A digital twin scientific research equipment intelligent management system fusing internet of things and artificial intelligence

Abstract

The present application relates to the technical field of digital twin device management, in particular to a digital twin scientific research device intelligent management system combining Internet of Things and artificial intelligence. It comprises a multi-source heterogeneous data acquisition unit for collecting real-time information on the operating environment and device state of scientific research equipment; a digital twin body mapping unit constructs a temperature and humidity coupling dynamic model by introducing a temperature and humidity coupling coefficient, which integrates a temperature dynamic model and a humidity dynamic model, and maps the data collected from multi-source equipment and laboratory environment to the temperature and humidity coupling dynamic model; an AI multidimensional prediction unit predicts the future operating state of the equipment and environmental risks; a risk warning unit analyzes the operating state of the air conditioning equipment and its environmental coupling risks in real time based on the prediction results of the AI multidimensional prediction unit, and issues a warning signal before potential abnormalities occur. It is used to predict equipment failure risks in advance, thereby achieving the purpose of advanced prediction and proactive intervention.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital twin device management, in particular to a digital twin scientific research device intelligent management system fusing Internet of Things and artificial intelligence. BACKGROUND

[0002] In the scientific research activities in the fields of modern life science, biological medicine, material chemistry, etc., high-precision and high-reliability scientific research devices are the fundamental guarantee for the accuracy of experimental data and the safety of samples. Among them, the running state of sample storage type devices (such as ultra-low temperature freezers, medical refrigerators, constant temperature and humidity incubators) and reaction process type devices (such as PCR instruments, biochemical incubators, constant temperature shakers) is crucial. The core task of such devices is to maintain the temperature and humidity of the internal cavity (or working cavity) at the set target value for a long time, and any slight, continuous or sudden temperature and humidity deviation may cause irreparable losses such as sample inactivation, reagent failure, and experimental data invalidation.

[0003] Although the existing devices are equipped with basic sensors, the data is usually limited to local display or independent alarm of the device, and the fusion perception of multi-source data cannot be achieved. For example, the compressor power consumption, door opening and closing times, environmental laboratory temperature and humidity, and surrounding air flow of an ultra-low temperature freezer are mutually influenced. The traditional device management system cannot truly reflect the virtual image of the device with coupled dynamics characteristics of temperature and humidity, cannot make advanced simulation and intelligent decision based on the image, and thus lacks in the aspects of precise control of the running state of scientific research devices, fault prediction, etc. Therefore, there is an urgent need for a digital twin scientific research device intelligent management system fusing Internet of Things and artificial intelligence. SUMMARY

[0004] The present application aims to provide a digital twin scientific research device intelligent management system fusing Internet of Things and artificial intelligence to solve the problem that the traditional device management system cannot truly reflect the virtual image of the device with coupled dynamics characteristics of temperature and humidity, cannot make advanced simulation and intelligent decision based on the image, and thus lacks in the aspects of precise control of the running state of scientific research devices, fault prediction, etc.

[0005] To achieve the above-mentioned purpose, the present application aims to provide a digital twin scientific research device intelligent management system fusing Internet of Things and artificial intelligence, comprising:

[0006] A multi-source heterogeneous data acquisition unit is used to collect real-time information of the running environment and device state of the scientific research device, and obtain running state parameters and multi-dimensional external parameters.

[0007] The digital twin mapping unit constructs a temperature and humidity coupling dynamic model by introducing a temperature and humidity coupling coefficient based on the operating state parameters and multi-dimensional external parameters, and maps the data collected from the multi-source equipment and laboratory environment to the temperature and humidity coupling dynamic model, and finally outputs real-time state data;

[0008] The temperature dynamic model is based on a first-order dynamic system and is constructed by introducing external environmental temperature and instantaneous disturbance terms;

[0009] The humidity dynamic model is based on a first-order dynamic system, takes the temperature change rate as a coupling influencing factor, and introduces external environmental humidity and instantaneous humidity disturbance to construct;

[0010] The AI multi-dimensional prediction unit constructs a CNN-LSTM prediction model based on the real-time state data generated by the digital twin mapping unit and the collected multi-source heterogeneous environment data to predict the future operating state of the equipment and the environmental risk;

[0011] The risk warning unit analyzes the operating state of the scientific research equipment and its environmental coupling risk in real time based on the prediction results of the AI multi-dimensional prediction unit, and issues a warning signal before potential abnormalities occur.

[0012] As a further improvement of the technical solution, the multi-source heterogeneous data acquisition unit includes an equipment data perception module and an environment perception module;

[0013] The equipment data perception module is used to collect operating state parameters of multi-source equipment and its execution components, and the operating state parameters include compressor parameters, air supply system parameters, heat exchanger parameters, energy consumption data, and operating mode information.

[0014] The environment perception module is used to collect multi-dimensional external parameters in the laboratory environment where the equipment is located, and the multi-dimensional external parameters include environmental parameters, air quality parameters, air dynamics parameters, and environmental interference factors.

[0015] As a further improvement of the technical solution, the digital twin mapping unit includes a physical mechanism modeling module and a real-time data mapping module;

[0016] The physical mechanism modeling module establishes a temperature and humidity coupling dynamic model according to the type of laboratory equipment, and the temperature and humidity coupling dynamic model is adapted to different equipment through parameter configuration;

[0017] The real-time data mapping module maps the data collected from the multi-source equipment and laboratory environment to the temperature and humidity coupling dynamic model, and is used to realize the synchronous output of the real-time state data of the equipment.

