A clean room intelligent operation and maintenance system and method based on digital twin

Abstract

The present invention discloses a clean room intelligent operation and maintenance system and method based on digital twins, which relates to the field of intelligent operation and maintenance technology. The system and method include acquiring clean room status data by arranging sensors, acquiring clean room equipment power data through external data sources, and preprocessing at edge nodes to generate a multi-source time series data set; constructing a multi-scale prediction model by integrating long short-term memory networks, autoregressive integral moving average models, and graph neural networks, and generating a multi-scale sequence data set based on the multi-source time series data set; simulating the physical field coupling in the clean room through a quantum computing platform to generate a clean room digital twin simulation model; loading the clean room digital twin simulation model into a cloud analysis platform, using the gradient descent method to optimize the clean room operating status, and generating an optimized operation and maintenance plan. The present invention simulates the physical field coupling in the clean room through a quantum computing platform, greatly improving the authenticity and real-time performance of the simulation model.

Description

Technical Field

[0001] The present invention relates to the field of intelligent operation and maintenance technology, and in particular to a clean room intelligent operation and maintenance system and method based on digital twins. Background Art

[0002] Digital twin technology, a real-time mapping tool for physical entities and virtual spaces, has been widely used in recent years in fields such as industrial manufacturing. In cleanroom operations and maintenance, traditional methods rely primarily on manual inspections and fixed threshold controls, making them incapable of handling the dynamic changes in complex environmental parameters. With the maturity of the Internet of Things (IoT) and edge computing technologies, intelligent operations and maintenance systems based on multi-source sensor data collection and real-time processing are gradually emerging.

[0003] Currently, most digital twin simulations are based on classical computing frameworks (such as finite element analysis), and are still immature in handling high-dimensional nonlinear coupling problems. For example, the distribution of air pollutant concentration in a clean room needs to consider filter efficiency, equipment power fluctuations, and spatial convection and diffusion. However, the computational complexity of classical simulation tools increases exponentially with the variable dimension, resulting in insufficient real-time performance. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a clean room intelligent operation and maintenance method based on digital twins to solve the problem of insufficient real-time performance of traditional digital twin simulations.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a clean room intelligent operation and maintenance method based on digital twins, which includes:

[0008] Acquire cleanroom status data and cleanroom equipment power data through external data sources, and pre-process them at the edge node to generate a multi-source time series dataset;

[0009] By integrating long short-term memory networks, autoregressive integrated moving average models, and graph neural networks, a multi-scale prediction model is constructed to generate multi-scale sequence datasets based on multi-source time series datasets.

[0010] On the quantum computing platform, we encoded multi-scale sequence data sets, configured quantum circuits, and calculated the power-efficiency coupling factor using a quantum Monte Carlo algorithm. Using three-dimensional linear interpolation and a convection-diffusion algorithm, we estimated the temperature, humidity, and air pollutant concentration distributions within the cleanroom. We generated temperature and humidity distribution maps and air pollutant concentration distribution maps, and verified their accuracy.

[0011] A 3D model of the cleanroom is constructed using 3D modeling software. The temperature and humidity distribution maps and air pollutant concentration distribution maps are then integrated into the 3D model to obtain a digital twin simulation model of the cleanroom.

[0012] The cleanroom digital twin simulation model is loaded onto the cloud analysis platform, and the gradient descent method is used to optimize the cleanroom operating status and generate an optimized operation and maintenance plan.

[0013] The optimized operation and maintenance plan is sent to the clean room equipment for execution, and after execution, the equipment will feedback the real-time status to the clean room digital twin simulation model.

[0014] As a preferred solution of the clean room intelligent operation and maintenance method based on digital twins of the present invention, the specific steps of generating a multi-scale sequence data set are as follows:

[0015] The multi-scale prediction model includes a data allocation layer, a short-term state prediction layer, a medium-term state prediction layer and a long-term state prediction layer;

[0016] Through the data interface, multi-source time series data sets are transmitted to the data distribution layer in chronological order second by second for data diversion and distribution;

[0017] The short-term, medium-term and long-term states of the clean room are predicted respectively at the short-term state prediction layer, the medium-term state prediction layer and the long-term state prediction layer to obtain the short-term, medium-term and long-term state prediction sequences of the clean room;

[0018] A structured aggregation method is used to classify and store short-term state prediction sequences, medium-term state prediction sequences, and long-term state prediction sequences according to time scale and target variable, and a multi-scale sequence dataset is output.

[0019] As a preferred solution of the clean room intelligent operation and maintenance method based on digital twins of the present invention, the specific steps of calculating the power and efficiency coupling factor are as follows:

[0020] The probability sequences of cleanroom equipment power exceeding the standard and air filter efficiency decreasing in the multi-scale sequence dataset are mapped to quantum states respectively, and the quantum encoding of the probability of cleanroom equipment power exceeding the standard and the probability of air filter efficiency decreasing are obtained.

[0021] Based on the quantum coding of the probability of clean room equipment power exceeding the standard and the probability of air filter efficiency decreasing, the quantum Monte Carlo algorithm is used to calculate the power and efficiency coupling factor;

[0022] According to the timestamp, the power and efficiency coupling factors at all times are integrated to obtain the coupling factor sequence.

