alarm.

[0056] Judgment logic: the system scans the simulation data for the entire first preset duration (e.g. 2 hours in the future): For the compressor: check if there is an interval where the load rate is >90% in the continuous time window.

[0057] For the fan: check if the speed is >1500 RPM at any sampling point.

[0058] If any condition is met, it is determined that "there is continuous exceedance".

[0059] Conditional issuance of regulation instructions: Safe path: if all device states are not out of limits, the first instruction in the device regulation instruction sequence obtained by the MPC solution (i.e. the instruction that should be executed immediately at the current time) is issued to the field PLC actuator through the OPC UA protocol. Abnormal handling: if there is an exceedance, the following reconstruction process is triggered.

[0060] In a preferred embodiment, after determining whether the expected operating state of any of the key devices continuously exceeds its safe operating boundary within the first preset duration, the following steps are further included: If yes, identify the affected environmental regulation loop, and based on the digital twin, dynamically reconstruct the control of the environmental regulation loop to the reconstructed system architecture in the virtual environment, the reconstructed system architecture including one or more healthy backup devices or redundant control loops; Take the reconstructed system architecture as the updated boundary condition, re-run the model predictive controller to obtain an alternative device regulation instruction sequence for a second preset duration in the future; Issue the first device regulation instruction in the alternative device regulation instruction sequence to the field actuator, and simultaneously generate a device warning and system reconstruction report.

[0061] Specifically, environmental regulation loop: refers to the complete control chain that completes the adjustment of a specific environmental parameter, including sensors, controllers, actuors and controlled objects. For example, the temperature regulation loop includes temperature sensors, PID controllers, VAV terminal devices and corresponding space areas. Dynamic reconstruction: the process of adjusting the control architecture in real time according to the device state during system operation, including the transfer of control rights, the redistribution of device roles, etc., to ensure that the system can maintain basic functions when part of the components are abnormal. The system architecture after reconstruction: the control system organization form after optimization and recombination, which clearly defines the control responsibilities and collaboration of each device, usually represented in the form of device connection matrix and control logic tree. Alternative device regulation instruction sequence: the device control instruction time sequence generated based on the architecture after reconstruction, which is optimized to be as close as possible to the original control target under the premise of ensuring safety.

[0062] Specifically, identify the affected environmental regulation loop: when the digital twin detects that the expected running state of the key device continuously exceeds the safety boundary, the system immediately starts impact analysis: Establish a device-loop mapping table: for example, compressor C1 serves the "east zone temperature regulation loop R1", and fan F2 serves the "north zone fresh air regulation loop R2" Topology propagation analysis: based on the pipe network connection relationship in the building BIM model, identify the affected upstream and downstream devices. For example, compressor overload will affect associated components such as condensers, expansion valves, etc.

[0063] Loop importance evaluation: according to the area function served by the loop (such as priority for critical areas such as operating rooms, data centers, etc.), the impact is classified: Output results: generate a list of affected loops, label the failure level (mild / severe) and impact range of each loop Dynamic reconstruction of control rights based on digital twin: Perform system architecture reconstruction in the virtual environment of the digital twin: Backup device activation: if the main compressor is overloaded, automatically activate the backup redundant compressor and recalculate the refrigerant distribution; Redundant control loop switching: when a certain temperature control loop fails, transfer control rights to the redundant loop in the adjacent area, and achieve coverage through air valve adjustment; Load redistribution: based on device capacity, real-time calculation is performed to proportionally distribute the load of the overloaded device to multiple healthy devices; Reconstruction verification: simulate the operation of the architecture after reconstruction in the virtual environment for 1-2 control cycles to confirm system stability; Re-run the model predictive controller: take the system architecture after reconstruction as the new boundary condition: Update the device state constraints of MPC: such as the power range of the backup device, the control range of the redundant loop; Adjusting the optimization target weight: increasing the device balanced use weight on the basis of the lowest energy consumption; Solving the alternative device control instruction sequence: the time range is usually shortened to 1 / 2 of the original first preset time length (such as from 2 hours to 1 hour), so as to improve the response speed; Issuing instructions and generating reports: issuing the first instruction of the alternative instruction sequence to the field execution mechanism through industrial Ethernet, and synchronously generating a structured report, including: Device early warning: detailed description of overloaded devices, overrun degree, and expected impact; System reconstruction report: reconstruction strategy, enabled device list, and performance expectation; Maintenance suggestions: suggest inspection time, spare parts preparation, etc.