[0018] As a further improvement of the technical solution, the specific steps of the temperature and humidity coupling dynamic model construction and data mapping are:

[0019] Obtain the device historical operation data: temperature, humidity, compressor state and fan speed, form the original input vector , and pretreat the original input vector;

[0020] Based on the pretreated original input vector , the rate of change of temperature with time and the deviation between the current temperature and the target temperature are calculated;

[0021] Based on the rate of change of temperature with time and the deviation between the current temperature and the target temperature, a first-order dynamic system is used to construct a temperature dynamic model, and the external environment temperature is introduced as a disturbance input of heat penetration, and the opening event is taken as a transient disturbance term , for predicting the dynamic change of the cavity temperature with time;

[0022] Based on the pretreated original input vector , the rate of change of the cavity relative humidity with time is calculated based on the deviation between the current humidity and the target humidity, and the temperature change rate is introduced as a coupling influence factor, a first-order dynamic system is used to construct a humidity dynamic model, and the external environment humidity is introduced as a disturbance input of heat penetration, and the transient humidity disturbance is taken as a disturbance term, for predicting the dynamic change of the cavity relative humidity with time;

[0023] The temperature dynamic model and the humidity dynamic model are superimposed and fused, and a temperature and humidity coupling dynamic model is constructed by introducing a temperature and humidity coupling coefficient;

[0024] The temperature and humidity coupling dynamic model is parameter corrected based on the least square method, for calibrating the cavity heat capacity, heat transfer coefficient, temperature and humidity coupling coefficient and disturbance correction coefficient;

[0025] The real-time operation parameters collected from the device sensors and the environment sensors are received by the real-time data mapping module, and the real-time operation parameters are pretreated, and the pretreated real-time operation parameters are taken as the input of the calibrated temperature and humidity coupling dynamic model, and the real-time state data is output;

[0026] The temperature and humidity coupling dynamic model calculates the device cavity state at the current time based on the current real-time input and the state at the last time;

[0027] A temperature and humidity combined PID controller is constructed, taking the real-time temperature and humidity error output by the temperature and humidity coupling dynamic model as input, and outputting the compressor duty ratio, fan setting, and dehumidification / humidification instructions.

[0028] As a further improvement of the technical solution, the temperature and humidity combined PID controller is composed of a temperature control channel and a humidity control channel in parallel;

[0029] The temperature control channel is based on the deviation between the target temperature of the cavity and the output temperature of the temperature dynamic model, and introduces a humidity feedforward compensation term, to construct a temperature PID controller for generating a compressor duty ratio adjustment signal.

[0030] The humidity control channel is based on the deviation between the target humidity of the cavity and the output humidity of the humidity dynamic model, and introduces a temperature feedforward compensation term, to construct a humidity PID controller for generating a humidification / dehumidification adjustment signal.

[0031] As a further improvement of the technical solution, the AI multi-dimensional prediction unit includes a device state prediction module, an environmental risk prediction module, and a multi-dimensional prediction fusion module.

[0032] The device state prediction module is based on the real-time state data output by the digital twin mapping unit to construct a prediction input sequence, and based on the prediction input sequence to construct a CNN-LSTM prediction model to predict the temperature, humidity, compressor operating state, and fan speed of the scientific research equipment cavity, and output a device state prediction sequence.

[0033] The environmental risk prediction module is based on the influence of external environmental conditions on the operation of the scientific research equipment, and by introducing a Fourier basis function, constructs a multivariate time series prediction model to output an environmental risk index prediction sequence for early warning of abnormal temperature and humidity fluctuations.

[0034] The multi-dimensional prediction fusion module is used to fuse the prediction results output by the device state prediction module and the environmental risk prediction module to obtain a comprehensive risk prediction index.

[0035] As a further improvement of the technical solution, the specific steps involved in the device state prediction sequence output by the CNN-LSTM prediction model are as follows:

[0036] Based on the real-time state data, a prediction input sequence is constructed , and the prediction input sequence is spliced by a sliding window to obtain an input prediction sample , and principal component analysis is used to obtain low-dimensional features .

[0037] A CNN-LSTM prediction model including a convolution feature extraction subnetwork and a recurrent neural network is constructed.

[0038] The low-dimensional features are extracted by a convolution feature extraction sub-network The local feature vector of the temperature and humidity change over time is extracted by performing convolution operation, and a local feature mapping matrix is output;

[0039] The local feature mapping matrix is input into a recurrent neural network to generate a time sequence feature sequence;

[0040] And by introducing an attention mechanism, different weights are assigned to the features corresponding to each time in the time sequence feature sequence, and a weighted time sequence feature is obtained;

[0041] The weighted time sequence feature is sent to a fully connected output layer to obtain a device operating state prediction value at each time within a future Step prediction length;

[0042] Among them, the device operating state prediction value includes the temperature prediction value and the humidity prediction value of the target device cavity.

[0043] As a further improvement of the technical solution, the risk warning unit includes an anomaly detection module and a risk warning module;

[0044] Among them, the anomaly detection module respectively receives the model reference output from the digital twin mapping unit and the prediction result from the AI multi-dimensional prediction unit, and combines the actual collection value to perform difference analysis;

[0045] The risk warning module outputs multi-level warning information according to the determination result of the anomaly detection module combined with the risk grading rules.

[0046] As a further improvement of the technical solution, in the anomaly detection module, the prediction result is combined with the actual collection value to perform difference analysis, which involves the following specific steps:

[0047] The temperature and humidity theoretical reference value output by the digital twin mapping unit at time The prediction value of the monitoring variable at time Output by the AI multi-dimensional prediction unit and the actual collection value of the monitoring variable collected by the multi-source heterogeneous data collection unit are obtained;

[0048] When the deviation between the actual collection value and the twin model reference value exceeds the preset threshold, the chi-square test is used to determine whether there is a trend deviation in the system;

[0049] When the residual error between the actual collection value and the AI prediction result exceeds the preset threshold within multiple time steps, the mean square error is used for quantitative evaluation, and the residual error sequence is subjected to abnormal pattern recognition combined with the one-class support vector machine;

[0050] If any of the above items are found to be abnormal, an abnormality signal will be output to the risk warning module.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] 1. In this intelligent management system for digital twin scientific research equipment that integrates the Internet of Things and artificial intelligence, a temperature and humidity coupled dynamic model is constructed, and the rate of temperature change is introduced as a coupling influencing factor in the humidity dynamic model. A feedforward compensation term based on humidity deviation is introduced into the output of the temperature PID controller, so that the compressor control can respond to humidity changes in advance (e.g., high humidity indicates an increase in latent heat load, so the cooling can be increased in advance). Similarly, the humidity PID controller introduces a feedforward compensation term based on temperature deviation, which can effectively suppress system oscillations caused by mutual interference of control actions, so that the system can recover to the set value more quickly and smoothly after dealing with disturbances (such as opening a door).

[0053] 2. In this intelligent management system for digital twin scientific research equipment that integrates the Internet of Things and artificial intelligence, a CNN-LSTM prediction model is constructed by an AI multidimensional prediction unit. The CNN layer is responsible for extracting local features from the sliding window sequence (such as the pattern of frequent start-stop of the compressor in a short period of time), and the LSTM layer is responsible for learning long-term temporal dependencies (such as the performance degradation trend of the equipment after 24 hours of continuous operation). By introducing Fourier basis functions to explicitly model the periodic features in the environmental data (such as day-night and seasonal temperature differences), the stability of long-term prediction is improved. The system outputs the equipment status prediction sequence and environmental risk prediction sequence for the next H steps and integrates them into a comprehensive risk index. This is used to predict equipment failure risks several hours or even days in advance (such as the risk of temperature rise due to compressor failure), thereby achieving the purpose of advance prediction and proactive intervention. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the overall process of the present invention.