[0023] As a preferred solution of the clean room intelligent operation and maintenance method based on digital twins described in the present invention, the temperature, humidity and air pollutant concentration distribution of the clean room indoor space are estimated by using the three-dimensional linear interpolation method and the convection diffusion algorithm. The specific steps are as follows:

[0024] Based on the coupling factor sequence, the correction values ​​of clean room indoor temperature and humidity and air pollutant concentration are calculated respectively to obtain the corrected temperature sequence, corrected humidity sequence and corrected air pollutant concentration sequence;

[0025] The temperature and humidity distribution and air pollutant concentration distribution of the entire clean room space are estimated using the three-dimensional linear interpolation method;

[0026] The concentration distribution of air pollutants in the clean room is calculated by the convection diffusion algorithm to obtain the distribution of air pollutant concentrations;

[0027] The temperature and humidity distribution map and the air pollutant concentration distribution map are verified for accuracy. If the accuracy verification fails, the number of quantum Monte Carlo algorithm iterations is increased until the accuracy verification passes.

[0028] As a preferred solution of the clean room intelligent operation and maintenance method based on digital twins of the present invention, the temperature and humidity distribution map and the air pollutant concentration distribution map are integrated into the three-dimensional model of the clean room. The specific steps are as follows:

[0029] Based on the type and size of the cleanroom space, as well as the layout of various sensors and equipment, a 3D model of the cleanroom is constructed using 3D model building software.

[0030] The corrected temperature sequence, corrected humidity sequence, and corrected air pollutant concentration sequence are stored at each coordinate point of the clean room three-dimensional model and represented by a NumPy three-dimensional array;

[0031] Create a metadata set containing spatial information and simulation conditions;

[0032] The metadata dataset, NumPy three-dimensional array, and simulation parameter set are bundled and stored as Python dictionary objects. They are serialized into a single file using the HDF5 format and output as the cleanroom digital twin simulation model.

[0033] As a preferred solution of the clean room intelligent operation and maintenance method based on digital twins of the present invention, the specific steps of generating an optimized operation and maintenance plan are as follows:

[0034] Set cleanliness targets based on the cleanroom's requirements for indoor temperature, humidity, and air pollutant concentrations;

[0035] Define the cleanliness loss function;

[0036] The gradient descent method is used to optimize the cleanliness target and obtain the optimal power of the temperature and humidity control equipment and air filter.

[0037] As a preferred solution of the clean room intelligent operation and maintenance method based on digital twins described in the present invention, the preprocessing of the clean room status data and the clean room equipment power data includes denoising and normalization processing.

[0038] In the second aspect, the present invention provides a clean room intelligent operation and maintenance system based on digital twins, including a data acquisition module, a state prediction module, a distribution map module, a simulation module, an analysis and optimization module, and a solution execution module.

[0039] The data acquisition module is used to acquire clean room status data by arranging sensors, and simultaneously acquire clean room equipment power data through external data sources, and perform preprocessing at the edge node to generate a multi-source time series data set;

[0040] The state prediction module is used to construct a multi-scale prediction model by integrating a long short-term memory network, an autoregressive integrated moving average model and a graph neural network, and generate a multi-scale sequence dataset based on a multi-source time series dataset;

[0041] The distribution map module is used to encode multi-scale sequence data sets on the quantum computing platform, configure quantum circuits, calculate the power and efficiency coupling factor using the quantum Monte Carlo algorithm, and use the three-dimensional linear interpolation method and convection-diffusion algorithm to estimate the temperature, humidity and air pollutant concentration distribution in the clean room indoor space, obtain temperature and humidity distribution maps and air pollutant concentration distribution maps, and verify their accuracy;

[0042] The simulation module is used to construct a three-dimensional model of the clean room using three-dimensional model construction software, and integrate the temperature and humidity distribution map and the air pollutant concentration distribution map into the three-dimensional model of the clean room to obtain a digital twin simulation model of the clean room;

[0043] The analysis and optimization module is used to load the clean room digital twin simulation model into the cloud analysis platform, optimize the clean room operating status using the gradient descent method, and generate an optimized operation and maintenance plan;

[0044] The solution execution module is used to send the optimized operation and maintenance solution to the clean room equipment for execution, and after the equipment is executed, it feeds back the real-time status to the clean room digital twin simulation model.

[0045] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the clean room intelligent operation and maintenance method based on digital twins as described in the first aspect of the present invention is implemented.

[0046] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the clean room intelligent operation and maintenance method based on digital twins as described in the first aspect of the present invention.

[0047] The beneficial effects of the present invention are: by integrating long short-term memory networks, autoregressive integral moving average models and graph neural networks to construct a multi-scale prediction model, various dynamic characteristics from instantaneous fluctuations to long-term trends can be effectively captured to meet different operation and maintenance needs. The physical field coupling in the clean room is simulated through a quantum computing platform, which greatly improves the authenticity and real-time performance of the simulation model. The optimized operation and maintenance plan is sent to the clean room equipment for execution, and the real-time status is fed back after the equipment is executed, forming a complete closed-loop control system to ensure continuous self-adjustment and optimization in actual operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 The figure is a flow chart of the clean room intelligent operation and maintenance method based on digital twin.

[0050] Figure 2 This is a module diagram of the clean room intelligent operation and maintenance system based on digital twins.

[0051] Figure 3 Flowchart for building a multi-scale prediction model.