[0064] In a preferred embodiment, the dynamic reconstruction of the control right of the environment control loop to the reconstructed system architecture in the virtual environment based on the digital twin includes the following steps: Based on the device topology relationship of the system, at least two alternative reconstruction strategies for distributing the load of the environment control loop to healthy backup devices or redundant control loops are generated; For each of the alternative reconstruction strategies, forward simulation is performed in the digital twin to pre-play the process of executing the alternative device control instruction sequence, and the pre-play performance indicators of each of the alternative reconstruction strategies are calculated, including the impact intensity of the switching process on comfort, additional energy consumption cost, and device life loss coefficient; Based on the pre-play performance indicators, each of the alternative reconstruction strategies is comprehensively scored; According to the alternative reconstruction strategy with the highest score, the reconstructed system architecture is obtained.

[0065] Specifically, the device topology relationship: describes the network structure of the physical connection and functional association between components in the building device system, including pipe connection, electrical connection, control signal transmission, etc. Forward simulation: starting from the current state, simulating the future evolution process of the system in chronological order, predicting the system behavior by solving differential equations or proxy models. Pre-play performance indicators: a quantitative standard system for evaluating the effect of reconstruction strategies, covering comfort, economy, device reliability, etc. Comfort impact intensity: quantifies the degree of disturbance to indoor environmental comfort during system reconstruction, usually represented by the integral quantity of environmental parameter deviation from the set value. Device life loss coefficient: represents the influence of a specific operating condition on the remaining service life of the device, calculated based on the stress-life curve and cumulative damage model.

[0066] The following describes the specific implementation: Based on the system, the device topology relationship is automatically generated multiple reconstruction schemes: Topology relationship modeling: Using graph theory method, taking device as node, connection relationship as edge, building system topology graph. Weight represents physical parameters such as pipe resistance and cable capacity; Strategy generation algorithm: Load transfer strategy: Identify adjacent healthy devices of overloaded devices and calculate transferable load; Function replacement strategy: Find different but functionally equivalent device combinations, such as replacing a single large-capacity device with multiple small-capacity devices; Area recombination strategy: Redivide control area and adjust the service range of each device; Strategy diversity: Ensure that at least two essentially different strategies are generated, such as one focusing on device redundancy and one focusing on control loop recombination; Forward simulation in digital twin: Full-process simulation in virtual environment for each alternative strategy: Simulation environment setting: Replicate all state parameters of the current actual system as simulation initial conditions; Execution process rehearsal: Simulate system dynamic response according to alternative device control instruction sequence with high time resolution (e.g. 10 seconds / step). Multi-dimensional data acquisition: Record environmental parameters such as temperature, humidity and pressure during simulation, as well as operating parameters such as power and speed of each device. Calculate performance indicators: Quantitatively evaluate each strategy based on simulation data: Comfort impact intensity: Calculate the root mean square deviation of key environmental parameters (temperature, CO2 concentration) before and after strategy execution to quantify the impact on comfort. Additional energy consumption cost: Compare the energy consumption difference between the reconstruction strategy and the original strategy, and calculate the economic cost combined with real-time electricity price. Device life loss coefficient: Calculate the fatigue damage increment of key devices based on the Miner linear cumulative damage theory. Comprehensive score and strategy selection: Weight distribution: Dynamically adjust the weight of indicators according to the current operation mode of the building. For example, in hospital mode, the comfort weight is the highest, and in commercial building mode, the energy consumption weight is higher; Score calculation: Use the TOPSIS multi-objective decision-making method to calculate the relative closeness of each strategy to the ideal solution; Select the strategy with the highest comprehensive score as the final reconstruction scheme.

[0067] In a preferred embodiment, the reconstruction system architecture according to the alternative reconstruction strategy with the highest score comprises the following steps: Based on the digital twin, stress test the alternative reconstruction strategy with the highest score under multiple preset disturbance scenarios to obtain the performance retention of the alternative reconstruction strategy under each disturbance scenario, the preset disturbance scenarios include sudden fluctuations in human flow load and / or step changes in outdoor meteorological parameters, and the performance retention is the proportion of time that still meets the comfort constraints under disturbance; If the performance retention of the alternative reconstruction strategy under each disturbance scenario meets the standard, the architecture corresponding to the alternative reconstruction strategy with the highest score is taken as the reconstruction system architecture.