[0055] The meanings of the labels in the diagram are as follows:

[0056] 1. Multi-source heterogeneous data acquisition unit; 11. Equipment data sensing module; 12. Environmental sensing module;

[0057] 2. Digital twin mapping unit; 21. Physical mechanism modeling module; 22. Real-time data mapping module;

[0058] 3. AI multidimensional prediction unit; 31. Equipment status prediction module; 32. Environmental risk prediction module; 33. Multidimensional prediction fusion module;

[0059] 4. Risk warning unit; 41. Anomaly detection module; 42. Risk warning module. Detailed Implementation

[0060] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0061] Embodiment: Please refer to Figure 1 As shown in the figure, a digital twin scientific research equipment intelligent management system combining Internet of Things and artificial intelligence is provided, comprising a multi-source heterogeneous data acquisition unit 1, which is used for collecting real-time information of the running environment and equipment state of the scientific research equipment, obtaining running state parameters and multi-dimensional external parameters;

[0062] In the present embodiment, the multi-source heterogeneous data acquisition unit 1 comprises a device data perception module 11 and an environment perception module 12.

[0063] The device data perception module 11 is used for collecting running state parameters of multi-source equipment and its execution components, and the running state parameters comprise compressor parameters (including compressor start-stop state, duty ratio, rotating speed, current, voltage and power data), air supply system parameters (including air supply fan rotating speed, air volume and air pressure), heat exchanger parameters (including condenser / evaporator surface temperature and heat exchange efficiency estimated value), energy consumption data (including real-time power consumption and cumulative energy consumption) and running mode information (including refrigeration mode, heating mode, dehumidification mode and air supply mode, etc.). Specifically, the device data perception module 11 is realized by arranging current transformers, rotating speed sensors and voltage acquisition ports and other sensors at the compressor, fan, motor and other parts, so as to ensure high accuracy and high real-time performance of data acquisition.

[0064] The environment perception module 12 is used for collecting multi-dimensional external parameters in the laboratory environment where the equipment is located, and the multi-dimensional external parameters comprise environmental parameters (indoor and outdoor temperature and humidity), air quality parameters (carbon dioxide concentration, PM2.5 / PM10 particulate matter concentration and VOC concentration), aerodynamic parameters (wind speed and airflow distribution) and environmental interference factors (noise intensity, vibration signal and light intensity). The environment perception module 12 realizes multi-source heterogeneous perception of the environmental state by arranging temperature and humidity sensors, infrared gas detectors, laser dust sensors, acoustic sensors and vibration acceleration sensors, so as to ensure that the digital twin model can fully reflect the air conditioner running and environmental coupling characteristics.

[0065] The fusion of the Internet of Things and artificial intelligence digital twin scientific research equipment intelligent management system also includes a digital twin mapping unit 2, which constructs a temperature and humidity coupling dynamic model by introducing a temperature and humidity coupling coefficient based on the operating state parameters and multi-dimensional external parameters, and maps the data collected from the multi-source equipment and laboratory environment to the temperature and humidity coupling dynamic model, and finally outputs real-time state data;

[0066] The temperature dynamic model is based on a first-order dynamic system and is constructed by introducing external environmental temperature and instantaneous disturbance terms;

[0067] The humidity dynamic model is based on a first-order dynamic system, takes the temperature change rate as a coupling influencing factor, and introduces external environmental humidity and instantaneous humidity disturbance to construct;

[0068] Specifically, the digital twin mapping unit 2 includes a physical mechanism modeling module 21 and a real-time data mapping module 22.

[0069] In this embodiment, the physical mechanism modeling module 21 establishes a temperature and humidity coupling dynamic model according to the type of laboratory equipment. The temperature and humidity coupling dynamic model is adapted to different equipment (mainly adapted to sample storage type equipment, which has significant interaction between temperature and humidity and has a direct impact on scientific research experiments) through parameter configuration. Specifically, such as ultra-low temperature refrigerator, constant temperature and humidity box, biological sample freezer, culture / reaction type equipment (cell incubator, constant temperature oven, PCR instrument) and the like:

[0070] When the system is used for an ultra-low temperature refrigerator, the heat capacity and heat transfer coefficient in the temperature and humidity coupling dynamic model need to be parameter calibrated according to the cavity volume and thermal insulation performance of the refrigerator (the biological samples, reagents, and cells stored therein have dual requirements for temperature and humidity, temperature fluctuations can cause humidity condensation or evaporation, affecting the humidity of the cavity, and opening / closing the door will cause cold loss and also cause moisture to enter, causing condensation or frost);

[0071] When used for a PCR instrument or a cell incubator, the instantaneous disturbance term caused by frequent opening and closing of the door needs to be added to the model, and the weight of the humidity coupling factor needs to be increased (the precise control of the temperature and humidity environment directly affects the experimental results);

[0072] When used for a freeze dryer or a high-speed centrifuge, air flow disturbance and pressure compensation factors are further introduced into the model; and in a climate simulation box or an open culture rack, the weight of the external environmental humidity and temperature penetration term needs to be increased to ensure the accurate adaptability of the model;

[0073] Real-time data mapping module 22, real-time data mapping module 22 maps the data collected from the multi-source equipment and laboratory environment to the temperature and humidity coupling dynamic model, for realizing the synchronous output of the real-time state data of the equipment.