[0052] Figure 4 Flowchart for generating a cleanroom digital twin simulation model. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0056] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a clean room intelligent operation and maintenance method based on digital twins, comprising the following steps:

[0057] S1. By arranging sensors, clean room status data is obtained. At the same time, clean room equipment power data is obtained through external data sources and preprocessed at the edge node to generate a multi-source time series dataset.

[0058] The clean room status data includes clean room temperature, humidity, air flow rate and air pollutant concentration at the clean room air filter inlet, and air flow rate and air pollutant concentration at the clean room air filter outlet;

[0059] Temperature sensors are installed in key areas of the clean room (such as the workbench surface, etc.), with one installed every 1 meter to ensure coverage of the entire space. Temperature values ​​are collected to reflect the temperature distribution in the clean room. Humidity sensors are placed near the air circulation path, such as near the return air vent, with one installed every 2 meters to measure the relative humidity in the clean room. Air pollutant concentration sensors (such as particulate matter concentration and harmful gas concentration, etc.) are placed at the air inlet and outlet of the clean room air filter to detect the concentration of air pollutants in the air. Wind speed sensors are installed at the air inlet and outlet of the clean room air filter to measure the air flow rate. All sensors are connected to the data collector via wired connections;

[0060] The voltage and current data of the cleanroom power supply system are collected through smart meters to calculate the power consumption of cleanroom equipment. Cleanroom equipment includes temperature control equipment (such as air conditioners), humidity control equipment (such as humidifiers and dehumidifiers), and air pollutant concentration control equipment (such as air filters).

[0061] Set up an edge computing node (such as a Raspberry Pi 4) in the cleanroom, close to the data collector to reduce transmission delays, and connect the data collector to the edge computing node using a standard Ethernet cable.

[0062] Run the Kalman filter algorithm program on the edge computing node to remove noise from the clean room status data and clean room equipment power data respectively;

[0063] Each parameter sequence in the denoised clean room indoor state data and clean room equipment power data is normalized using the minimum and maximum normalization formula, and timestamps are aligned to obtain a multi-source time series dataset.

[0064] S2. A multi-scale prediction model is constructed by integrating long short-term memory networks, autoregressive integrated moving average models and graph neural networks, and a multi-scale sequence dataset is generated based on multi-source time series datasets.

[0065] The multi-scale prediction model includes a data allocation layer, a short-term state prediction layer, a medium-term state prediction layer, and a long-term state prediction layer;

[0066] The short-term state prediction layer is a long short-term memory network, which predicts the clean room's short-term state (minute level, such as within 10 minutes). The medium-term state prediction layer is an autoregressive integrated moving average model, which predicts the clean room's medium-term state (hour level, such as within 1 hour). The long-term state prediction layer is a graph neural network, which predicts the clean room's long-term state (day level, such as within 7 days).

[0067] Through the data interface, multi-source time series data sets are transmitted to the data distribution layer in chronological order second by second for data diversion and distribution;

[0068] Specifically, the clean room temperature and humidity are assigned to the short-term state prediction layer for predicting the short-term state for the next several minutes. The air flow rate and air pollutant concentration at the clean room air filter inlet, as well as the air flow rate and air pollutant concentration at the clean room air filter outlet, are assigned to the medium-term state prediction layer for predicting the medium-term state for the next several hours. The power data of the clean room equipment is assigned to the long-term state prediction layer for predicting the long-term state for the next several days.

[0069] Based on the data assigned to the short-term state prediction layer, a long short-term memory network is used to generate a short-term state prediction sequence;

[0070] Specifically, taking the short-term prediction of clean room temperature as an example, the clean room temperature is defined as the input sequence of the long short-term memory network. Start processing the input sequence second by second and calculate the forget gate output. The calculation formula is as follows:

[0071] ;

[0072] in, for The output of the momentary forget gate, the value range , represents the forgetting ratio, is the activation function of the long short-term memory network, Input weight matrix for forget gate, size is 50×1, (corresponding to 50 neurons and 1 input feature), value range , is the input sequence of the long short-term memory network, is the weight matrix of the hidden layer of the forget gate, with a size of 50×50 (corresponding to the connection between hidden layers) and a value range of , for The hidden state at each moment, the initial hidden state is a 50×1 zero vector, Forget gate bias, size is 50×1, value range , is the index variable of the moment;

[0073] Calculate the update gate output as follows:

[0074] ;

[0075] in, for Update the gate output at all times, the value range , represents the update ratio, To update the gate input weight matrix, the size is 50×1 (corresponding to 50 neurons and 1 input feature), the value range is , To update the gate hidden layer weight matrix, the size is 50×50 (corresponding to the connection between hidden layers), the value range is , is the update gate bias term;

[0076] Calculate the candidate state, the calculation formula is as follows:

[0077] ;

[0078] in, for Candidate state data value at the moment, value range , is the hyperbolic tangent function, Input weight matrix for candidate states, size is 50×1, (corresponding to 50 neurons and 1 input feature), value range , is the candidate state hidden layer weight matrix, is the candidate state bias item;

[0079] Update the unit state, the calculation formula is as follows:

[0080] ;

[0081] in, for The unit state at the moment, indicating the moment The initial cell state is a 50×1 zero vector. for The unit state at the moment;

[0082] Calculate the output gate, the calculation formula is as follows:

[0083] ;

