Simulation optimization method and system for virtual grid nitrogen storage and oxygen control system in combination with digital twinning

Abstract

The invention provides a virtual grid nitrogen storage and oxygen control system simulation optimization method and system combined with digital twinning, and belongs to the technical field of nitrogen storage and oxygen control system.The method comprises the steps that firstly, cross-domain dynamic coupling association of a physical nitrogen storage and oxygen control system and a virtual grid nitrogen storage and oxygen control model is built, related data is collected, and a coupling simulation control instruction is generated; then driving the virtual model to execute grid node co-simulation operation, and generating a multi-dimensional simulation data set; then, layered deviation analysis is conducted on the multi-dimensional simulation data set and a physical system actual operation data set, and core grid parameters influencing oxygen control stability are positioned; designing a parameter adaptive optimization rule system according to core grid parameters, and adjusting virtual model grid node configuration parameters to generate an optimization parameter set; and finally, the optimization parameter set is loaded to the virtual model for closed-loop simulation verification, a simulation optimization scheme meeting the oxygen control precision and energy consumption cooperation requirement is generated, a physical system execution control instruction is generated according to the simulation optimization scheme, the physical system execution control instruction is transmitted to an execution assembly, and comprehensive optimization of the nitrogen storage and oxygen control system is achieved.

Description

Technical Field

[0001] This invention relates to the field of nitrogen storage and oxygen control system optimization technology, and more specifically, to a simulation optimization method and system for a virtual grid nitrogen storage and oxygen control system that combines digital twins. Background Technology

[0002] In numerous fields such as industrial production and energy storage, nitrogen storage and oxygen control systems play a crucial role in ensuring production safety, improving product quality, and optimizing energy utilization efficiency. Traditional nitrogen storage and oxygen control systems mainly rely on physical equipment and fixed control strategies to operate, resulting in relatively simple control methods and a lack of flexibility.

[0003] Existing nitrogen storage and oxygen control systems struggle to comprehensively and accurately acquire operational status information for all internal components during operation, resulting in poor adaptability to complex and changing working conditions. When the system's operating environment or conditions change, traditional systems cannot adjust control parameters in a timely and effective manner, leading to difficulties in ensuring oxygen control accuracy. This can easily result in oxygen levels exceeding or falling below the standard, thereby affecting the safety of the production process and product quality.

[0004] Meanwhile, traditional systems also have shortcomings in energy consumption control. Due to the lack of precise understanding and dynamic optimization of the system's operating status, energy waste is often caused, increasing production costs. In addition, traditional systems also have limitations in fault diagnosis and predictive maintenance, making it difficult to detect potential problems in advance and take timely measures, which can easily lead to system failures and production interruptions. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a simulation optimization method for a virtual grid nitrogen storage and oxygen control system incorporating digital twins, the method comprising:

[0006] A cross-domain dynamic coupling relationship is constructed between the physical nitrogen storage and oxygen control system and the virtual grid nitrogen storage and oxygen control model. Multi-dimensional operation data of the physical nitrogen storage and oxygen control system and grid node feature data of the virtual grid nitrogen storage and oxygen control model are collected. Coupled simulation control commands are generated by aligning cross-domain data features.

[0007] Based on the coupled simulation control command, the virtual grid nitrogen storage and oxygen control model is driven to perform grid node collaborative simulation operation, and the status interaction data of grid nodes and the overall system operation data are collected in real time during the simulation process to generate a multi-dimensional simulation dataset.

[0008] A hierarchical deviation analysis was performed on the multi-dimensional simulation dataset and the actual operation dataset of the physical nitrogen storage and oxygen control system to construct a multi-dimensional deviation data matrix. Based on the deviation source tracing method, the core grid parameters affecting the stability of oxygen control were located.

[0009] Based on the influence characteristics of the core grid parameters, a parameter adaptive optimization rule system is designed. Based on the parameter adaptive optimization rule system, the grid node configuration parameters of the virtual grid nitrogen storage and oxygen control model are adjusted to generate an optimized parameter set. The parameter adaptive optimization rule system is constructed based on the multi-objective influence characteristics of the core grid parameters on oxygen control accuracy, nitrogen consumption, pressure stability and energy consumption level, and combined with the set multi-objective constraints. It includes an interactive influence compensation mechanism for handling the coupling influence between parameters, as well as rules for collaborative adjustment and comprehensive evaluation of the core grid parameters based on the parameter influence evaluation model.

[0010] The optimized parameter set is loaded into the virtual grid nitrogen storage and oxygen control model to perform closed-loop simulation verification, generating a simulation optimization scheme that meets the requirements of oxygen control accuracy and energy consumption coordination. Based on the simulation optimization scheme, the execution control command of the physical nitrogen storage and oxygen control system is generated and transmitted to the corresponding execution component.

[0011] Furthermore, embodiments of the present invention also provide a simulation and optimization system for a virtual grid nitrogen storage and oxygen control system combined with digital twins, characterized in that it includes:

[0012] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described simulation optimization method for a virtual grid nitrogen storage and oxygen control system incorporating digital twins by executing the machine-executable instructions.

[0013] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a virtual grid nitrogen storage and oxygen control system simulation optimization system incorporating digital twins reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the virtual grid nitrogen storage and oxygen control system simulation optimization system incorporating digital twins to execute the aforementioned virtual grid nitrogen storage and oxygen control system simulation optimization method incorporating digital twins.

[0014] Based on the above, by constructing a cross-domain dynamic coupling relationship between the physical nitrogen storage and oxygen control system and the virtual grid nitrogen storage and oxygen control model, data interaction and collaborative optimization between the physical system and the virtual model are realized. By collecting multi-dimensional operational data of the physical system and grid node feature data of the virtual model, and generating coupled simulation control commands through cross-domain data feature alignment, the system can comprehensively and accurately obtain system operation status information. Based on the coupled simulation control commands, the virtual model is driven to perform grid node collaborative simulation operation. Multi-dimensional data is collected in real time during the simulation process to generate a multi-dimensional simulation dataset. Layered deviation analysis is performed between the multi-dimensional simulation dataset and the actual operation dataset of the physical system to construct a multi-dimensional deviation data matrix. Based on the deviation source tracing method, the core grid parameters affecting oxygen control stability are located, and key problems in system operation can be identified. Based on the influence characteristics of the core grid parameters, a parameter adaptive optimization rule system is designed, and the grid node configuration parameters of the virtual model are adjusted to generate an optimized parameter set. The optimized parameter set is loaded into the virtual model to perform closed-loop simulation verification, generating a simulation optimization scheme that meets the requirements of oxygen control accuracy and energy consumption coordination. Based on the simulation optimization scheme, the execution control commands of the physical system are generated and transmitted to the corresponding execution components. Finally, a comprehensive, accurate, and dynamic optimization of the physical nitrogen storage oxygen control system is achieved, improving oxygen control accuracy, reducing energy consumption, and enhancing system stability and reliability. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the execution flow of the simulation optimization method for a virtual grid nitrogen storage and oxygen control system combined with digital twins provided in an embodiment of the present invention.

[0016] Figure 2 This is a schematic diagram of exemplary hardware and software components of the simulation and optimization system for a virtual grid nitrogen storage and oxygen control system combined with digital twins provided in an embodiment of the present invention. Detailed Implementation

[0017] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a simulation optimization method for a virtual grid nitrogen storage and oxygen control system combining digital twins, provided in one embodiment of the present invention. The following is a detailed description of this simulation optimization method for a virtual grid nitrogen storage and oxygen control system combining digital twins.

[0018] Step S110: Construct a cross-domain dynamic coupling relationship between the physical nitrogen storage and oxygen control system and the virtual grid nitrogen storage and oxygen control model. Collect multi-dimensional operation data of the physical nitrogen storage and oxygen control system and grid node feature data of the virtual grid nitrogen storage and oxygen control model. Generate coupled simulation control commands by aligning cross-domain data features.

[0019] This embodiment uses seven cold storage facilities in a regional cold chain logistics center project as an application scenario. In this scenario, each cold storage facility is an independent nitrogen storage and oxygen control protection zone. Digital twin technology is needed to achieve dynamic coupling between the virtual grid model and the physical system in order to optimize oxygen control accuracy and energy consumption. The implementation process of this step will be described in detail below.