[0068] Specifically, the top-scoring alternative reconstruction strategy is subjected to extreme condition testing: disturbance scenario library construction: multiple typical disturbance patterns are predefined: human flow load sudden fluctuation: scenarios such as sudden dispersal of large meetings, restaurant peak, etc. are simulated, and the human flow increases by 50%-100% within 5 minutes; outdoor meteorological parameter step change: scenarios such as cold front passage, sudden change of solar radiation, etc. are simulated, and the temperature changes by 3-5°C within 10 minutes; mixed disturbance scenario: the worst case of multiple disturbances occurring simultaneously; test execution: each disturbance scenario is applied in the digital twin in turn, and each scenario simulation duration covers the entire control period; performance retention calculation: quantify the performance of the strategy under disturbance: comfort constraint check: monitor whether the indoor temperature, CO2 concentration, etc. remain within the set range during simulation; time proportion calculation: performance retention = (number of time points meeting the comfort constraint / total number of time points) x 100%; set a threshold (e.g. ≥95%), below which the strategy is considered insufficiently robust; if all thresholds are met, the system architecture corresponding to the strategy is officially determined as the reconstructed system architecture.

[0069] In a preferred embodiment, after obtaining the performance retention of the alternative reconstruction strategy under each disturbance scenario, the following steps are further included: If the performance retention of the alternative reconstruction strategy under any of the preset disturbance scenarios does not meet the threshold, the key devices and their action periods that cause performance degradation are deduced based on the characteristics of the non-compliant preset disturbance scenario in the digital twin; Based on the device control instruction sequence obtained in the previous control period, a dynamic compensation instruction sequence for the key devices in the action period is generated, which includes a gradual adjustment curve of device output power and switching timing optimization; The dynamic compensation instruction sequence is fused with the alternative reconstruction strategy to form an enhanced reconstruction strategy; The enhanced reconstruction strategy is verified in the digital twin, and if it passes the verification, its corresponding architecture is determined as the reconstructed system architecture.

[0070] Specifically, based on the disturbance characteristics, the key devices and their action periods are deduced: Disturbance characteristic deep analysis: the system first analyzes the multi-dimensional characteristics of the non-compliant disturbance scenario. This includes analyzing the amplitude, change rate, duration, and propagation path of the disturbance in the building space. Special attention is paid to the directionality of the disturbance, such as whether the temperature rises or falls, and whether the human flow converges or diverges.

[0071] Sensitivity mapping: running a sensitivity analysis algorithm based on mathematical adjoint equations in the digital twin. This method calculates the sensitivity coefficients of all device parameters on system performance by constructing an adjoint model corresponding to the original system model. The system identifies key devices whose state changes have a decisive impact on system performance degradation.

[0072] Precise positioning of performance degradation period: using a sliding time window analysis technique, the system accurately scans the specific time interval when the system performance starts to deteriorate, reaches the worst, and gradually recovers during the entire simulation process with a time resolution of five minutes. By comparing the system response characteristics in multiple time windows, the system accurately locates the action period of the key device that has the maximum negative impact.

[0073] Specifically, a dynamic compensation instruction sequence is generated: Gradual power adjustment curve design: based on the device control instruction sequence obtained in the previous control period, a smooth power compensation curve is designed for the identified key devices within their action period. An S-shaped transition function is used to generate the adjustment trajectory of device output power, ensuring that the power change is continuous and smooth rather than stepwise. This design allows the device output power to gradually increase to the compensation value at a preset rate and smoothly return to the normal level after the action period. The system sets individualized maximum change rate limits for different devices based on their dynamic characteristics to prevent rapid power changes from causing mechanical or electrical shocks to the devices.

[0074] Multi-device coordination timing optimization: based on the critical path analysis method, the action timing of multiple devices that need to be compensated is optimized. The system establishes a dependency graph of device actions to avoid system oscillation or resource conflicts caused by multiple devices acting simultaneously. By introducing appropriate start-up delays and action intervals, the system ensures that the execution of compensation instructions is reasonably distributed in the time dimension, thereby ensuring the stability of the system during the entire compensation process.

[0075] Specifically, an enhanced reconstruction strategy is formed by fusion: Deep integration of control logic: the designed dynamic compensation instruction sequence is organically integrated with the original candidate reconstruction strategy to form a hierarchical control architecture. In this architecture, the upper layer is the global optimization instruction based on model predictive control, responsible for the overall running efficiency of the system; the lower layer is the local compensation instruction based on real-time state feedback, specifically handling performance degradation problems under specific disturbances. The instructions of the two levels are synthesized through a unified coordination mechanism to ensure consistency in control objectives.

[0076] Parameter coordination and conflict resolution: The system will establish a special parameter coordination module to detect and resolve potential conflicts between compensation instructions and original instructions. By setting priority rules and constraints, it ensures that there are no contradictory control requirements during instruction fusion. For potential conflicts detected, the system will automatically adjust based on multi-objective optimization principles to find the optimal compromise solution that meets all constraints.