[0084] in, for Output of the gate at all times, value range , The input weight matrix for the output gate is 50×1 (corresponding to 50 neurons and 1 input feature), with a value range of , is the output gate hidden layer weight matrix, with a size of 50×50 (corresponding to the connection between hidden layers), and a value range of , is the output gate bias term;

[0085] Based on the calculation result of the output gate and the cell state, the hidden state is updated. The calculation formula is as follows:

[0086] ;

[0087] in, for Always hide the status;

[0088] The predicted value is calculated through the pre-trained linear output layer. The calculation formula is as follows:

[0089] ;

[0090] in, for The predicted value at time, is the linear output layer weight matrix, is the linear output layer bias term;

[0091] The predicted clean room temperature and humidity are sorted into a clean room temperature and humidity prediction sequence according to the time series, i.e. the short-term state prediction result of the clean room;

[0092] Based on the data assigned to the mid-term state prediction layer, the autoregressive integrated moving average model is used to generate the probability series of air filter efficiency degradation;

[0093] Specifically, the probability of air filter efficiency decline is used as the prediction target to calculate the air filter efficiency sequence. The air filter efficiency is defined as:

[0094] ;

[0095] ;

[0096] ;

[0097] in, for Air filter efficiency at the moment, value range , for The particle flow rate at the air outlet at the time, for The particle flow rate at the air inlet at time and They are The air flow rate at the clean room air filter outlet and inlet at all times, and They are Check the concentration of air pollutants at the outlet and inlet of the clean room air filter at all times;

[0098] Calculate the first-order difference of the air filter efficiency series to obtain a first-order difference series, remove the linear trend (such as the slow decline caused by particulate matter accumulation) in it, and make the air filter efficiency series stable;

[0099] Input the first-order difference sequence into the pre-trained autoregressive integrated moving average model, and the air filter efficiency sequence for the next hour. Set the autoregressive order of the autoregressive integrated moving average model to 2, the difference order to 1, and the moving average order to 1. The calculation formula is as follows:

[0100] ;

[0101] in, for The predicted air filter efficiency value at all times, is the first-order autoregressive coefficient, and its value range is , and They are Moment and The air filter efficiency difference value at time, is the second-order autoregressive coefficient, and its value range is , for The white noise at the moment has a mean of 0 and a variance of 0.001, indicating random disturbance. is the sliding average coefficient, and its value range is ;

[0102] Similarly, the concentration of air pollutants at the air filter outlet of the clean room is predicted to generate an air pollutant concentration prediction sequence;

[0103] Calculate the probability of air filter efficiency decline using the following formula:

[0104] ;

[0105] in, for The probability that the air filter efficiency is lower than 0.9 at any given moment indicates a decreasing risk. is the cumulative function of the standard normal distribution, is the prediction standard deviation, with a value range of , based on ARIMA residual estimation;

[0106] The predicted probability of air filter efficiency degradation is organized into an air filter efficiency degradation probability sequence according to time series;

[0107] That is, the probability sequence of air filter efficiency decline and the air pollutant concentration prediction sequence together constitute the mid-term state prediction sequence of the clean room;

[0108] In the long-term state prediction layer, a graph neural network is used to generate a long-term state prediction sequence of the clean room based on the power of the clean room equipment;

[0109] Specifically, the main equipment components of the clean room are defined as nodes in the graph, such as the air conditioning compressor, fan, and electronic control unit (ECU), a total of three nodes. The edges between the nodes are defined as physical or electrical associations. For example, the air conditioning compressor drives the fan through current, and the fan affects the ECU through voltage, forming an edge set. A feature vector is initialized for each node. The power characteristics of the node are calculated based on the power sequence of the clean room equipment. Taking the air conditioning compressor as an example, it is assumed that its power accounts for 70% of the total power, the fan accounts for 20%, and the ECU accounts for 10%. The node power is calculated and the average value of the voltage and current is taken as the feature to complete the construction of the equipment association graph.

[0110] Based on pre-trained graph neural networks, the device association graph node status is updated and power trends are predicted;

[0111] Specifically, two hidden layers are set to capture the long-term dependencies between devices, and the calculation formula for updating the node state is:

[0112] ;

[0113] in, For nodes In the The hidden state of the layer, dimensionless, represents the updated value of the node feature, is the hidden layer index, is the device association graph node index, Device association graph node Neighbor node index, For nodes The set of neighbor nodes of is the graph neural network weight matrix, is the graph neural network offset, represents the graph neural network activation function, node In the The hidden state of the layer;

[0114] Based on the updated two hidden layers of the graph neural network ,Through linear output layer mapping, the probability sequence of clean room equipment power exceeding the standard is calculated, that is, the long-term state prediction sequence of the clean room;

[0115] ;

[0116] in, for Clean room equipment at all times Power at all times Exceeding the rated power of cleanroom equipment The probability of is the linear layer bias term, is the base of natural logarithm;

[0117] At the edge computing node, a structured aggregation method is used to classify and store short-term state prediction sequences, medium-term state prediction sequences, and long-term state prediction sequences according to time scale and target variable, and output a multi-scale sequence dataset.