[0020] Step S111: Divide the physical nitrogen storage and oxygen control system into functional subsystems. Each functional subsystem is equipped with multiple types of sensors to collect multi-dimensional operating data of the functional subsystem. The multi-dimensional operating data includes oxygen concentration distribution data, nitrogen flow dynamic data, system pressure gradient data, temperature field distribution data, and real-time energy consumption data.

[0021] First, according to the process flow of the physical nitrogen storage and oxygen control system, each cold storage in a certain region's cold chain logistics center is divided into a nitrogen generation subsystem, a nitrogen transportation subsystem, a nitrogen release subsystem, an oxygen concentration monitoring subsystem, a pressure regulation subsystem, a temperature control subsystem, and an energy consumption metering subsystem. Each functional subsystem is equipped with corresponding sensors based on its operating characteristics: the nitrogen generation subsystem is equipped with a nitrogen purity sensor, a raw material air flow sensor, a molecular sieve bed pressure sensor, and a nitrogen generator operating current sensor; the nitrogen delivery subsystem is equipped with electromagnetic flow meters and pressure transmitters on the main and branch pipelines; the nitrogen release subsystem is equipped with miniature flow regulating valves and valve status sensors at the release nozzles in each grid area; the oxygen concentration monitoring subsystem uses a combination of a suction sampling network and optical oxygen sensors, arranging sampling points in the cold storage at a spatial grid density of 10m × 10m × 5m; the pressure regulation subsystem is equipped with absolute pressure sensors at the top, middle, and bottom of the cold storage; the temperature control subsystem uses distributed fiber optic temperature sensors, with a monitoring point every 2 meters along the height of the cold storage; the energy consumption metering subsystem is equipped with smart meters for each electrical device to achieve separate metering. Data collected by the sensors is transmitted to the data acquisition gateway via industrial Ethernet, with the sampling frequency set to 1Hz to ensure real-time data transmission.

[0022] Step S112: Extract the grid node feature data of the virtual grid nitrogen storage and oxygen control model. The grid node feature data includes the virtual functional attributes, virtual spatial location parameters, virtual medium transmission coefficient, virtual pressure regulation threshold and virtual energy consumption calculation coefficient of each grid node.

[0023] The virtual grid-based nitrogen and oxygen control model employs a 1:1 3D modeling approach compared to the physical cold storage, dividing each cold storage unit into 3D grid nodes. The virtual functional attributes of each grid node are determined based on its corresponding area function in the physical system; for example, a node directly below the nitrogen release nozzle is labeled as the "nitrogen release core area," and a node in a corner of the cold storage is labeled as the "airflow stagnation area." Virtual spatial position parameters are described using a 3D Cartesian coordinate system, with the origin set at the lower left corner of the cold storage's bottom surface. The X-axis runs along the length of the cold storage, the Y-axis along the width, and the Z-axis along the height. The virtual medium transmission coefficient comprehensively considers the cargo density, airflow resistance, and nitrogen diffusion coefficient in the area where the grid node is located. The oxygen diffusion coefficient, nitrogen permeability coefficient, and turbulent diffusion coefficient of each node are pre-calculated through CFD simulation. Virtual pressure regulation thresholds are set according to the airtightness level and design pressure range of the cold storage, and are divided into upper and lower limits for normal operating pressure and emergency pressure reduction thresholds. The virtual energy consumption calculation coefficient includes the equipment power coefficient, operating time coefficient, and efficiency decay coefficient of the region corresponding to the node. The equipment power coefficient is determined with reference to the relationship curve between the rated power and load rate of the physical equipment.

[0024] Step S113: Perform data standardization processing on the multi-dimensional operating data and the grid node feature data to generate standardized physical data and standardized virtual data.

[0025] For multi-dimensional operational data, a range standardization method is used: oxygen concentration distribution data is converted from raw measurements to a percentage deviation relative to the target oxygen concentration, i.e., (measured value - target value) / target value; nitrogen flow dynamic data is normalized based on the maximum design flow rate of the subsystem; system pressure gradient data is converted to relative pressure, i.e., (measured absolute pressure - local atmospheric pressure) / design working pressure; temperature field distribution data is calculated as (measured temperature - design temperature) / (design temperature fluctuation range); real-time energy consumption data is standardized based on the rated power of the equipment, according to (real-time power / rated power). For grid node feature data, virtual spatial location parameters are normalized to (X / Lx, Y / Ly, Z / Lz), where Lx, Ly, and Lz are the length, width, and height of the cold storage, respectively; the virtual medium transmission coefficient is normalized using a logarithmic standardization method, converting the original coefficients into dimensionless relative coefficients; the virtual pressure regulation threshold and virtual energy consumption calculation coefficients are both normalized based on the maximum value of their respective parameters. All standardization processes are implemented through the data preprocessing module. The processed datasets are stored in a distributed time-series database, using a columnar storage format to improve query efficiency.

[0026] Step S114: Construct a cross-domain data feature mapping model, perform feature association between the oxygen concentration distribution data in the standardized physical data and the virtual medium transmission coefficient in the standardized virtual data, and perform feature association between the nitrogen flow dynamic data in the standardized physical data and the virtual pressure regulation threshold in the standardized virtual data.

[0027] The cross-domain data feature mapping model employs a deep neural network architecture, comprising an input layer, a feature fusion layer, an association inference layer, and an output layer. The input layer receives standardized physical data and standardized virtual data. Oxygen concentration distribution data and nitrogen flow dynamic data constitute the physical feature vector, while the virtual medium transmission coefficient and virtual pressure regulation threshold constitute the virtual feature vector. The feature fusion layer uses an attention mechanism to weightedly fuse the physical and virtual feature vectors. The attention weights are learned through training; for example, for regions with large oxygen concentration gradients, the model automatically increases the weight of the corresponding virtual medium transmission coefficient. The association inference layer contains three fully connected sublayers, each followed by batch normalization and ReLU activation functions to learn the nonlinear mapping relationship between cross-domain features. The output layer outputs a feature association strength matrix, where matrix elements represent the degree of association between physical and virtual features. Model training uses the Adam optimizer, with the loss function being a weighted sum of mean squared error and cross-entropy loss. The training dataset consists of paired samples of historical operating data from the physical system and simulation data from the virtual model, with a sample size of no less than 100,000 sets.

[0028] Step S1141: Perform spatial interpolation on the oxygen concentration distribution data in the standardized physical data to obtain the specific value of oxygen concentration at each spatial point in the physical system, forming an oxygen concentration spatial distribution matrix.

[0029] Kriging interpolation was used to spatially interpolate discrete oxygen concentration sampling data. First, a variogram model was constructed based on the three-dimensional coordinates of the sampling points. The experimental variogram was fitted using the least squares method to determine the Nugget effect, sill value, and range parameters. Then, based on the spatial location of the grid nodes, ordinary Kriging interpolation was used to calculate the estimated oxygen concentration for each node, and the variance of the estimation error was calculated to evaluate the interpolation accuracy. For regions where the interpolation error variance exceeded a preset threshold, virtual sampling points were added for secondary interpolation. The final generated spatial distribution matrix of oxygen concentration has the same dimension as the number of nodes in the virtual grid model, and the matrix elements are the standardized oxygen concentration values ​​for the corresponding grid nodes. The interpolation process was accelerated using a GPU to ensure that the interpolation calculation for a single cold storage unit was completed within 10 seconds.

[0030] Step S1142: Perform grid node matching processing on the virtual medium transmission coefficients in the standardized virtual data, and associate the virtual medium transmission coefficients of each grid node with the corresponding spatial points to form a virtual medium transmission coefficient space matrix.

[0031] The virtual medium transport coefficients were pre-calculated using multiphysics coupling simulation. The specific process was as follows: Based on a three-dimensional geometric model of the cold storage, the cargo area was set as a porous medium, air as a continuous medium, and nitrogen as the component transport medium. Steady-state simulations were performed using Fluent software to solve the Navier-Stokes equations and the component transport equations, obtaining the oxygen diffusion coefficient, nitrogen permeability coefficient, and turbulent diffusion coefficient for each grid node. The simulation results were indexed according to the spatial coordinates of the grid nodes to generate a virtual medium transport coefficient spatial matrix. Each row of the matrix corresponds to a grid node, and each column corresponds to a transport coefficient. To ensure spatial matching with the physical data, the spatial coordinates of the virtual grid nodes were strictly consistent with the interpolation grid of the oxygen concentration sampling points, with coordinate deviation controlled within ±0.1 meters.