[0074] The specific steps of the temperature and humidity coupling dynamic model construction and data mapping are:

[0075] Obtain equipment historical operation data: temperature, humidity, compressor state and fan speed and other key operation parameters to form an original input vector ; In the formula, represents the actual measured temperature of the equipment internal cavity at time ; represents the external environment temperature at time ; is the on-off state of the compressor at time , wherein 0 represents off and 1 represents on; represents the cumulative running time of the compressor from start to the current time ; represents the real-time speed of the fan in the equipment cavity at time ; represents the relative humidity (unit: %) of the air inside the equipment cavity at time ; represents the temperature measurement value of the equipment refrigerant pipeline at time , and the original input vector is preprocessed;

[0076] Among them, the collection sources of key operation parameters are: equipment sensors (compressor, fan), environmental sensors (temperature, humidity), internal cavity state, refrigerant temperature and calibration parameters (equivalent heat capacity of scientific research equipment cavity (unit: J / ℃, used to describe the cavity heat absorption / heat release capacity) and equivalent heat transfer coefficient (unit: W / ℃, used to represent the cavity and external environment heat conduction efficiency);

[0077] Specifically, the original input vector is time resampled (linear interpolation alignment is performed with fixed time step (1-3 seconds)), outlier rejection (Hampel algorithm) and normalized (z-score or min-max algorithm is adopted), to obtain the preprocessed original input vector ;

[0078] Based on the preprocessed original input vector , the rate of change of temperature with time and the deviation between the current temperature and the target temperature are calculated;

[0079] Based on the rate of change of temperature with time and the deviation between the current temperature and the target temperature, a first-order dynamic system is adopted to construct a temperature dynamic model (the first-order dynamic system is a first-order dynamic system modeling based on energy balance, the state variable of which is the cavity temperature, the control input is the compressor duty cycle signal, and the disturbance input includes the external environment temperature and the door opening event), and the external environment temperature is introduced as a disturbance input of heat penetration, and the door opening event is taken as a transient disturbance term , for predicting the dynamic change of the cavity temperature with time;

[0080] In this embodiment, a first-order dynamic system (the first-order dynamic system is essentially a first-order differential equation, which is used to describe the smooth change response of the state variable with time, and in this embodiment, it is suitable for scientific research equipment such as refrigerator cavity, thermostat, PCR instrument, etc.) is used to model the temperature dynamic model:

[0081]

[0082] In the formula, represents the cavity temperature , the instantaneous change rate of the continuous time , the real-time temperature of the scientific research equipment cavity within the continuous time ; and represents the equivalent heat capacity of the equipment cavity; represents the equivalent heat transfer coefficient; represents the equivalent heat transfer coefficient; represents the external environment temperature within the continuous time ; and represents the compressor refrigeration power conversion coefficient, which is used to convert the duty cycle of the compressor into the actual refrigeration power acting on the cavity (unit: watt W); represents the compressor duty cycle within the continuous time ; and represents the instantaneous disturbance power, which is a short-time influence on the cavity temperature at a certain sampling time, specifically refers to the impact of sudden events, such as door opening and heater action (unit: watt W); represents the continuous time (unit: second);

[0083] Real-time data is processed by discrete time:

[0084]

[0085] In the formula, represents the cavity temperature of the scientific research equipment within the discrete sampling time ; and represents the cavity temperature of the scientific research equipment within the discrete sampling time . denotes the sampling time interval; denotes the discrete sampling time ambient temperature; denotes the discrete sampling time compressor duty cycle; denotes the instantaneous disturbance power at the discrete sampling time;

[0086] Based on the pre-processed original input vector , the cavity relative humidity is calculated, and the rate of change over time is based on the deviation of the current humidity from the target humidity, and the temperature change rate is introduced as a coupling influence factor. A first-order dynamic system is used to construct a humidity dynamic model, and the external environment humidity is introduced as a disturbance input of heat penetration. The instantaneous humidity disturbance is used as a disturbance term (such as opening the door causing moisture to enter, sudden operation of the humidifier / dehumidifier), to predict the dynamic change of the cavity relative humidity over time;

[0087] In this embodiment, the humidity dynamic model is: ;

[0088] In the formula, denotes the cavity relative humidity the rate of change over continuous time , and denotes the real-time relative humidity of the cavity (unit: %RH) within the continuous time ; denotes the humidity first-order time constant (unit: seconds), which is used to describe the humidity system response speed and the smoothness of humidity change; denotes the target temperature of the cavity (i.e. the temperature target value set by the device); denotes the temperature deviation coupling coefficient (unit: %RH / ℃); denotes the external environment humidity within the continuous time ; denotes the humidity deviation coefficient (dimensionless), which represents the influence weight of the deviation between the current humidity of the cavity and the target humidity on humidity adjustment; denotes the target relative humidity of the cavity within the continuous time , i.e. the humidity control target set by the device (unit: %RH); denotes the environmental humidity coupling coefficient (dimensionless), which represents the penetration effect of the external environment humidity on the cavity humidity, such as the influence of laboratory air humidity change on the cavity humidity; denotes the instantaneous humidity disturbance (unit: %RH);

[0089] By using Euler forward method to process real-time data: ;

[0090] In the formula, represents the relative humidity of the cavity at the discrete sampling time ; represents the relative humidity of the cavity at the discrete sampling time ; represents the external environment humidity at the discrete sampling time ; represents the instantaneous humidity disturbance at the discrete sampling time ; represents the temperature-humidity coupling coefficient, which is calibrated based on historical operation data by batch least squares method , and for the same reason);

[0091] The temperature dynamic model and the humidity dynamic model are superimposed and fused, and a temperature-humidity coupling dynamic model is constructed by introducing a temperature-humidity coupling coefficient (obtained by artificial calibration or calculation). In this embodiment, the temperature-humidity coupling dynamic model calculates the temperature node prediction value based on the current device power and environmental conditions, takes the temperature prediction value as the input, calculates the humidity prediction value in combination with the environmental humidity and the device heat exchange state, adjusts the humidity prediction value based on the temperature-humidity interaction, generates the corrected coupling prediction result, and takes the temperature, humidity and key environmental parameters output by the temperature-humidity coupling dynamic model as the theoretical reference value for system operation;

[0092] The temperature-humidity coupling dynamic model is parameter corrected based on the least squares method, which is used to calibrate the cavity heat capacity, heat transfer coefficient, temperature-humidity coupling coefficient and disturbance correction coefficient (including instantaneous humidity disturbance , instantaneous disturbance term (i.e. temperature disturbance term));

[0093] The real-time operation parameters (temperature, humidity, compressor state, fan speed and other operation parameters) collected from the device sensors and environmental sensors are received by the real-time data mapping module 22, and the real-time operation parameters are time resampled, outlier removed and normalized pretreated. The pretreated real-time operation parameters are taken as the input of the calibrated temperature-humidity coupling dynamic model, and the real-time state data (including cavity real-time temperature, cavity real-time humidity and real-time state vector corrected by the temperature-humidity coupling dynamic model; also including the prediction value of the compressor duty ratio, fan speed and other control variables) are output.