[0118] To verify the accuracy of the multi-scale prediction model, multi-scale prediction simulation experiments were conducted. The experimental data showed that the model exhibited high accuracy in short-term temperature, humidity, medium-term pollutant concentration, and long-term power prediction. Representative data are shown in Table 1-3 (partial time step data):

[0119] Table 1 Comparison of short-term temperature and humidity forecasts and actual values

[0120]

[0121] Table 2 Comparison of mid-term pollutant forecasts and actual values

[0122]

[0123] Table 3 Comparison of long-term power prediction and actual values

[0124]

[0125] The overall accuracy evaluation is performed by calculating the coefficient of determination between the predicted value and the actual value, as shown in Table 4:

[0126] Table 4 Prediction accuracy evaluation table

[0127]

[0128] S3. Encode the multi-scale sequence data set on the quantum computing platform, configure the quantum circuit, and calculate the power and efficiency coupling factor through the quantum Monte Carlo algorithm. Use the three-dimensional linear interpolation method and the convection-diffusion algorithm to estimate the temperature, humidity and air pollutant concentration distribution in the clean room indoor space, obtain the temperature and humidity distribution map and the air pollutant concentration distribution map, and verify the accuracy.

[0129] Transferring multi-scale sequence datasets to a quantum computing platform (e.g., IBM Qiskit), configuring quantum circuits, and calculating power and efficiency coupling factors;

[0130] Specifically, the probability sequence of clean room equipment power exceeding the standard and the probability sequence of air filter efficiency degradation in the multi-scale sequence dataset are mapped to quantum states, respectively, and the calculation formula is as follows:

[0131] ;

[0132] ;

[0133] ;

[0134] in, for Quantum encoding of the probability of clean room equipment power exceeding the standard at any moment, for The quantum code of the probability of the air filter efficiency decreasing at any moment, is the imaginary unit, defined as , for Phase angle at the moment, unit radian, value range ,

[0135] The quantum Monte Carlo algorithm is used to calculate the power and efficiency coupling factors, and the power and efficiency coupling factors at all times are integrated to obtain the coupling factor sequence. The calculation formula is as follows:

[0136] ;

[0137] in, for The power and efficiency coupling factor at the moment, is the Hamiltonian operator;

[0138] Based on the variable dimension and time span of the multi-scale sequence data set transmitted to the quantum computing platform, the required number of quantum bits and the ratio of Hadamard gates are calculated as follows:

[0139] ;

[0140] ;

[0141] in, is the required number of qubits, represents the number of independent variables in the multi-scale sequence dataset, which is 4 here (short-term state prediction sequence, including the clean room indoor temperature and humidity prediction sequence, the clean room air filter efficiency degradation probability sequence, and the clean room equipment power exceeding probability sequence). is the maximum time span, i.e. the time span of the long-term state prediction layer, is the Hadamard gate ratio, with a value range of , represents the time span of the simulation task, is the coupling complexity coefficient, with a value range of ;

[0142] Based on the coupling factor sequence, the correction values ​​of clean room indoor temperature and humidity and air pollutant concentration are calculated respectively. The calculation formula is as follows:

[0143] ;

[0144] ;

[0145] ;

[0146] ;

[0147] in, for Corrected value of clean room temperature at all times, for The predicted value of clean room temperature at all times, is the heat conversion coefficient, the value range is , The heat generated by the excessive power, Exceeding the power limit, is the power exceeding time step, for Corrected humidity value of clean room at all times, for The predicted value of indoor humidity in the clean room at all times, The influence coefficient of excessive power on humidity, the value range is , for Corrected value of air pollutant concentration at any moment, for The predicted value of air pollutant concentration at each moment, Is the influence coefficient of air filter efficiency reduction, the value range is ;

[0148] The correction values ​​of clean room indoor temperature, humidity and air pollutant concentration are integrated according to the timestamp to obtain the correction temperature sequence, correction humidity sequence and correction air pollutant concentration sequence;

[0149] Based on the temperature sensor layout, the three-dimensional linear interpolation method is used to estimate the temperature distribution of the entire clean room space and obtain the temperature distribution map. The calculation formula is as follows:

[0150] ;

[0151] in, For coordinates Department Corrected value of clean room indoor temperature at the moment, For the Temperature sensor locations Temperature correction value at the moment, For the Temperature sensor locations Temperature correction value at the moment, is the index variable of the temperature sensor serial number, For the The index variable of the temperature sensor number closest to the temperature sensor. Indicates the A temperature sensor and The distance between the temperature sensors, Indicates the Temperature sensors and coordinates distance, 、 and They represent the coordinate values ​​in the three-dimensional model of the clean room;

[0152] Based on the layout of humidity sensors, the humidity distribution of the entire clean room space is estimated using the three-dimensional linear interpolation method to obtain a humidity distribution map.