[0032] Step S1143: Construct a correlation function between oxygen concentration and medium transport coefficient, using the values ​​in the spatial distribution matrix of oxygen concentration as input and the values ​​in the spatial matrix of virtual medium transport coefficient as output, and determine the coefficients of the correlation function through linear regression.

[0033] First, paired samples were extracted from the spatial distribution matrix of oxygen concentration and the spatial matrix of virtual medium transport coefficients. Each sample contained the standardized oxygen concentration value of a grid node and the corresponding standardized values ​​of the three medium transport coefficients. Then, a multiple linear regression model was constructed, using the standardized oxygen concentration value as the dependent variable and the standardized values ​​of the three medium transport coefficients as independent variables. The regression coefficients were estimated using the least squares method. The regression model took the form: Oxygen concentration = Coefficient 1 × Oxygen diffusion coefficient + Coefficient 2 × Nitrogen permeability coefficient + Coefficient 3 × Turbulent diffusion coefficient + Constant term. To avoid multicollinearity, the variance inflation factor was tested on the independent variables before regression. Principal component analysis was used to reduce the dimensionality of independent variables with inflation factors greater than 10. The significance of the regression coefficients was verified by a t-test with a significance level set at 0.05. After removing insignificant independent variables, the model was re-estimated.

[0034] Step S1144: The correlation function is validated using cross-validation. The oxygen concentration spatial distribution matrix is ​​divided into a training set and a validation set. The correlation function is trained using the training set and the prediction accuracy of the correlation function is tested using the validation set.

[0035] A k-fold cross-validation method is used, randomly dividing the sample set into k subsets, with k-1 subsets serving as the training set and the remaining subset as the validation set. For each training set, the correlation function coefficients are estimated according to step S1143. Then, the root mean square error (RMSE) and coefficient of determination (R²) between the predicted and actual oxygen concentrations are calculated using the validation set data. Cross-validation is repeated k times, and the average RMSE and R² are used as the model's generalization performance metrics. When k=10, the model's validation accuracy is relatively stable; therefore, this embodiment selects 10-fold cross-validation. If the R² of the validation set is less than 0.8, the correlation function accuracy is deemed insufficient, and the model structure needs adjustment.

[0036] Step S1145: If the prediction accuracy does not reach the preset accuracy threshold, then adjust the type of the correlation function and reconstruct the correlation function using a nonlinear regression method until the prediction accuracy reaches the preset accuracy threshold.

[0037] When the predictive accuracy of the linear regression model is insufficient, a nonlinear regression model is used for optimization. First, multinomial regression is attempted, incorporating quadratic and interaction terms of the independent variables into the model, such as adding a squared term for the oxygen diffusion coefficient or a product term of the oxygen diffusion coefficient and nitrogen permeability coefficient. If multinomial regression still fails to meet the accuracy requirements, a support vector regression (SVR) model is used, with a radial basis function (RBF) kernel. The penalty coefficient C and kernel parameter γ are optimized using a grid search method. The SVR model is trained using 5-fold cross-validation, with the RMSE of the validation set as the optimization objective. If the accuracy of the SVR model is still insufficient, a gradient boosting tree (GBDT) model is further employed, integrating multiple decision trees to improve predictive performance. Hyperparameters such as the depth of the decision trees, the number of leaf nodes, and the learning rate are determined through Bayesian optimization. Throughout the model selection process, the R² of 10-fold cross-validation is consistently used as the accuracy benchmark.

[0038] Step S1146: Perform time series decomposition on the nitrogen flow dynamic data in the standardized physical data, extract the trend term, periodic term and random term of nitrogen flow, and form a nitrogen flow dynamic feature vector.

[0039] The STL (Season on Earth and Trend Decomposition using Loess) method was employed to decompose the dynamic nitrogen flow data into a time series. First, the decomposition window parameters were set: the trend term window length was determined based on the diurnal fluctuation cycle of nitrogen flow and was set to 24 hours; the periodic term window length was set to 1 hour, corresponding to a typical valve regulation cycle; and the residual term window length was set to 15 minutes to capture high-frequency noise. Then, the trend term, periodic term, and random term were separated using local weighted regression (Loess) smoothing. The trend term reflects the long-term trend of nitrogen flow, the periodic term reflects hourly periodic fluctuations, and the random term reflects unpredictable random disturbances. The three components obtained from the decomposition were standardized separately and then concatenated in chronological order to form a dynamic feature vector of nitrogen flow. The vector length is the number of sampling points within the decomposition window, and each time point corresponds to the standardized value of the three components.

[0040] Step S1147: Perform time response analysis on the virtual pressure regulation threshold in the standardized virtual data, calculate the response change of the virtual pressure regulation threshold under different nitrogen flow inputs, and form a pressure regulation response feature vector.

[0041] A dynamic response model of the pressure regulation subsystem is constructed in a virtual mesh model. This model includes a proportional-integral-derivative (PID) controller, an electric regulating valve, and a pipeline fluid dynamics module. The nitrogen flow input is set as a combination signal of the trend term, periodic term, and random term obtained from step S1146. By changing the amplitude and frequency of the input signal, the output response of the virtual pressure regulation threshold is recorded. Response analysis employs the step response method and the frequency response method: In the step response method, step inputs of 5%, 10%, and 15% of the rated flow are applied, and the rise time, overshoot, and steady-state error of the pressure regulation threshold are recorded; in the frequency response method, a sinusoidal sweep input is applied within the frequency range of 0.01Hz to 1Hz, and the amplitude-frequency and phase-frequency characteristics are recorded. The response characteristic parameters (rise time, overshoot, steady-state error, amplitude-frequency gain, and phase shift) are arranged sequentially to form a pressure regulation response characteristic vector, with the vector dimension consistent with the nitrogen flow dynamic characteristic vector.

[0042] Step S1148: Construct a correlation model between nitrogen flow rate and pressure regulation threshold, calculate the vector similarity between the dynamic feature vector of nitrogen flow rate and the feature vector of pressure regulation response, and determine the correlation weight between the two.

[0043] The cosine similarity algorithm is used to calculate the similarity between the dynamic feature vector of nitrogen flow rate and the feature vector of pressure regulation response. First, the two vectors are standardized to unit vectors. Then, the vector dot product is calculated to obtain the cosine similarity value; the closer the value is to 1, the higher the correlation. For feature vectors containing multiple time points, the cosine similarity for each corresponding time point is calculated, and the average value is taken as the overall correlation. Correlation weights are assigned based on the correlation magnitude; for example, a similarity of 0.9 is assigned a weight of 0.9, and a similarity of 0.6 is assigned a weight of 0.6. For correlations with time lags, the similarity at different lag times is calculated using a sliding time window. The lag time corresponding to the maximum similarity is selected as the optimal lag, and the response feature vector at that lag time is correlated with the flow rate feature vector.

[0044] Step S1149: Establish a vector mapping relationship based on the association weights, so that each element in the dynamic feature vector of nitrogen flow can correspond to a specific element in the pressure regulation response feature vector.

[0045] Based on the correlation weights and optimal lag time determined in step S1148, a vector mapping matrix is ​​constructed. The rows of the mapping matrix correspond to the elements of the nitrogen flow dynamic feature vector (the components of the trend, periodic, and random terms at each time point), and the columns correspond to the elements of the pressure regulation response feature vector (response parameters such as rise time and overshoot). The matrix elements are the correlation weights. Vector mapping is achieved through matrix multiplication: Pressure regulation response feature vector = Mapping matrix × Nitrogen flow dynamic feature vector. To verify the accuracy of the mapping relationship, the mean square error between the mapped response feature vector and the actual output of the virtual model is calculated. If the error exceeds a preset threshold, iterative optimization is performed by adjusting the element weights of the mapping matrix until the error meets the requirements.

[0046] Step S11410: By changing the dynamic data of nitrogen flow, detect whether the pressure regulation response feature vector changes as expected according to the mapping relationship. If the expected change is not achieved, adjust the correlation weights and re-establish the mapping relationship.