[0094] The temperature and humidity coupling dynamic model is based on current real-time input (real-time running parameters (compressor, fan, environmental conditions, etc.) collected at the current time) and the state at the last time (cavity temperature, humidity), to calculate the device cavity state (output real-time temperature and real-time humidity) at the current time;

[0095] A temperature and humidity joint PID controller is constructed, taking the real-time temperature and humidity error output by the temperature and humidity coupling dynamic model as input, and outputting the compressor duty ratio, fan setting and dehumidification / humidification instructions, which are converted into actual switch instructions through duty ratio-switch mapping logic.

[0096] Further, the temperature and humidity joint PID controller is composed of a temperature control channel and a humidity control channel in parallel;

[0097] The temperature control channel is based on the deviation between the target temperature of the cavity and the temperature output by the temperature dynamic model, and introduces a humidity feedforward compensation term, to construct a temperature PID controller for generating a compressor duty ratio adjustment signal.

[0098] The humidity control channel is based on the deviation between the target humidity of the cavity and the humidity output by the humidity dynamic model, and introduces a temperature feedforward compensation term, to construct a humidity PID controller for generating a humidification / dehumidification adjustment signal.

[0099] In this embodiment, to realize accurate adjustment of the temperature and humidity of the device cavity, a discrete sampling time and a sampling interval (unit: seconds, indicating the time resolution of the controller execution and sensor sampling) are defined in advance; for the temperature control channel, the corresponding discrete error is defined as follows:

[0100]

[0101]

[0102] In the formula, represents the temperature error, and a positive value indicates overheating, which requires increased refrigeration; represents the temperature prediction value (or real-time measurement value) of the cavity at the th sampling time; represents the set target temperature; represents the set target humidity; represents the relative humidity prediction value (or real-time measurement value) of the cavity at the th sampling time; represents the humidity error, and a positive value indicates low humidity, which requires increased humidification;

[0103] In the formula, the temperature channel is based on the temperature error The temperature PID controller is constructed, and the humidity deviation is introduced as a feedforward compensation term, and the control law is:

[0104]

[0105]

[0106] In the formula, represents the original control quantity of the compressor duty cycle, the range is , wherein 0 represents off, and 1 represents full-power operation; represents the proportional gain parameter of the temperature control channel; represents the integral gain parameter of the temperature control channel; represents the differential gain parameter of the temperature control channel; represents the humidity feedforward compensation coefficient in the temperature control channel (when the humidity is high, the condensation load of the cavity will increase, and the feedforward compensation can be compensated in advance to avoid lag); represents the humidity feedforward compensation term of the temperature channel; represents the temperature error at the discrete sampling time ; represents the temperature control error at the th discrete sampling time;

[0107] The actual compressor duty cycle control instruction is obtained through limiting processing:

[0108]

[0109] In the formula, represents the actual duty cycle control instruction applied to the compressor after limiting processing; represents a limiting function (saturation function), that is, the original calculated compressor duty cycle control quantity is limited in the range of ;

[0110] Further, for the humidity control channel, based on the humidity error , a humidity PID controller is constructed, and the temperature deviation and the temperature change rate are introduced as feedforward compensation terms, and the control law is:

[0111]

[0112] , wherein

[0113] In the formula, represents the original control quantity of the humidifying / dehumidifying equipment, the range is , wherein -1 represents maximum dehumidification, and +1 represents maximum humidification; represents the temperature feedforward compensation term of the humidity channel; a proportional gain parameter representing the humidity control channel; an integral gain parameter representing the humidity control channel; a differential gain parameter representing the humidity control channel; a humidity control error representing the discrete sampling time instant; a humidity error representing the humidity error at the discrete sampling time instant; a temperature prediction value (or real-time measurement) representing the temperature within the chamber at the sampling time instant; the above represents the current discrete sampling time, an index representing the discrete sampling time instant;

[0114] an actual humidification / dehumidification control instruction obtained after limiting;

[0115]

[0116] wherein, represents the actual duty cycle control instruction applied to the compressor after limiting; represents a limiting function (saturation function), i.e., limiting the original calculated control quantity to the interval;

[0117] the feedforward compensation coefficients , and are calibrated through historical operation data based on a least square identification method.

[0118] The digital twin scientific research equipment intelligent management system fusing the Internet of Things and artificial intelligence further comprises an AI multi-dimensional prediction unit 3, which constructs a CNN-LSTM prediction model based on real-time state data generated by the digital twin mapping unit 2 and collected multi-source heterogeneous environment data, predicts future operation states of the equipment and environmental risks, and provides input for the risk early warning unit 4;

[0119] In the embodiment, the AI multi-dimensional prediction unit 3 comprises an equipment state prediction module 31, an environmental risk prediction module 32, and a multi-dimensional prediction fusion module 33;

[0120] The equipment state prediction module 31 constructs a prediction input sequence based on real-time state data output by the digital twin mapping unit 2, constructs a CNN-LSTM prediction model based on the prediction input sequence to predict key parameters such as temperature, humidity, compressor operation state, and fan speed of the scientific research equipment chamber, and outputs an equipment state prediction sequence to realize prediction of future operation trends.

[0121] The environmental risk prediction module 32 is based on the impact of external environmental conditions (multi-source heterogeneous environmental data collected by sensors, specifically including laboratory ambient temperature, laboratory ambient humidity, laboratory air pressure, air quality parameters, time characteristic parameters, and seasonal characteristic parameters) on the operation of scientific research equipment. By introducing Fourier basis functions (considering environmental parameters with periodic fluctuations such as diurnal temperature difference and seasonal humidity changes), it constructs a multivariate time series prediction model and outputs an environmental risk indicator prediction sequence for early warning of abnormal temperature and humidity fluctuations.