[0153] The convection diffusion algorithm is used to calculate the indoor air pollutant concentration distribution in the clean room based on the modified air pollutant concentration sequence and the layout of the air filter outlet. The air pollutant concentration distribution map is obtained. The calculation formula is as follows:

[0154] ;

[0155] in, For coordinates Department Corrected value of air pollutant concentration at any moment, 、 and is the three-dimensional coordinate value of the air filter outlet, is the diffusion length;

[0156] The relative error method is used to verify the accuracy of temperature distribution maps, humidity distribution maps, and air pollutant concentration distribution maps;

[0157] Specifically, the ratio of the absolute deviation between the temperature observation value and the temperature correction value to the total temperature observation value is calculated to verify the accuracy of the temperature distribution map. The calculation formula is as follows:

[0158] ;

[0159] in, is the temperature distribution map accuracy, is the total number of temperature sensors, For the Temperature sensor locations Temperature observation value at time;

[0160] Humidity distribution map accuracy The calculation is the same as ;

[0161] The ratio of the absolute deviation between the observed air pollutant concentration and the corrected air pollutant concentration to the observed air pollutant concentration is calculated to verify the accuracy of the air pollutant concentration distribution map. The calculation formula is as follows:

[0162] ;

[0163] in, is the accuracy of the air pollutant concentration distribution map, for Observed values ​​of air pollutant concentrations at each moment;

[0164] If the accuracy of any type of distribution map is lower than 0.99, increase the number of quantum Monte Carlo algorithm iterations and rerun the correction and distribution map calculation of the clean room indoor temperature and humidity and air pollutant concentration prediction values ​​until the accuracy of all three distribution maps reaches 0.99 or above;

[0165] S4. Use 3D model building software to build a 3D model of the clean room, and integrate the temperature and humidity distribution map and the air pollutant concentration distribution map into the 3D model of the clean room to obtain a digital twin simulation model of the clean room.

[0166] Based on the cleanroom interior space type (e.g., cuboid, cylinder, etc.) and size, as well as the layout of various sensors and equipment, a 3D model of the cleanroom is constructed using 3D modeling software (e.g., CAD).

[0167] Based on the temperature distribution map, humidity distribution map and air pollutant concentration distribution map, the corrected temperature sequence, humidity sequence and air pollutant concentration sequence are stored at each coordinate point in the clean room three-dimensional model and represented by a NumPy three-dimensional array;

[0168] Record the number of qubits, Hadamard gate ratio, and rotation gate angle (phase angle) and store them in JSON format as a simulation parameter set;

[0169] Create a metadata dataset containing spatial information and simulation conditions to describe the cleanroom 3D model and data source. Spatial information includes cleanroom dimensions, sensor locations, and air outlet locations. Simulation conditions include time points and accuracy indicators.

[0170] The metadata dataset, NumPy three-dimensional array, and simulation parameter set are bundled and stored as Python dictionary objects. They are serialized into a single file using the HDF5 format and output as the cleanroom digital twin simulation model.

[0171] To evaluate the advantages of quantum computing platforms in improving simulation realism and real-time performance, we conducted comparative experiments in multiple scenarios. As shown in Table 5 (for some representative scenarios), quantum computing outperforms classical computing frameworks (such as finite element analysis) in both error rate and time consumption, confirming that quantum computing can efficiently handle high-dimensional nonlinear problems and improve simulation real-time performance.

[0172] Table 5 Comparison data of clean room simulation authenticity and real-time performance

[0173]

[0174] Among them, for the calculation of the error rate, it is expressed as the ratio of the difference between the predicted value and the actual value to the actual value; for the error rate reduction rate, it is expressed as the ratio of the difference between the quantum computing error rate and the classical computing error rate to the classical computing error rate; for the time reduction rate, it is expressed as the ratio of the difference between the quantum computing time and the classical computing time to the classical computing time.

[0175] S5. Load the cleanroom digital twin simulation model to the cloud analysis platform, use the gradient descent method to optimize the cleanroom operating status, and generate an optimized operation and maintenance plan.

[0176] Read the cleanroom digital twin simulation model from local storage on the edge computing node, connect it to the gateway of the edge computing node via a standard Ethernet cable, and upload it to the cloud analysis platform (such as AWS EC2) using a secure file transfer protocol (such as SFTP);

[0177] Based on the cleanroom's requirements for indoor temperature, humidity, and air pollutant concentration, set cleanliness targets (including target values ​​for temperature and humidity and target values ​​for air pollutant concentration);

[0178] Define the cleanliness loss function:

[0179] ;

[0180] in, is the cleanliness loss function value, is the average temperature of all coordinate points in the temperature distribution diagram, is the temperature target value, is the average humidity of all coordinate points in the humidity distribution diagram, is the humidity target value, is the average air pollutant concentration of all coordinate points in the air pollutant concentration distribution map, is the target value of air pollutant concentration, 、 and are the weights of temperature, humidity and air pollutant concentration respectively, and their value ranges are , and satisfies ;

[0181] Adopt gradient descent method to optimize cleanliness target;

[0182] Calculate the temperature target value as follows:

[0183] ;

[0184] in, The influence coefficient of temperature regulating equipment on temperature, the value range is , Real-time power for temperature regulation equipment, Initial power of the temperature regulating device;

[0185] Calculate the humidity target value as follows:

[0186] ;

[0187] in, The influence coefficient of humidity control equipment on temperature, the value range is , Real-time power of humidity control equipment, is the initial power of the humidity control equipment;

[0188] Calculate the target value of air pollutant concentration using the following formula:

[0189] ;

[0190] in, is the filtration efficiency coefficient, the value range is , Real-time power of the air filter, is the initial power of the air filter, The real-time air flow rate at the air filter outlet. is the initial air flow rate at the air filter outlet;

[0191] Based on the cleanliness loss function and gradient formula, the real-time power of the temperature and humidity control equipment and air filter is iteratively updated. The calculation formula is as follows:

[0192] ;

[0193] ;

[0194] ;

[0195] in, is the gradient descent learning rate, Real-time power of the temperature regulating device after the update, The real-time power of the updated humidity control equipment, For updated air filter real-time power, 、 and Respectively express 、 and Find partial derivatives;

[0196] When the maximum number of iterations is reached, the update is stopped and the optimized operation and maintenance plan is obtained, that is, the optimal power of the temperature and humidity control equipment and the air filter.