[0047] Typical operating conditions (such as normal operation, peak load, and valve failure) are selected in the physical system, and corresponding dynamic data of nitrogen flow are collected and input into the correlation model to calculate the expected pressure regulation response feature vector. Simultaneously, the same operating conditions are reproduced in the virtual model, and the actual pressure regulation response feature vector is recorded. The expected values ​​are compared with the actual values, and the relative error of each response parameter is calculated. If the relative error of a parameter exceeds 10%, the correlation weight of that parameter is deemed unreasonable, and the cosine similarity between that parameter and the nitrogen flow feature vector needs to be recalculated, adjusting the weights of the corresponding elements in the mapping matrix. This detection and adjustment process is repeated until the relative errors of all response parameters are less than 10%, ensuring the accuracy of the mapping relationship.

[0048] Step S115: Calculate the correlation matching degree of cross-domain data features. If the correlation matching degree does not reach the preset matching threshold, adjust the feature weight allocation of the cross-domain data feature mapping model and re-associate the features until the correlation matching degree reaches the preset matching threshold.

[0049] The correlation matching degree is evaluated using a comprehensive metric, including feature vector similarity, response time deviation, and error accumulation rate. Feature vector similarity is a weighted average of the similarity between oxygen concentration and medium transport coefficient correlations and the similarity between nitrogen flow rate and pressure regulation threshold correlations, with weights set to 0.6 and 0.4, respectively. Response time deviation is the absolute difference between the virtual model's response time and the physical system's response time, normalized and used as a negative indicator. The error accumulation rate is the ratio of the sum of squared errors from 100 consecutive sampling points to the signal energy. The correlation matching degree is obtained by weighted summing of the three indicators, with a maximum score of 100 points and a preset matching threshold of 85 points. If the calculated correlation matching degree is lower than 85 points, the weight parameters of each layer in the cross-domain data feature mapping model are adjusted using the backpropagation algorithm, with a focus on adjusting the weight allocation of the attention mechanism and the connection weights of the correlation inference layer. After adjustment, feature association and matching degree calculations are performed again, with an upper limit of 100 iterations to ensure that the matching threshold is reached within a limited number of iterations.

[0050] Step S116: Based on the cross-domain data feature association results that reach the preset matching threshold, construct the dynamic coupling association between the physical nitrogen storage and oxygen control system and the virtual grid nitrogen storage and oxygen control model to determine the collaborative parameters between the two in terms of data interaction, state feedback, and parameter adjustment.

[0051] Dynamic coupling is achieved through a real-time data synchronization mechanism and a two-way control channel. Regarding data interaction, the types and timing of data transmitted from the physical system to the virtual model are defined: oxygen concentration distribution data is transmitted every 5 seconds, nitrogen flow dynamic data every 1 second, system pressure gradient data every 2 seconds, temperature field distribution data every 10 seconds, and real-time energy consumption data every 30 seconds. The virtual model feeds back status data to the physical system, including predicted oxygen concentration, pressure regulation suggestions, and energy consumption optimization coefficients from grid nodes. The feedback cycle is consistent with the transmission cycle of the corresponding physical data. Status feedback employs an event-triggered mechanism; feedback is triggered immediately when the virtual model's predicted oxygen concentration deviation exceeds ±0.5% or the pressure fluctuation exceeds ±200 Pa. Parameter adjustment and coordination parameters include a data synchronization delay compensation coefficient, a virtual model prediction confidence weight, and a physical actuator response speed correction coefficient. These parameters are obtained offline through historical data identification; for example, the delay compensation coefficient is set based on the statistical average of network transmission delay, and the confidence weight is determined based on the reciprocal of the prediction error.

[0052] Step S117: Collect the multi-dimensional operation data change trends of the physical nitrogen storage and oxygen control system under different operating conditions and the grid node characteristic data change trends of the virtual grid nitrogen storage and oxygen control model, and analyze the consistency of their dynamic responses.

[0053] Five typical operating conditions of a cold chain logistics center in a certain region were selected for dynamic response consistency analysis: normal operating condition (24-hour continuous operation, average oxygen concentration 15.5%), peak load operating condition (goods entering the warehouse from 8-10 am, door opening frequency increased by 50%), low load operating condition (1-5 am, no goods entering or leaving the warehouse), sudden disturbance operating condition (simulated nitrogen release valve failure, flow rate drops sharply by 30%), and long-term operation condition (7 days of continuous operation). Under each operating condition, multi-dimensional operational data of the physical system and grid node characteristic data of the virtual model were collected simultaneously, covering the complete cycle of the operating condition. The dynamic response consistency analysis indicators include trend similarity, peak delay time, and steady-state deviation: trend similarity is determined by calculating the dynamic time warping (DTW) distance between the physical data and the virtual data, the smaller the distance, the more consistent the trend; peak delay time is the difference between the peak occurrence time of the virtual model and the peak occurrence time of the physical system; steady-state deviation is the average deviation between the physical data and the virtual data during the stable phase of the operating condition. Set acceptable ranges for each indicator, such as DTW distance less than 5, peak delay time less than 2 seconds, and steady-state deviation less than 2%.

[0054] Step S118: Adjust the collaborative parameters of the dynamic coupling association according to the dynamic response consistency results, so that the response delay of physical system data changes and virtual model data changes is controlled within a preset range.

[0055] If the dynamic response consistency analysis results show that a certain indicator exceeds the acceptable range, the corresponding coordination parameters are adjusted: when the trend similarity is insufficient, the weights of the feature fusion layer of the cross-domain data feature mapping model are adjusted, increasing the weights of high-similarity features; when the peak latency is too long, the data synchronization mechanism is optimized, edge computing nodes are used to reduce data transmission latency, and the prediction step size of the virtual model is adjusted, changing from one-step prediction to multi-step rolling prediction; when the steady-state deviation is too large, the grid node feature data of the virtual model is corrected, such as recalibrating the virtual medium transmission coefficient or the virtual pressure regulation threshold. After adjustment, the operating condition test is repeated until all indicators are within the acceptable range. At this point, the response latency between the physical system and the virtual model is controlled within 5 seconds, meeting the requirements of real-time simulation.

[0056] Step S119: Based on the adjusted dynamic coupling correlation and coordination parameters, generate coupled simulation control instructions that include data interaction frequency, state feedback cycle, and parameter adjustment trigger conditions.

[0057] The coupled simulation control commands adopt a structured message format, including a command header, a data interaction configuration section, a status feedback configuration section, a parameter adjustment configuration section, and a checksum. The command header includes the command number, transmission time, and target model identifier; the data interaction configuration section defines the transmission frequency of various data types (e.g., 5Hz for oxygen concentration data, 10Hz for nitrogen flow rate data), the data compression algorithm (using LZ77 compression), and the transmission protocol (Modbus TCP / IP); the status feedback configuration section sets the feedback period (consistent with the data interaction frequency), feedback content (e.g., predicted oxygen concentration, suggested adjustment amount), and feedback channel (independent redundant channel); the parameter adjustment configuration section defines the trigger conditions (e.g., oxygen concentration deviation >1% for 3 seconds, energy consumption exceeding the threshold by 5%), adjustment priority (safety-related parameters are highest, followed by energy consumption parameters), and adjustment step size limit (a single adjustment cannot exceed 10% of the current value); the checksum is generated using the CRC32 algorithm to ensure the integrity of command transmission. After the control commands are generated, they are transmitted to the control interface of the virtual mesh nitrogen storage and oxygen control model through an encrypted channel. The model parses the commands and configures the corresponding simulation parameters.

[0058] Step S120: Based on the coupled simulation control command, drive the virtual grid nitrogen storage and oxygen control model to perform grid node collaborative simulation operation, collect the status interaction data of grid nodes and the overall system operation data in real time during the simulation process, and generate a multi-dimensional simulation dataset.

[0059] In a virtual mesh model of a cold chain logistics center in a certain region, after the coupled simulation control command is triggered, the model will start the collaborative simulation of mesh nodes according to the parameters configured in the command. The implementation logic of this process is described in detail below.

[0060] Step S121: Analyze the data interaction frequency, state feedback cycle and parameter adjustment trigger conditions in the coupled simulation control command, and convert them into simulation operation parameters that can be recognized by the virtual grid nitrogen storage and oxygen control model.