[0122] In this embodiment, the environmental risk prediction module 32 predicts risk factors in the operating environment of scientific research equipment that may lead to abnormal equipment status based on multi-source heterogeneous environmental data, and outputs future risk prediction results. Step-by-step environmental risk prediction module sequence ;

[0123] Specifically, it is based on multi-source heterogeneous environmental data (based on laboratory environmental parameters collected by the digital twin mapping unit 2, including ambient temperature, relative humidity, airflow velocity, air particle concentration, and vibration (noise index)), and incorporates external environmental disturbance information. Constructing the environment state vector ;

[0124] Based on the environmental state vector sequence, a sliding window (length is...) is used... Concatenate the input prediction samples At the same time, for each time point Constructing Fourier features (Example periodic set) Retain each cycle (one harmonic), and then use the Fourier features with the input prediction samples. The extended features are obtained by concatenating them along the on / off dimension (while, to ensure model reliability, the Fourier features and the original sensor channels are normalized separately (Min-Max normalization), and the higher-order harmonic coefficients are regularized (L2 regularization)). An extended feature sample sequence is constructed using a sliding window, where the Fourier features... :

[0125]

[0126] In the formula, Indicates the current sampling time step (time index); Represents a periodic set; This represents the number of harmonics in each cycle; Indicates the harmonic index, with a value of 1- , indicating the retention of the former One harmonic;

[0127] The deep learning method is used to construct a multivariate time series prediction model (the same architecture of CNN-LSTM prediction model is selected as the device state prediction module 31, and on the basis of the CNN-LSTM prediction model of the device state prediction module 31, a new input head is extended (an environmental risk feature parallel input head is added), so that the device state feature and the environmental risk feature (that is, the extended feature sample sequence) are received in parallel through the multiple input heads, and then the device state prediction sequence and the environmental risk prediction module sequence are output through the multiple output heads), the time sequence evolution characteristics of the environmental state sequence are modeled, and the influence of the key disturbance moment on the environmental risk is highlighted through the attention mechanism to obtain the prediction output ;

[0128] The multi-dimensional prediction fusion module 33 is used to fuse the prediction results output by the device state prediction module 31 and the environmental risk prediction module 32 to obtain a comprehensive risk prediction index.

[0129] In this embodiment, the prediction results of the device state prediction module 31 and the environmental risk prediction module 32 are fused:

[0130]

[0131] In the formula, represents the sampling moment The comprehensive risk prediction index of the future step is used to reflect the comprehensive evaluation result of the device state abnormal risk and the environmental risk; represents the future step prediction device state output by the device state prediction module 31 is mapped into a risk index function, which can be calculated by normalization, threshold judgment or empirical risk scoring method, and reflects the abnormal degree of the device running state; represents the weight coefficient of the device state prediction result in the comprehensive risk index, which can be set by historical data or expert experience, and the value range is , which is used to adjust the contribution degree of the device state prediction to the comprehensive risk; represents the weight coefficient of the environmental risk prediction module result in the comprehensive risk index, which is set by historical data or expert experience, and the value range is , which is used to adjust the contribution degree of the environmental risk prediction module to the comprehensive risk, and ; represents the environmental risk index of the future step output by the environmental risk prediction module 32, which reflects the influence degree of the laboratory environment or external conditions on the device running risk.

[0132] The specific steps of the CNN-LSTM prediction model outputting the device state prediction sequence are:

[0133] Based on real-time state data (operating state data output by the digital twin mapping unit 2: including the temperature in the cavity, the relative humidity in the cavity, the compressor operating state parameters, the fan operating parameters, the device door operation parameters, and the humidity mutation indication parameters), a prediction input sequence is constructed , and the prediction input sequence is spliced with a sliding window to obtain an input prediction sample , and principal component analysis (PCA) or an autoencoder is used for dimension reduction to obtain low-dimensional features ;

[0134] In this embodiment,

[0135]

[0136] In the formula, represents the temperature in the cavity at the sampling time ; represents the relative humidity in the cavity at the sampling time ; represents the compressor operating state parameter (on-off, speed, or power) at the sampling time ; represents the fan operating parameter (speed, air volume) at the sampling time ; represents the relative temperature of the laboratory environment at the sampling time ; represents the relative humidity of the laboratory environment at the sampling time ; represents the device door operation parameter at the sampling time ; represents the humidity mutation indication parameter (humidity mutation signal caused by opening the door) at the sampling time ; represents the state vector at the sampling time , which represents the comprehensive state of the device ontology and the environment at that time;

[0137] The sliding window length is set to , and the input prediction sample ;

[0138] In the formula, represents the state vector at the sampling time , which represents the combination of the device operating state and the environmental characteristics at that time; represents the prediction input sequence constructed at the current sampling time , which contains the state vectors of the past times;

[0139] A CNN-LSTM prediction model including a convolutional feature extraction subnetwork and a recurrent neural network is constructed;

[0140] The low-dimensional features are extracted by the convolutional feature extraction subnetwork The convolutional operation is performed, the input data at adjacent time points are calculated by sliding (the length of the one-dimensional convolution kernel is , the number of convolution kernels (the number of channels) , and the local feature vector of the temperature and humidity change over time is extracted, and a local feature mapping matrix is output;

[0141] The local feature mapping matrix is input into the recurrent neural network to generate a time series feature sequence;

[0142] In this embodiment, the recurrent neural network is a long short-term memory network (LSTM), which is composed of a plurality of layers of gate units in series, the hidden state dimension of each layer is , and the number of layers is ; at each time point , the recurrent neural network subnetwork calculates the activation values of the input gate, the forget gate, the output gate and the candidate memory unit based on the current input feature vector and the hidden state at the previous time point, thereby updating the cell state and the hidden state, and modeling the long-term and short-term dependence relationship of the dynamic evolution process of the temperature and humidity;

[0143] And by introducing an attention mechanism (preferably using additive attention (Bahdanau type), wherein the attention mechanism subnetwork includes attention heads, the hidden dimension of each head is the same as that of the recurrent neural network , and the weight parameters are learned through the training data), which is used to assign different weights to the features corresponding to each time point in the time series feature sequence, to obtain weighted time series features, so as to highlight the key disturbance time points that have a significant impact on the device running state;

[0144] The weighted time series features are sent to a fully connected output layer to obtain the device running state prediction value at each time point within the future step prediction period;

[0145] The device running state prediction output is:

[0146]

[0147] wherein, ; in the formula, indicates the device state (temperature, humidity, etc.) predicted in the future steps; indicates a neural network prediction function, wherein the neural network prediction function includes a convolutional feature extraction subnetwork, a recurrent neural network, and an attention mechanism subnetwork;

[0148] The CNN-LSTM prediction model is trained and its parameters are optimized using a weighted mean square error loss function. The weighted mean square error measures the difference between the predicted and actual values. By minimizing this loss function, the parameters of the CNN-LSTM prediction model are iteratively updated. Specifically, the weighted mean square error loss function is used to calculate the deviation between the predicted output value and the actual observed value. Different weight parameters are assigned to different prediction times according to the prediction step size, so that the prediction model can ensure both short-term prediction accuracy and long-term prediction stability. The weight parameters are manually set based on empirical rules, or they can be automatically calculated based on the training dataset through parameter optimization methods (calculation methods based on historical running data include determining the weight factors based on the error variance of each prediction time, adaptive decay function, etc.) so as to balance short-term prediction accuracy and long-term prediction stability in different application scenarios.