[0077] Specifically, verification and validation in digital twins: Comprehensive regression testing: The enhanced refactoring strategy is re-placed in the digital twin for comprehensive verification. First, re-run those disturbance scenarios that did not pass before, and focus on observing whether the system's performance under the same disturbance conditions meets the predetermined standards after the compensation mechanism is added.

[0078] Boundary performance testing: Test the stability of the enhanced strategy under more severe working conditions, including simultaneous application of multiple disturbances, prolonged disturbance duration, increased disturbance amplitude, and other extreme tests to verify the robust boundary of the strategy under extreme conditions.

[0079] Long-term running stability test: Simulate the scenario of continuous running of the enhanced strategy for multiple control cycles, and check whether the system has cumulative errors or performance degradation. Only the enhanced strategy that passes all these verification links and maintains a stable performance of 97% or more will be finally confirmed to be adopted.

[0080] This embodiment can convert most substandard refactoring strategies into effective and usable control schemes through precise etiology diagnosis and targeted compensation design, significantly reducing the consumption of manpower and resources for strategy abandonment and redesign. The system can maintain the stability of environmental parameters and the reliable operation of equipment when facing abnormal working conditions such as sudden changes in human flow and drastic changes in weather conditions.

[0081] In this paper, specific examples are used to illustrate the principles and embodiments of the application. The above examples are only used to help understand the method and its core idea. The above description is only the preferred embodiment of the application. It should be noted that due to the limited nature of the language, there are infinite specific structures. For ordinary skilled persons in the art, without departing from the principles of the application, some improvements, refinements or changes can be made, or the above technical features can be combined in an appropriate manner; these improvements, refinements, changes or combinations, or the application of the inventive concept and technical solution to other fields without improvement, shall be regarded as the protection scope of the application.

Claims

1. A method for adaptive regulation of a smart building environment based on digital twinning and AI, characterized in that, The method comprises the following steps: extracting dynamic features based on the time-aligned multi-source system data to construct a dynamic feature set; the multi-source system data comprises environmental data, equipment operation data, and people flow data; the dynamic features comprise thermal inertia delay features and people flow impact conduction features; inputting the multi-source system data into a digital twin, taking the current building equipment state as a boundary condition, iteratively simulating key environmental parameters at multiple time steps in the future, and outputting a multi-step prediction sequence; the key environmental parameters comprise indoor temperature and CO2 concentration; inputting the multi-step prediction sequence into a model predictive controller, taking the minimum overall energy consumption in the prediction period as the optimization objective, and taking the key environmental parameters not exceeding the comfort interval as the constraint condition, to obtain the equipment control instruction sequence in the first preset time period in the future; the first equipment control instruction in the equipment control instruction sequence is issued to the on-site execution mechanism to realize adaptive regulation of the building environment; the digital twin is a physical information neural network constructed by fusing the dynamic feature set and a building cyber-physical model; the building cyber-physical model is obtained by discretizing thermodynamic differential equations, and the loss function of the physical information neural network comprises a data fitting term and a physical equation constraint term.

2. The method of claim 1, wherein the method further comprises: determining a current state of the building environment; and determining a current state of the digital twin of the building environment. extracting thermal inertia delay features based on the time-aligned multi-source system data, comprising the following steps: based on the environmental temperature sequence and the HVAC power sequence in the target historical time period, calculating the mutual information value at different time offsets, and taking the time offset that makes the mutual information value reach the maximum as the thermal inertia delay time of the building; for each collection time, the thermal inertia delay feature is obtained from the real-time environmental temperature collected at the collection time and the HVAC energy consumption delayed by the thermal inertia delay time from the collection time.

3. The method of claim 2, wherein the method further comprises: determining a current state of the building environment; and determining a current state of the digital twin of the building environment. the mutual information value is obtained by the following formula: ； wherein, represents a mutual information value; represents a time offset; T represents an ambient temperature sequence; t represents a real-time ambient temperature; p represents an HVAC power sequence; represents a temperature value t i and a delayed power value at the jth collection time represents a joint probability of simultaneous occurrence; represents a temperature marginal probability; represents a delayed power marginal probability; N represents a total number of data of the ambient temperature sequence T; and M represents a total number of data of the HVAC power sequence p.

4. The method of claim 2, wherein the method further comprises: determining a current state of the building environment; and determining a current state of the digital twin of the building environment. before calculating the mutual information value at different time offsets based on the environmental temperature sequence and the HVAC power sequence in the target historical time period, the following steps are included: obtain the mode feature vector of the current day to determine the building operation mode category to which the current day belongs; the mode feature vector comprises daily average outdoor temperature, daily HVAC energy consumption peak, and people flow density distribution histogram; according to the building operation mode category to which the current day belongs, the corresponding target historical time period is obtained.