[0061] The coupled simulation control instruction parsing module is implemented using a syntax analyzer. First, lexical analysis is performed on the instructions, decomposing the instruction string into lexical units such as keywords, identifiers, and constants. Then, syntax analysis is performed, constructing a syntax tree based on preset instruction syntax rules (using Backus-Noor normal form) to verify the correctness of the instruction format. Finally, semantic analysis is performed, converting the syntax tree into simulation parameters recognizable by the virtual model. Data interaction frequency is converted into the model's sampling time step, such as 5Hz corresponding to a 0.2-second step; the state feedback cycle is converted into the model's output cycle, consistent with the sampling time step; parameter adjustment trigger conditions are converted into event trigger rules for the model, stored in the rule engine, and represented using production rules (e.g., "IF oxygen concentration > 16% AND duration > 3s THEN trigger nitrogen flow adjustment"). The parsed simulation parameters are stored in the model's configuration file in JSON format for easy reading and modification.

[0062] Step S122: Initialize the grid node state of the virtual grid nitrogen storage and oxygen control model according to the simulation operation parameters, and assign initial virtual oxygen concentration, initial virtual nitrogen flow rate, initial virtual pressure and initial virtual energy consumption value to each grid node.

[0063] The initialization of grid node states employs a hierarchical initialization strategy: First, global initial values ​​are set based on the current operating state of the physical system, such as setting the initial virtual oxygen concentration to the current average oxygen concentration of the physical system and the initial virtual pressure to the current system pressure. Then, local corrections are made based on the virtual spatial location parameters of the grid nodes; for example, the initial virtual temperature of the top node in the cold storage is 1°C lower than that of the bottom node (considering the effect of cold air sinking). Finally, initial values ​​for special nodes are set based on virtual functional attributes; for example, the initial virtual nitrogen flow rate of the nitrogen release node is set to 30% of its rated flow rate, and the initial virtual oxygen concentration of the oxygen monitoring node is subject to a ±0.5% random perturbation to simulate measurement noise. Initialization data is written to the state variables of the grid nodes through the model's API interface. The initialization process is executed automatically upon model startup, with a time limit of less than 10 seconds.

[0064] Step S123: Start the grid node collaborative simulation of the virtual grid nitrogen storage and oxygen control model. During the simulation, collect the state interaction data of each grid node and its adjacent grid nodes in real time. The state interaction data includes medium transmission data, pressure transmission data, temperature exchange data and energy consumption sharing data.

[0065] The virtual grid-based nitrogen and oxygen storage and control model employs a distributed simulation architecture, with each grid node acting as an independent simulation unit, achieving collaborative operation through message passing. Upon simulation startup, the master node broadcasts a startup signal to all slave nodes. Upon receiving the signal, each slave node loads the corresponding simulation module (such as a media transfer module or pressure calculation module) based on its virtual functional attributes. State interaction between grid nodes is achieved through a neighbor list; each node exchanges data only with its six adjacent nodes (front, back, left, right, up, and down). Media transfer data is calculated by solving the diffusion equations between adjacent nodes, considering the effects of concentration gradients and transfer coefficients. Pressure transfer data is calculated based on the fluid continuity equation, considering pressure differences and flow resistance between nodes. Temperature exchange data is calculated using the heat conduction equation, considering temperature gradients and heat conduction coefficients between nodes. Energy sharing data is calculated based on the energy flow direction between nodes (e.g., energy transfer from high-energy-consuming nodes to low-energy-consuming nodes). The acquisition frequency of interactive data is consistent with the data exchange frequency in the simulation operation parameters. After acquisition, the data is stored in a distributed in-memory database, supporting real-time querying.

[0066] Step S124: Collect the overall system operation data of the virtual grid nitrogen storage and oxygen control model. The overall system operation data includes the system average oxygen concentration, the system total nitrogen consumption, the system pressure fluctuation amplitude, the system temperature distribution uniformity, and the system total energy consumption data.

[0067] The overall system operation data is obtained through aggregation calculations of grid node status data: the system average oxygen concentration is the weighted average of the virtual oxygen concentrations of all grid nodes, with the weight being the spatial volume of the node; the total nitrogen consumption is the sum of the virtual nitrogen flow rates of all nitrogen-releasing nodes, calculated by integration; the system pressure fluctuation amplitude is the ratio of the standard deviation to the mean of the pressure data; the uniformity of the system temperature distribution is represented by the coefficient of variation of the temperature field, i.e., the ratio of the standard deviation to the mean; the total system energy consumption data is the sum of the virtual energy consumption values ​​of all energy-related nodes. The aggregation calculations are performed periodically by the master node, with the calculation cycle consistent with the status feedback cycle. The calculation results are stored in the system performance index library for subsequent deviation analysis and optimization.

[0068] Step S125: According to the state feedback cycle in the coupled simulation control command, the collected state interaction data and the overall system operation data are time-marked to form time-series interaction data and time-series overall data.

[0069] The time-series stamping uses UTC timestamps, accurate to the millisecond, and is synchronized with the data timestamps of the physical system. For state interaction data, each interaction event records the sending node ID, receiving node ID, interaction data value, and timestamp; for overall system operation data, each time point records the calculated values ​​and timestamps of all performance indicators. The time-series data is stored in a time-series database, organized by time partitions, and supports querying and aggregation operations by time range. To ensure data consistency, a clock synchronization protocol (such as PTP) is used during the time-series stamping process to keep the clock deviation between the virtual model and the physical system within 1 millisecond.

[0070] Step S126: Perform data correlation analysis on the time-series interactive data, identify strong and weak correlation paths of state interaction between grid nodes, and record the interaction frequency and interaction amount of the strong correlation path.

[0071] In this embodiment, the time-series interactive data covers the media transmission volume, pressure transmission volume, temperature exchange volume, and energy consumption sharing data of all nodes in the virtual grid model of Cold Storage No. 1 in a cold chain logistics center of a certain region. Data correlation analysis is achieved through the following sub-steps:

[0072] Step S1261: Classify the medium transmission data, pressure transmission data, temperature exchange data and energy consumption sharing data in the time-series interactive data according to grid node pairs to form an interactive data subset for each grid node pair.

[0073] First, iterate through all time-series interaction data, extracting the source node ID, target node ID, and interaction data value (e.g., media transfer volume of 0.5 m³ / h) from each data record. Based on the combination (i, j) of the source node ID and target node ID, classify all interaction data involving that node pair into the same data subset. For example, the interaction data subset for node pair (101, 102) includes media transfer volume, pressure transfer volume, and other data at all times between them. For bidirectional interaction node pairs (e.g., node A transfers media to node B, while node B simultaneously feeds back pressure data to node A), classify them into two different interaction data subsets according to (A, B) and (B, A), respectively. After classification, each interaction data subset contains a timestamp sequence and a sequence of various interaction data values ​​at the corresponding time.

[0074] Step S1262: Perform statistical analysis on the subset of interaction data for each grid node pair, and calculate the total number of interactions, average interaction amount, maximum interaction amount, and interaction amount fluctuation coefficient during the simulation period.

[0075] For each node pair's interaction data subset, statistical analysis metrics include: total number of interactions (the total number of records with non-zero interactions in the subset); average interaction amount (the arithmetic mean of all interaction data values); maximum interaction amount (the maximum interaction data value in the subset); and interaction amount fluctuation coefficient (the ratio of the standard deviation of interaction data values ​​to the average interaction amount). For example, node pair (101, 102) has 1440 media transmission data records (sampling frequency 1Hz) within a 24-hour simulation period, of which 1200 are non-zero (total number of interactions), with an average transmission amount of 0.8 m³ / h, a maximum of 2.5 m³ / h, a standard deviation of 0.3 m³ / h, and a fluctuation coefficient of 0.3 / 0.8 = 0.375. All statistical metrics are stored categorized by node pair and interaction data type (media transmission, pressure transmission, etc.).

[0076] Step S1263: Set association strength evaluation indicators, which include interaction frequency score, interaction volume score and interaction stability score. The interaction frequency score, interaction volume score and interaction stability score are calculated according to the corresponding statistical data and preset scoring standards.