[0149] The predicted values ​​for equipment operating status include the predicted values ​​for temperature and humidity of the target equipment cavity.

[0150] The intelligent management system for digital twin scientific research equipment that integrates the Internet of Things and artificial intelligence also includes a risk warning unit 4. Based on the prediction results of the AI ​​multidimensional prediction unit 3, the risk warning unit 4 performs real-time analysis on the operating status of scientific research equipment and its environmental coupling risks, and issues a warning signal before potential anomalies occur.

[0151] In this embodiment, the risk warning unit 4 includes an anomaly detection module 41 and a risk warning module 42;

[0152] Among them, the anomaly detection module 41 receives the model baseline output from the digital twin mapping unit 2 and the prediction result from the AI ​​multidimensional prediction unit 3, and performs difference analysis in combination with the actual collected values.

[0153] Specifically, in the anomaly detection module 41, the predicted results are combined with the actual collected values ​​to perform difference analysis. The specific steps involved are as follows:

[0154] The output of the digital twin mapping unit 2 at time 2 is obtained respectively. Theoretical reference values ​​for temperature and humidity, and the output of the AI ​​multidimensional prediction unit 3 at time 3. The predicted values ​​of the monitored variables (predicted values ​​of comprehensive operating parameters such as temperature, humidity and key environmental parameters) and the actual collected values ​​of the monitored variables (monitored variables refer to comprehensive operating parameters such as temperature, humidity and key environmental parameters) collected by the multi-source heterogeneous data acquisition unit 1.

[0155] When the deviation of the actual collected value from the twin model reference value (used to represent the deviation degree of the actual operation of the system from the mechanism model) exceeds the preset threshold value, it is determined by chi-square test whether there is a trend deviation, and if the chi-square statistic is greater than the critical value corresponding to the significance level, it is determined that there is a trend deviation;

[0156] When the residual error of the actual collected value and the AI prediction result exceeds the preset threshold value in multiple time steps (used to represent the difference between the system operation and the prediction trend), the mean square error (MSE) is used for quantitative evaluation, and the single-class support vector machine is used for abnormal pattern recognition on the residual error sequence, and if the recognition result is abnormal, it indicates that the prediction and the actual operation trend deviate significantly;

[0157] If any of the above is abnormal, an abnormal signal is output to the risk warning module 42;

[0158] Specifically, the risk warning module 42 outputs multi-level warning information according to the determination result of the abnormality detection module 41 and in combination with the risk grading rules:

[0159] Primary warning: when a single detection method determines an abnormality and the deviation amplitude is small, the system prompts mild risk information;

[0160] Secondary warning: when at least two detection methods simultaneously determine an abnormality and the deviation trend is obvious, the system prompts moderate risk and provides operation optimization suggestions to the control mechanism;

[0161] Tertiary warning: when multiple detection methods consistently determine a serious abnormality and the deviation exceeds a high threshold value, the system prompts a serious risk and triggers an emergency control strategy through a control linkage mechanism, including reducing the load, switching the operation mode or stopping the protection.

[0162] The basic principles, main features and advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited by the above embodiments, and the above embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application.

Claims

1. A digital twin intelligent management system for scientific research equipment integrating the Internet of Things and artificial intelligence, characterized in that, include: Multi-source heterogeneous data acquisition unit (1) is used to collect real-time information on the operating environment and equipment status of scientific research equipment, and obtain operating status parameters and multi-dimensional external parameters; The digital twin mapping unit (2) is based on the operating status parameters and multi-dimensional external parameters. By introducing the temperature and humidity coupling coefficient, it constructs a temperature and humidity coupled dynamic model that integrates the temperature dynamic model and the humidity dynamic model. It then maps the data collected from multiple sources and the laboratory environment to the temperature and humidity coupled dynamic model and finally outputs real-time status data. Meanwhile, a temperature and humidity joint PID controller is constructed, which takes the real-time temperature and humidity error output by the temperature and humidity coupled dynamic model as input and outputs the compressor duty cycle, fan setting and dehumidification / humidification commands. Among them, the temperature dynamic model is based on a first-order dynamic system and is constructed by introducing external ambient temperature and instantaneous disturbance terms; The humidity dynamic model is based on a first-order dynamic system, with the rate of temperature change as a coupling influencing factor, and incorporates external environmental humidity and instantaneous humidity disturbances. Furthermore, the temperature and humidity combined PID controller is composed of a temperature control channel and a humidity control channel connected in parallel; The temperature control channel is based on the deviation between the target temperature of the cavity and the output temperature of the temperature dynamic model, and introduces a humidity feedforward compensation term to construct a temperature PID controller, which is used to generate the compressor duty cycle adjustment signal. The humidity control channel is based on the deviation between the target humidity of the cavity and the output humidity of the humidity dynamic model, and a temperature feedforward compensation term is introduced to construct a humidity PID controller for generating humidification / dehumidification adjustment signals. AI multidimensional prediction unit (3), the AI ​​multidimensional prediction unit (3) constructs a CNN-LSTM prediction model based on the real-time status data generated by the digital twin mapping unit (2) and the collected multi-source heterogeneous environmental data, and predicts the future operating status and environmental risks of the equipment; Risk warning unit (4) Based on the prediction results of AI multidimensional prediction unit (3), the risk warning unit (4) performs real-time analysis on the operating status of scientific research equipment and its environmental coupling risks, and issues a warning signal before potential anomalies occur.

2. The intelligent management system for digital twin scientific research equipment integrating the Internet of Things and artificial intelligence as described in claim 1, characterized in that, The multi-source heterogeneous data acquisition unit (1) includes a device data sensing module (11) and an environmental sensing module (12). The equipment data sensing module (11) is used to collect the operating status parameters of multi-source equipment and its execution components. The operating status parameters include compressor parameters, air supply system parameters, heat exchanger parameters, energy consumption data and operating mode information. The environmental sensing module (12) is used to collect multi-dimensional external parameters of the laboratory environment in which the equipment is located. The multi-dimensional external parameters include environmental parameters, air quality parameters, aerodynamic parameters and environmental interference factors.