5. The digital-twin and AI-based adaptive regulation method for a smart building environment according to claim 4, characterized in that: the steps of obtaining the mode feature vector of the current day to determine the building operation mode category to which the current day belongs include: obtain the mode feature vector of the current day, retrieve and traverse the historical mode database to obtain the building operation mode category corresponding to the mode feature vector of the current day, and take it as the building operation mode category to which the current day belongs; the historical mode database comprises a plurality of historical time periods and a building operation mode category corresponding to each historical time period.

6. The digital-twin and AI-based adaptive regulation method for a smart building environment according to claim 1, characterized in that: the method further comprises the following steps: The model predictive controller is used to solve the device control instruction sequence, and at the same time, based on the digital twin, a synchronous simulation is performed on an expected running state of a key device when the device control instruction sequence is executed, the expected running state including an expected load rate of a compressor and an expected rotating speed of a fan; After the solving of the device control instruction sequence in the future for the first preset time length, the method further includes the following steps: It is judged whether the expected running state of any key device in the first preset time length continuously exceeds a safe running boundary thereof; The first device control instruction in the device control instruction sequence is issued to a field execution mechanism. If not, the first device control instruction in the device control instruction sequence is issued to the field execution mechanism.

7. The digital-twin and AI-based adaptive regulation method for a smart building environment according to claim 6, characterized in that: After the judgment of whether the expected running state of any key device in the first preset time length continuously exceeds the safe running boundary thereof, the method further includes the following steps: If yes, an affected environmental control loop is identified, and based on the digital twin, a control right of the environmental control loop is dynamically reconstructed to a reconstructed system architecture in a virtual environment, the reconstructed system architecture including one or more healthy backup devices or redundant control loops; The model predictive controller is re-run with the reconstructed system architecture as an updated boundary condition to solve a candidate device control instruction sequence in the future for a second preset time length; The first device control instruction in the candidate device control instruction sequence is issued to the field execution mechanism, and a device early warning and system reconstruction report is synchronously generated.

8. The digital-twin and AI-based adaptive regulation method for a smart building environment according to claim 7, characterized in that: The dynamic reconstruction of the control right of the environmental control loop to the reconstructed system architecture in the virtual environment based on the digital twin includes the following steps: Based on a device topology relationship of the system, at least two candidate reconstruction strategies of distributing a load of the environmental control loop to the healthy backup devices or the redundant control loops are generated; For each candidate reconstruction strategy, a forward simulation is performed in the digital twin to pre-play a process of executing the candidate device control instruction sequence, and a pre-play performance index of each candidate reconstruction strategy is calculated, the pre-play performance index including an impact intensity of a switching process on comfort, an additional energy consumption cost, and a device life loss coefficient; Based on the pre-play performance index, each candidate reconstruction strategy is comprehensively scored; The reconstructed system architecture is obtained according to the candidate reconstruction strategy with the highest score.

9. The digital-twin and AI-based adaptive regulation method for a smart building environment according to claim 8, characterized in that: The reconstructed system architecture obtained according to the candidate reconstruction strategy with the highest score includes the following steps: Based on the digital twin, the candidate reconstruction strategy with the highest score is subjected to stress testing of multiple preset disturbance scenarios to obtain a performance retention degree of the candidate reconstruction strategy under each disturbance scenario, the preset disturbance scenario including an abrupt fluctuation of a human flow load and / or a step change of an outdoor meteorological parameter, and the performance retention degree being a time proportion still meeting a comfort constraint under the disturbance. If the performance retention degrees of the alternative reconstruction strategies under each of the disturbance scenarios are all up to the standard, the architecture corresponding to the alternative reconstruction strategy with the highest score is determined as the reconstructed system architecture.

10. The digital-twin and AI-based adaptive regulation method for a smart building environment according to claim 9, characterized in that: After obtaining the performance retention degrees of the alternative reconstruction strategies under each of the disturbance scenarios, the following steps are further included: If the performance retention degree of the alternative reconstruction strategy under any of the preset disturbance scenarios does not reach the standard, the key equipment and its action period that cause performance degradation are deduced in the digital twin based on the characteristics of the preset disturbance scenario that does not reach the standard; Based on the equipment control instruction sequence obtained in the previous control period, a dynamic compensation instruction sequence for the key equipment in the action period is generated, which includes a gradual adjustment curve of equipment output power and switching timing optimization; The dynamic compensation instruction sequence is fused with the alternative reconstruction strategy to form an enhanced reconstruction strategy; The enhanced reconstruction strategy is verified in the digital twin, and the architecture corresponding to the enhanced reconstruction strategy is determined as the reconstructed system architecture after verification.

Images