[0077] The correlation strength evaluation index uses a 100-point scale, with the specific scoring criteria as follows: Interaction frequency score is determined by the ratio of the total number of interactions to the total simulation duration (interaction frequency). The higher the ratio, the higher the score, with a maximum of 100 points (interaction frequency ≥ 50%) and a minimum of 0 points (interaction frequency < 1%). Linear interpolation is used for intermediate values. Interaction quantity score is determined by the ratio of the average interaction quantity to the maximum value of that type of interaction data. The higher the ratio, the higher the score, with a maximum of 100 points (ratio ≥ 80%) and a minimum of 0 points (ratio < 5%). Interaction stability score is calculated as (1 - interaction quantity fluctuation coefficient) × 100. The lower the fluctuation coefficient, the higher the score, with a maximum of 100 points (fluctuation coefficient = 0) and a minimum of 0 points (fluctuation coefficient ≥ 1). For example, a node pair with an interaction frequency of 30% receives a frequency score of 60 points; an average interaction quantity ratio of 50% receives a quantity score of 50 points; and a fluctuation coefficient of 0.2 receives a stability score of 80 points.

[0078] Step S1264: Calculate the comprehensive score of the association strength of each grid node pair according to the association strength evaluation index. The comprehensive score is obtained by adding the interaction frequency score, interaction amount score and interaction stability score according to the preset weight.

[0079] The preset weights are set according to the interaction characteristics: interaction frequency weight 0.4, interaction quantity weight 0.3, and interaction stability weight 0.3. The overall score = interaction frequency score × 0.4 + interaction quantity score × 0.3 + interaction stability score × 0.3. Taking the above node pair as an example, the overall score = 63 points. After calculation, each node pair's interaction path corresponds to an overall score, and different interaction types within the same node pair (such as media transmission and pressure transmission) are calculated separately for each overall score.

[0080] Step S1265: Grid node pairs whose overall association strength score exceeds the association strength threshold are identified as strongly associated node pairs, and the corresponding interaction paths are identified as strongly associated paths.

[0081] The association strength threshold is set according to system requirements; in this embodiment, it is set to 70 points. The comprehensive score of all node pairs is traversed. If the comprehensive score of the interaction path of a node pair is ≥70 points, it is determined to be a strongly associated path. For example, the comprehensive score of the media transmission path of node pair (205, 206) is 75 points, exceeding the threshold, and is marked as a strongly associated path; its pressure transmission path comprehensive score is 65 points, below the threshold, and is not included in the strongly associated path list. The identification information of strongly associated paths (source node ID, target node ID, interaction type) is stored in the strongly associated path list.

[0082] Step S1266: Grid node pairs whose overall association strength score does not exceed the association strength threshold are identified as weakly associated node pairs, and the corresponding interaction paths are identified as weakly associated paths.

[0083] For node pair interaction paths with a comprehensive score of <70, they are uniformly marked as weakly related paths. Weakly related paths only record the node pair identifier and interaction type, and are not subject to further detailed analysis.

[0084] Step S1267: Further analyze the strongly correlated paths, record the interaction frequency of each strongly correlated path in different time intervals, and calculate the deviation rate between the interaction frequency and the average interaction frequency in each time interval.

[0085] The simulation period is divided into multiple time intervals by hours (e.g., 24 hours divided into 24 intervals). For each strongly correlated path, the number of interactions within each time interval is counted, and the interaction frequency for that interval is calculated as (interval interaction count / total interval duration). Then, the deviation rate between the interaction frequency of that interval and the overall average interaction frequency of strongly correlated paths is calculated using the formula: (interval interaction frequency - average interaction frequency) / average interaction frequency × 100%. For example, if the overall average interaction frequency of a strongly correlated path is 50%, and the interaction frequency in the 3rd hour interval is 60%, then the deviation rate is 20%. A positive deviation rate indicates that the interaction frequency of that interval is higher than the average level, while a negative deviation rate indicates that it is lower than the average level.

[0086] Step S1268: Record the interaction volume of each strongly associated path in different time intervals, and calculate the deviation rate between the interaction volume in each time interval and the average interaction volume.

[0087] Similarly, the average interaction volume of each strongly correlated path is calculated within each time interval, and the deviation rate between the average interaction volume of the interval and the overall average interaction volume of the path is calculated using the formula: (Interval average interaction volume - Overall average interaction volume) / Overall average interaction volume × 100%. For example, if the overall average interaction volume of a certain path is 1.2 m³ / h, and the average interaction volume in the 8th hour interval is 1.08 m³ / h, the deviation rate is 10%.

[0088] Step S1269: Compile the identification information of strongly correlated paths, the interaction frequency and deviation rate of each time interval, and the interaction volume and deviation rate of each time interval into a strongly correlated path interaction report.

[0089] The strong correlation path interaction report includes basic path information (source node ID, target node ID, interaction type, and overall score) and time series analysis results (interaction frequency, frequency deviation rate, interaction volume, and volume deviation rate for each time interval). The report uses a structured data format and supports searching by path ID or time interval. For example, the report shows that the interaction frequency deviation rate of the strong correlation path (205, 206) is -30% and the interaction volume deviation rate is -25% between 2-3 AM, indicating a possible airflow stagnation problem during this period.

[0090] Step S12610: Briefly record the weakly related paths, including the weakly related node pair identifiers, the total number of interactions, and the average number of interactions, to form a summary table of weakly related paths.

[0091] The weakly correlated path summary table retains only key information, which is used to evaluate the importance of paths during subsequent grid node optimization, thereby reducing data storage.

[0092] Step S127: Perform trend analysis on the time-series overall data, extract the upward trend segment, downward trend segment and stable trend segment from the overall system operation data, and determine the simulation time interval corresponding to each trend segment.

[0093] Trend analysis employs a piecewise linear regression method. First, a minimum length for each trend segment is set (e.g., 10 sampling points). Then, a sliding window is used to linearly fit the overall time-series data, calculating the slope and correlation coefficient of the fitted line. A trend segment is identified as upward when the slope is greater than 0 and the correlation coefficient is greater than 0.8; a trend segment is identified as downward when the slope is less than 0 and the correlation coefficient is greater than 0.8; and a stationary trend segment is identified when the absolute value of the slope is less than 0.1 and the correlation coefficient is less than 0.5. The start and end points of each trend segment are determined by inflection point detection, where a significant change in the first derivative occurs (e.g., the absolute value of the derivative exceeds a threshold). The start and end times of each trend segment are recorded as simulation time intervals and associated with the corresponding trend type for subsequent condition division and optimization analysis.

[0094] Step S128: Integrate the time-series interactive data, time-series overall data, strongly correlated path information, and trend segment time interval information to construct a multi-dimensional simulation dataset according to the simulation time sequence.

[0095] The multi-dimensional simulation dataset adopts a data cube structure, with dimensions including time, node, metric, and path dimensions. The time dimension corresponds to the simulation time interval; the node dimension contains the identifiers of all grid nodes; the metric dimension contains various metrics of state interaction data and overall system operation data; and the path dimension contains the identifiers of strongly correlated paths. The metric of the data cube is the specific data value under the corresponding dimension combination. For example, the metric of "Time T1 - Node A - Oxygen Concentration - Path A → B" is the oxygen concentration data transmitted by node A through path A → B at time T1. The dataset is constructed using ETL tools, extracting data from time-series databases, path feature libraries, and other data sources. After cleaning, transformation, and loading, a multi-dimensional data model oriented towards analysis is finally formed, supporting multi-dimensional analysis operations such as slicing, dicing, and drill-down.

[0096] Step S130: Perform hierarchical deviation analysis on the multi-dimensional simulation dataset and the actual operation dataset of the physical nitrogen storage and oxygen control system, construct a multi-dimensional deviation data matrix, and locate the core grid parameters affecting the stability of oxygen control based on the deviation source tracing method.

[0097] This step aims to identify key parameters affecting oxygen control stability by comparing the deviations between simulation data and physical data. The following section uses simulation data and actual data from Cold Storage Unit 1 of a regional cold chain logistics center as an example to describe in detail the stratified deviation analysis and deviation tracing process.

[0098] Step S131: Collect the actual operating dataset of the physical nitrogen storage and oxygen control system under the same operating conditions. The actual operating dataset includes actual oxygen concentration distribution data, actual nitrogen flow rate data, actual pressure gradient data, actual temperature field data, and actual energy consumption data, and has a time stamp corresponding to the multi-dimensional simulation dataset.