[0197] S6. The optimized operation and maintenance plan is sent to the clean room equipment for execution. After the equipment is executed, the real-time status is fed back to the clean room digital twin simulation model.

[0198] The optimized operation and maintenance plan generated by the cloud analysis platform is transmitted back to the clean room equipment. After confirmation by the user, the power of the clean room equipment is automatically adjusted according to the optimized operation and maintenance plan.

[0199] A lightweight web server runs on the edge computing node to generate a software interface that displays the cleanroom digital twin simulation model in real time. The interface can also be accessed through a terminal device (such as a tablet) connected to the edge computing node to customize the power of the cleanroom equipment.

[0200] Based on the optimized operation and maintenance plan sent to the clean room equipment for execution, the clean room status data and clean room equipment power data are fed back in real time through the edge computing node, and the clean room digital twin simulation model is updated on the cloud analysis platform to complete the closed-loop process of clean room intelligent operation and maintenance.

[0201] This embodiment also provides a clean room intelligent operation and maintenance system based on digital twins, including: a data acquisition module, a state prediction module, a distribution map module, a simulation module, an analysis and optimization module, and a solution execution module. The data acquisition module is used to obtain clean room state data by arranging sensors, and at the same time obtain clean room equipment power data through external data sources, and pre-process it at the edge node to generate a multi-source time series data set; the state prediction module is used to construct a multi-scale prediction model by integrating long short-term memory networks, autoregressive integral moving average models and graph neural networks, and generate a multi-scale sequence data set based on the multi-source time series data set; the distribution map module is used to encode the multi-scale sequence data set on the quantum computing platform, configure quantum circuits, and calculate power and power through quantum Monte Carlo algorithms. The efficiency coupling factor uses the three-dimensional linear interpolation method and the convection-diffusion algorithm to estimate the temperature, humidity and air pollutant concentration distribution in the clean room indoor space, obtain the temperature and humidity distribution map and the air pollutant concentration distribution map, and verify the accuracy; the simulation module is used to use the three-dimensional model construction software to build a three-dimensional model of the clean room, and integrate the temperature and humidity distribution map and the air pollutant concentration distribution map into the three-dimensional model of the clean room to obtain a clean room digital twin simulation model; the analysis and optimization module is used to load the clean room digital twin simulation model to the cloud analysis platform, use the gradient descent method to optimize the clean room operating status, and generate an optimized operation and maintenance plan; the plan execution module is used to send the optimized operation and maintenance plan to the clean room equipment for execution, and after the equipment is executed, the real-time status is fed back to the clean room digital twin simulation model.

[0202] This embodiment also provides a computer device, which is suitable for the clean room intelligent operation and maintenance method based on digital twins, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the clean room intelligent operation and maintenance method based on digital twins proposed in the above embodiment.

[0203] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0204] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the clean room intelligent operation and maintenance method based on digital twins as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.

[0205] In summary, the present invention constructs a multi-scale prediction model by integrating long short-term memory networks, autoregressive integral moving average models and graph neural networks, effectively capturing various dynamic characteristics from instantaneous fluctuations to long-term trends, meeting different operation and maintenance needs, and simulating the physical field coupling in the clean room through a quantum computing platform, which greatly improves the authenticity and real-time performance of the simulation model. The optimized operation and maintenance plan is sent to the clean room equipment for execution, and the real-time status is fed back by the equipment after execution, forming a complete closed-loop control system to ensure continuous self-adjustment and optimization in actual operation.

[0206] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A clean room intelligent operation and maintenance method based on digital twins, characterized by: include, Acquire cleanroom status data and cleanroom equipment power data through external data sources, and pre-process them at the edge node to generate a multi-source time series dataset; By integrating long short-term memory networks, autoregressive integrated moving average models, and graph neural networks, a multi-scale prediction model is constructed to generate multi-scale sequence datasets based on multi-source time series datasets. On the quantum computing platform, we encode the multi-scale sequence data set, configure the quantum circuit, and calculate the power and efficiency coupling factor using the quantum Monte Carlo algorithm. The specific steps are as follows: The probability sequences of cleanroom equipment power exceeding the standard and air filter efficiency decreasing in the multi-scale sequence dataset are mapped to quantum states respectively, and the quantum encoding of the probability of cleanroom equipment power exceeding the standard and the probability of air filter efficiency decreasing are obtained. Based on the quantum coding of the probability of clean room equipment power exceeding the standard and the probability of air filter efficiency decreasing, the quantum Monte Carlo algorithm is used to calculate the power and efficiency coupling factor; According to the timestamp, the power and efficiency coupling factors at all times are integrated to obtain the coupling factor sequence; Using the three-dimensional linear interpolation method and convection diffusion algorithm, the temperature, humidity and air pollutant concentration distribution of the clean room indoor space are estimated, and the temperature and humidity distribution map and air pollutant concentration distribution map are obtained, and the accuracy is verified. The specific steps are as follows: Based on the coupling factor sequence, the correction values ​​of clean room indoor temperature and humidity and air pollutant concentration are calculated respectively to obtain the corrected temperature sequence, corrected humidity sequence and corrected air pollutant concentration sequence; The temperature and humidity distribution and air pollutant concentration distribution of the entire clean room space are estimated using the three-dimensional linear interpolation method; The concentration distribution of air pollutants in the clean room is calculated by the convection diffusion algorithm to obtain the distribution of air pollutant concentrations; Perform accuracy verification on the temperature and humidity distribution map and the air pollutant concentration distribution map. If the accuracy verification fails, increase the number of quantum Monte Carlo algorithm iterations until the accuracy verification passes. A 3D model of the cleanroom is constructed using 3D modeling software. The temperature and humidity distribution maps and air pollutant concentration distribution maps are then integrated into the 3D model to obtain a digital twin simulation model of the cleanroom. The cleanroom digital twin simulation model is loaded onto the cloud analysis platform, and the gradient descent method is used to optimize the cleanroom operating status and generate an optimized operation and maintenance plan. The optimized operation and maintenance plan is sent to the clean room equipment for execution, and after execution, the equipment will feedback the real-time status to the clean room digital twin simulation model.