3. The intelligent management system for digital twin scientific research equipment integrating the Internet of Things and artificial intelligence as described in claim 1, characterized in that, The digital twin mapping unit (2) includes a physical mechanism modeling module (21) and a real-time data mapping module (22). The physical mechanism modeling module (21) establishes a temperature and humidity coupled dynamic model according to the type of laboratory equipment. The temperature and humidity coupled dynamic model is adapted to different equipment through parameter configuration. The real-time data mapping module (22) maps the data collected from multiple sources and the laboratory environment to the temperature and humidity coupled dynamic model, which is used to realize the synchronous output of the real-time status data of the equipment.

4. The intelligent management system for digital twin scientific research equipment integrating the Internet of Things and artificial intelligence as described in claim 3, characterized in that, The specific steps for constructing the temperature and humidity coupled dynamic model and data mapping are as follows: Acquire historical operating data of the equipment: temperature, humidity, compressor status and fan speed, form the original input vector, and preprocess the original input vector; Based on the preprocessed original input vector, the rate of change of temperature over time and the deviation between the current temperature and the target temperature are calculated. Based on the rate of change of temperature over time and the deviation between the current temperature and the target temperature, a temperature dynamic model is constructed using a first-order dynamic system. The external ambient temperature is introduced into the temperature dynamic model as a disturbance input for heat penetration, and the door opening event is used as an instantaneous disturbance term to predict the dynamic change of cavity temperature over time. Based on the preprocessed original input vector, the rate of change of relative humidity of the cavity over time is calculated. Based on the deviation between the current humidity and the target humidity, and with the temperature change rate as a coupling influence factor, a first-order dynamic system is used to construct a humidity dynamic model. The external ambient humidity is introduced into the humidity dynamic model as a disturbance input for heat penetration. The instantaneous humidity disturbance is used as a disturbance term to predict the dynamic change of relative humidity of the cavity over time. The temperature dynamic model and the humidity dynamic model are superimposed and fused, and a temperature and humidity coupled dynamic model is constructed by introducing a temperature and humidity coupling coefficient. The parameters of the temperature and humidity coupled dynamic model are corrected based on the least squares method, which are used to calibrate the cavity heat capacity, heat transfer coefficient, temperature and humidity coupling coefficient and disturbance correction coefficient. The real-time data mapping module (22) receives real-time operating parameters collected from equipment sensors and environmental sensors, preprocesses the real-time operating parameters, uses the preprocessed real-time operating parameters as input to the calibrated temperature and humidity coupled dynamic model, and outputs real-time status data. The temperature and humidity coupled dynamic model calculates the current state of the equipment cavity based on the current real-time input and the state at the previous moment.

5. The intelligent management system for digital twin scientific research equipment integrating the Internet of Things and artificial intelligence as described in claim 1, characterized in that, The AI ​​multidimensional prediction unit (3) includes an equipment status prediction module (31), an environmental risk prediction module (32), and a multidimensional prediction fusion module (33). Among them, the equipment status prediction module (31) constructs a prediction input sequence based on the real-time status data output by the digital twin mapping unit (2), and constructs a CNN-LSTM prediction model based on the prediction input sequence to predict the temperature, humidity, compressor operating status and fan speed of the scientific research equipment cavity, and outputs the equipment status prediction sequence. The environmental risk prediction module (32) is based on the impact of external environmental conditions on the operation of scientific research equipment, and by introducing Fourier basis functions, it constructs a multivariate time series prediction model and outputs environmental risk index prediction sequences for early warning of abnormal temperature and humidity fluctuations. The multidimensional prediction fusion module (33) is used to fuse the prediction results output by the equipment status prediction module (31) and the environmental risk prediction module (32) to obtain a comprehensive risk prediction index.

6. The intelligent management system for digital twin scientific research equipment integrating the Internet of Things and artificial intelligence as described in claim 5, characterized in that, The specific steps involved in the output device state prediction sequence of the CNN-LSTM prediction model are as follows: Based on real-time state data, a predicted input sequence is constructed, and a sliding window is used to stitch the predicted input sequence to obtain the input predicted sample. Principal component analysis is then used to obtain low-dimensional features. Construct a CNN-LSTM prediction model that includes a convolutional feature extraction subnetwork and a recurrent neural network; The convolutional feature extraction subnetwork performs convolution operations on low-dimensional features to extract local feature vectors of temperature and humidity changes over time, and outputs a local feature mapping matrix. The local feature mapping matrix is ​​input into a recurrent neural network to generate a temporal feature sequence. Furthermore, by introducing an attention mechanism, different weights are assigned to the features corresponding to each time step in the time-series feature sequence to obtain weighted time-series features; The weighted temporal features are fed into the fully connected output layer to obtain the predicted values ​​of the device operating status at each time point within the future step prediction time. The predicted values ​​for equipment operating status include the predicted values ​​for temperature and humidity of the target equipment cavity.

7. The intelligent management system for digital twin scientific research equipment integrating the Internet of Things and artificial intelligence as described in claim 1, characterized in that, The risk warning unit (4) includes an anomaly detection module (41) and a risk warning module (42). Among them, the anomaly detection module (41) receives the model baseline output from the digital twin mapping unit (2) and the prediction result from the AI ​​multidimensional prediction unit (3), and performs difference analysis in combination with the actual collected values; The risk warning module (42) outputs multi-level warning information based on the judgment result of the anomaly detection module (41) and the risk classification rules.

8. The intelligent management system for digital twin scientific research equipment integrating the Internet of Things and artificial intelligence as described in claim 7, characterized in that, In the anomaly detection module (41), the predicted results are combined with the actual collected values ​​for difference analysis. The specific steps involved are as follows: The theoretical reference values ​​of temperature and humidity at time t are obtained from the output of the digital twin mapping unit (2), the predicted values ​​of the monitoring variables at time t are obtained from the output of the AI ​​multidimensional prediction unit (3), and the actual collected values ​​of the monitoring variables collected by the multi-source heterogeneous data acquisition unit (1). When the deviation between the actual collected value and the reference value of the twin model exceeds a preset threshold, the chi-square test is used to determine whether there is a trend of deviation in the system. When the residual between the actual collected value and the AI ​​prediction result exceeds the preset threshold in multiple time steps, the mean square error is used for quantitative evaluation, and a single-class support vector machine is combined to identify abnormal patterns in the residual sequence. If any of the above items are found to be abnormal, an abnormal signal is output to the risk warning module (42).

Images