[0099] The actual operational dataset was strictly synchronized with the multi-dimensional simulation dataset, using the same operating conditions (such as the same cargo storage volume, ambient temperature, and number of door openings) and time windows (such as continuous 24-hour data collection). Actual oxygen concentration distribution data was collected through a network of optical oxygen sensors deployed within the cold storage facility, with sensor calibration performed weekly to ensure measurement errors were less than ±0.3%. Actual nitrogen flow rate data was collected using electromagnetic flowmeters with an accuracy class of 0.5. Actual pressure gradient data was measured using absolute pressure sensors, covering a range from -1000 Pa to 1000 Pa. Actual temperature field data was collected using distributed fiber optic sensors with a spatial resolution of 1 meter. Actual energy consumption data was collected using smart meters with a metering accuracy class of 1.0. All actual data were stamped with the same UTC timestamps as the simulation data, with time synchronization errors controlled within 1 second to ensure data comparability.

[0100] Step S132: Divide the multi-dimensional simulation dataset and the actual running dataset into three levels according to data type: basic parameter layer, interaction relationship layer, and system performance layer. Each level contains a corresponding subset of simulation data and a subset of actual data.

[0101] The basic parameter layer contains parameters describing the system's basic state. The simulation data subset includes virtual oxygen concentration, virtual nitrogen flow rate, virtual pressure, and virtual temperature for each grid node; the actual data subset includes actual oxygen concentration, actual nitrogen flow rate, actual pressure, and actual temperature collected by sensors. The interaction layer contains parameters describing the system's internal interaction behavior. The simulation data subset includes media transfer, pressure transfer, and temperature exchange between grid nodes; the actual data subset is calculated from indirect measurement data of the physical system, such as calculating media transfer based on the pressure difference and flow rate at both ends of a pipe. The system performance layer contains parameters describing the overall system performance. The simulation data subset includes the system's average oxygen concentration, total nitrogen consumption, pressure fluctuation amplitude, temperature distribution uniformity, and total energy consumption; the actual data subset represents the overall performance indicators of the physical system, such as calculating the actual average oxygen concentration from the average value of all oxygen sensors and calculating the actual total energy consumption from the total meter reading.

[0102] Step S133: In the basic parameter layer, calculate the absolute deviation between the oxygen concentration data of the grid nodes in the simulation data subset and the oxygen concentration data of the physical region in the actual data subset, and calculate the relative deviation between the nitrogen flow rate data of the grid nodes in the simulation data subset and the nitrogen flow rate data of the physical system in the actual data subset.

[0103] The basic parameter layer deviation calculation adopts a point-to-point comparison method: for oxygen concentration, the virtual oxygen concentration of each grid node is compared with the actual oxygen concentration at the corresponding physical location of that node, and the absolute deviation (simulation value - actual value) is calculated; for nitrogen flow rate, the virtual nitrogen flow rate is compared with the actual nitrogen flow rate, and the relative deviation ((simulation value - actual value) / actual value) × 100% is calculated. To improve the comparison accuracy, a spatial interpolation method is used to interpolate the discrete sensor data of the physical system to the positions of the virtual grid nodes. The interpolation method is the same as in step S1141. The deviation calculation results are stored according to the grid node number, forming a basic parameter deviation matrix, where the matrix elements are the deviation values ​​of the corresponding nodes.

[0104] Step S134: In the interaction layer, calculate the deviation of medium transmission between grid nodes in the simulation data subset and the deviation of medium transmission between physical systems in the actual data subset, and calculate the deviation of pressure transmission between grid nodes in the simulation data subset and the deviation of pressure transmission between physical systems in the actual data subset.

[0105] The deviation calculation for the interaction relationship layer employs a path comparison method: For media transmission volume, a key transmission path in the physical system (such as the transmission path from the main pipeline to a branch pipeline) is selected, and the absolute deviation between the simulated and actual media transmission volume along this path is calculated; for pressure transmission deviation, the relative deviation between the simulated and actual pressure drop along the same path is calculated. The actual value of the media transmission volume is calculated using the flow difference between the two ends of the pipeline, and the actual value of the pressure drop is obtained using the difference between the measurements of the pressure sensors at both ends of the pipeline. The deviation calculation results are stored according to the transmission path number, forming an interaction relationship deviation matrix.

[0106] Step S135: At the system performance layer, calculate the deviation of the total system energy consumption in the simulation data subset and the deviation of the actual total system energy consumption in the actual data subset, and calculate the deviation of the system oxygen control accuracy in the simulation data subset and the deviation of the system oxygen control accuracy in the actual data subset.

[0107] The system performance layer deviation calculation adopts an overall index comparison method: the system total energy consumption deviation is the relative deviation between the simulated total energy consumption and the actual total energy consumption; the system oxygen control accuracy deviation is defined as the difference between the standard deviation of the simulated average oxygen concentration and the standard deviation of the actual average oxygen concentration, reflecting the deviation of oxygen control stability. In addition, the system pressure fluctuation amplitude deviation and temperature distribution uniformity deviation are also calculated using similar methods. The system performance layer deviation results form a system performance deviation vector, containing the deviation values ​​of each performance index.

[0108] Step S136: Based on the deviation calculation results of the three levels, construct a multi-dimensional deviation data matrix. The rows of the multi-dimensional deviation data matrix represent different data types, the columns of the multi-dimensional deviation data matrix represent different time nodes, and the elements of the multi-dimensional deviation data matrix represent the deviation values ​​of the corresponding data types at the corresponding time nodes.

[0109] The construction process of the multi-dimensional deviation data matrix is ​​as follows: The row dimension includes oxygen concentration, nitrogen flow rate, pressure, and temperature in the basic parameter layer; medium transmission volume, pressure transfer volume, and temperature exchange volume in the interaction relationship layer; and total energy consumption, oxygen control accuracy, pressure fluctuation amplitude, and temperature distribution uniformity in the system performance layer, totaling 11 data types; the column dimension is the time node, sampled at 1-minute intervals, covering the entire operating cycle; the matrix elements are the deviation values ​​of the corresponding data type at the corresponding time node (absolute deviations in the basic parameter layer, and relative deviations in the interaction relationship layer and system performance layer). The matrix is ​​stored using a two-dimensional array, with floating-point data types, and missing values ​​are filled using linear interpolation.

[0110] Step S137: Perform deviation intensity analysis on the multi-dimensional deviation data matrix, calculate the mean deviation, maximum deviation, and deviation fluctuation frequency for each data type, and determine the data types whose deviation intensity exceeds the preset deviation intensity threshold.

[0111] Deviation intensity analysis indicators include: the mean deviation, which is the arithmetic mean of the deviation values ​​of a certain data type at all time points; the maximum deviation, which is the maximum deviation value of that data type; and the deviation fluctuation frequency, which is the ratio of the number of times the deviation value exceeds the preset allowable range (e.g., oxygen concentration deviation ±0.5%) to the total number of time points. Preset deviation intensity thresholds are set according to system requirements, for example, a mean deviation threshold of ±0.3%, a maximum deviation threshold of ±1%, and a deviation fluctuation frequency threshold of 5%. For each data type, if any one of its indicators exceeds the threshold, the deviation intensity of that data type is determined to be excessive, and deviation tracing is required.

[0112] Step S138: For data types where the deviation intensity exceeds the preset deviation intensity threshold, analyze the correlation between the virtual mesh node characteristic parameters and the physical system operating parameters corresponding to the data type, based on the deviation tracing method.

[0113] Taking the oxygen concentration deviation data type (deviation intensity exceeding the standard) as an example, the deviation tracing process is implemented through the following sub-steps:

[0114] For example, step S1381: Determine the virtual grid node feature parameter category corresponding to the data type whose deviation intensity exceeds the preset deviation intensity threshold. For example, the oxygen concentration deviation corresponds to the virtual medium transmission coefficient, and the nitrogen flow deviation corresponds to the virtual pressure regulation threshold.

[0115] According to the system's association rules, different data type deviations correspond to specific categories of virtual grid node characteristic parameters: oxygen concentration distribution data deviations are mainly associated with the virtual medium transmission coefficient (affecting the oxygen diffusion rate); nitrogen flow dynamic data deviations are associated with the virtual pressure regulation threshold (affecting valve opening control); system pressure gradient data deviations are associated with the virtual pressure regulation threshold and virtual spatial location parameters; temperature field distribution data deviations are associated with the virtual medium transmission coefficient and virtual energy consumption calculation coefficient; and real-time energy consumption data deviations are associated with the virtual energy consumption calculation coefficient. In this embodiment, the oxygen concentration deviation corresponds to the virtual medium transmission coefficient category, and this type of parameter requires focused analysis.