2. The clean room intelligent operation and maintenance method based on digital twin according to claim 1, characterized in that: The specific steps of generating a multi-scale sequence dataset are as follows: The multi-scale prediction model includes a data allocation layer, a short-term state prediction layer, a medium-term state prediction layer and a long-term state prediction layer; Through the data interface, multi-source time series data sets are transmitted to the data distribution layer in chronological order second by second for data diversion and distribution; The short-term, medium-term and long-term states of the clean room are predicted respectively at the short-term state prediction layer, the medium-term state prediction layer and the long-term state prediction layer to obtain the short-term, medium-term and long-term state prediction sequences of the clean room; A structured aggregation method is used to classify and store short-term state prediction sequences, medium-term state prediction sequences, and long-term state prediction sequences according to time scale and target variable, and a multi-scale sequence dataset is output.

3. The clean room intelligent operation and maintenance method based on digital twin according to claim 1, characterized in that: The specific steps of integrating the temperature and humidity distribution map and the air pollutant concentration distribution map into the three-dimensional model of the clean room are as follows: Based on the type and size of the cleanroom space, as well as the layout of various sensors and equipment, a 3D model of the cleanroom is constructed using 3D model building software. The corrected temperature sequence, corrected humidity sequence, and corrected air pollutant concentration sequence are stored at each coordinate point of the clean room three-dimensional model and represented by a NumPy three-dimensional array; Create a metadata set containing spatial information and simulation conditions; The metadata dataset, NumPy three-dimensional array, and simulation parameter set are bundled and stored as Python dictionary objects. They are serialized into a single file using the HDF5 format and output as the cleanroom digital twin simulation model.

4. The clean room intelligent operation and maintenance method based on digital twin according to claim 1, characterized in that: The specific steps for generating an optimized operation and maintenance plan are as follows: Set cleanliness targets based on the cleanroom's requirements for indoor temperature, humidity, and air pollutant concentrations; Define the cleanliness loss function; The gradient descent method is used to optimize the cleanliness target and obtain the optimal power of the temperature and humidity control equipment and air filter.

5. The clean room intelligent operation and maintenance method based on digital twin according to claim 1, characterized in that: The preprocessing of clean room status data and clean room equipment power data includes denoising and normalization.

6. A cleanroom intelligent operation and maintenance system based on digital twins, based on the cleanroom intelligent operation and maintenance method based on digital twins according to any one of claims 1 to 5, characterized in that: Including data acquisition module, state prediction module, distribution map module, simulation module, analysis and optimization module and solution execution module, The data acquisition module is used to acquire clean room status data by arranging sensors, and simultaneously acquire clean room equipment power data through external data sources, and perform preprocessing at the edge node to generate a multi-source time series data set; The state prediction module is used to construct a multi-scale prediction model by integrating a long short-term memory network, an autoregressive integrated moving average model and a graph neural network, and generate a multi-scale sequence dataset based on a multi-source time series dataset; The distribution map module is used to encode multi-scale sequence data sets on the quantum computing platform, configure quantum circuits, calculate the power and efficiency coupling factor using the quantum Monte Carlo algorithm, and use the three-dimensional linear interpolation method and convection-diffusion algorithm to estimate the temperature, humidity and air pollutant concentration distribution in the clean room indoor space, obtain temperature and humidity distribution maps and air pollutant concentration distribution maps, and verify their accuracy; The simulation module is used to construct a three-dimensional model of the clean room using three-dimensional model construction software, and integrate the temperature and humidity distribution map and the air pollutant concentration distribution map into the three-dimensional model of the clean room to obtain a digital twin simulation model of the clean room; The analysis and optimization module is used to load the clean room digital twin simulation model into the cloud analysis platform, optimize the clean room operating status using the gradient descent method, and generate an optimized operation and maintenance plan; The solution execution module is used to send the optimized operation and maintenance solution to the clean room equipment for execution, and after the equipment is executed, it feeds back the real-time status to the clean room digital twin simulation model.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the clean room intelligent operation and maintenance method based on digital twins are implemented in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the clean room intelligent operation and maintenance method based on digital twins according to any one of claims 1 to 5 are implemented.

Images