[0116] Step S1382: Extract the value sequence of all virtual grid node feature parameters under the virtual grid node feature parameter category, and at the same time extract the corresponding data value sequence in the physical system operation parameters.

[0117] Extract the virtual medium transport coefficient value sequence (the sequence of coefficient values ​​for each node changing over time within the simulation period) from the virtual mesh model; extract the corresponding oxygen concentration distribution data value sequence (the sequence of physical sensor measurements matching the spatial location of the virtual node changing over time) from the physical system database. For example, the virtual medium transport coefficient sequence for a virtual mesh node (x=10m, y=5m, z=3m) is [0.9, 0.92, 0.88, ...], and the oxygen concentration sequence for its corresponding physical sensor (with the same installation location) is [15.2%, 15.3%, 15.1%, ...]. The time length of the value sequence is consistent with the operating cycle of the deviation analysis (e.g., 24 hours).

[0118] Step S1383: Construct a correlation analysis model, taking the sequence of virtual grid node characteristic parameters as independent variables and the sequence of physical system operating parameters as dependent variables, and calculate the linear correlation coefficient between the two through correlation analysis.

[0119] The Pearson correlation coefficient is used to calculate the linear correlation between the virtual medium transport coefficient (X) and the oxygen concentration (Y), with the formula r = Σ[(Xi-X̄)(Yi-Ȳ)] / √[Σ(Xi-X̄)²Σ(Yi-Ȳ)²], where X̄ and Ȳ are the average values ​​of the X and Y sequences, respectively. The correlation coefficient r ranges from -1 to 1, with |r| closer to 1 indicating a stronger linear correlation. For example, a node with r = -0.85 indicates a strong negative correlation between the virtual medium transport coefficient and the oxygen concentration (an increase in the coefficient corresponds to a decrease in the oxygen concentration). The correlation coefficients are calculated for all grid nodes to form a correlation coefficient matrix.

[0120] Step S1384: Determine the degree of correlation based on the magnitude of the linear correlation coefficient. An absolute value of the correlation coefficient close to 1 indicates a strong correlation, while an absolute value of the correlation coefficient close to 0 indicates a weak correlation.

[0121] A correlation coefficient threshold of 0.7 is set. Nodes with |r| ≥ 0.7 are considered tightly correlated, while those with |r| < 0.7 are considered loosely correlated. For example, 80% of the grid nodes have an absolute correlation coefficient between 0.75 and 0.9, indicating a generally strong correlation between the virtual medium transmission coefficient and oxygen concentration; a few nodes (such as those in corner areas) have a correlation coefficient of 0.4, indicating a loose correlation, which may be affected by cargo obstruction.

[0122] Step S1385: Perform lag correlation analysis on closely related parameter pairs, calculate the correlation coefficients under different lag times, and determine the lag time of the influence of changes in virtual grid node characteristic parameters on changes in physical system operating parameters.

[0123] Lag correlation analysis is used to determine the time interval at which changes in virtual parameters precede changes in physical parameters. Taking a closely related node pair as an example, the virtual medium transport coefficient sequence (X) is shifted backward by τ time units (τ = 1, 2, ..., 10 seconds). The correlation coefficient between the shifted X sequence and the original oxygen concentration sequence (Y) is calculated, and the τ value corresponding to the maximum correlation coefficient is found as the lag time. For example, the correlation coefficient is maximum at τ = 3 seconds (r = -0.88), indicating that the physical oxygen concentration responds 3 seconds after the virtual medium transport coefficient changes. The lag time distribution is used for time compensation during subsequent parameter adjustments.

[0124] Step S1386: Based on the lag time and correlation coefficient, construct a parameter influence transmission model to describe how the characteristic parameters of virtual grid nodes affect the deviation of physical system operating parameters through the internal mechanism of the model.

[0125] The parameter impact propagation model uses a causal graph: virtual medium transport coefficient (cause) → oxygen diffusion rate (intermediate variable) → oxygen concentration distribution (effect). The intensity of the impact is quantified by the correlation coefficient, and the propagation speed is quantified by the lag time. For example, the virtual medium transport coefficient of node A (correlation coefficient -0.85, lag time 3 seconds) → oxygen diffusion rate decreases by 20% → oxygen concentration increases by 0.5%, forming a complete impact chain. The model also needs to consider the coupling effect between nodes. For example, if the parameter change of node A affects the oxygen concentration of node B through a strongly correlated path, an indirect impact path needs to be added to the model.

[0126] Step S1387: Perform sensitivity analysis on the parameter influence transmission model. By changing the values ​​of the characteristic parameters of the virtual mesh nodes, detect the magnitude of the change in the deviation of the physical system operating parameters and determine the sensitivity of the parameters.

[0127] Sensitivity analysis employed the controlled variable method: In the virtual model, other parameters were fixed, and the transmission coefficient of a certain virtual medium was increased by 10%. After running the simulation, the change in physical oxygen concentration deviation (change in deviation / change in parameter) was calculated. The larger the change, the higher the sensitivity of that parameter. For example, after increasing the virtual medium transmission coefficient of node C by 10%, the oxygen concentration deviation decreased from 0.8% to 0.5%, a change of (0.5-0.8) / 0.8 = -37.5%, indicating a relatively high sensitivity. Similarly, when the parameter of node D changed by 10%, the deviation change was only -5%, indicating a relatively low sensitivity. The sensitivity was converted to a dimensionless value between 0 and 1 through standardization (dividing by the maximum change).

[0128] Step S1388: Mark the characteristic parameters of virtual grid nodes whose sensitivity exceeds the preset sensitivity threshold as key influencing parameters, and record their parameter identifier, correlation coefficient, lag time and sensitivity.

[0129] The preset sensitivity threshold is set to 0.6 (i.e., the change range reaches 60% of the maximum change range). The sensitivity of all virtual medium transmission coefficients is iterated, and parameters with a sensitivity ≥ 0.6 are marked as key influencing parameters. For example, node A (sensitivity 0.85) and node B (sensitivity 0.72) are marked as key influencing parameters, and their node ID, correlation coefficient (-0.85, -0.78), lag time (3 seconds, 4 seconds), and sensitivity (0.85, 0.72) are recorded.

[0130] Step S1389: By changing the values ​​of the physical system operating parameters, detect whether the corresponding virtual mesh node characteristic parameters change as expected according to the parameter influence transmission model, and verify the accuracy of the correlation.

[0131] In the physical system, the opening of the nitrogen release valve is manually adjusted (changing the physical operating parameters) to increase the local oxygen concentration by 0.5%. The corresponding virtual medium transport coefficient is then observed to see if it decreases as expected by the model (because an increase in oxygen concentration should correspond to a lower diffusion coefficient). For example, after the physical oxygen concentration increases by 0.5%, the medium transport coefficient of node A in the virtual model is expected to decrease by 5%, but in the actual simulation it decreases by 4.8%, with an error of <5%, verifying the accuracy of the correlation. The expected decrease for node E is 3%, but the actual decrease is 1%, with an error >30%, requiring recalibration of the parameter influence transfer model for this node.

[0132] Step S13810: Organize the analysis results of all key influencing parameters, generate a deviation source tracing report, and determine the key influencing parameters and related characteristics corresponding to each data type whose deviation intensity exceeds the preset deviation intensity threshold.

[0133] The deviation source tracing report includes: deviation data type (oxygen concentration deviation), a list of key influencing parameters (node ​​ID, parameter name), correlation characteristics (correlation coefficient, lag time, sensitivity), impact transmission path, and verification results.

[0134] Step S139: Using the path tracing function of the deviation source tracing method, locate the grid node feature parameters that cause the deviation, calculate the contribution of each parameter to the deviation, sort according to the contribution, filter out the grid node feature parameters whose contribution exceeds the preset contribution threshold, and determine them as the core grid parameters affecting oxygen control stability.

[0135] The path tracing function is implemented using a Bayesian network inference algorithm. Given an observation of a bias data type (e.g., oxygen concentration bias = 0.8%), the posterior probability of the feature parameter of each virtual grid node is calculated. A higher posterior probability indicates a higher likelihood that the parameter will cause the bias. The contribution is calculated using variance decomposition, which represents the proportion of the bias variance explained by the parameter's change. A preset contribution threshold of 10% is set, and parameters with a contribution exceeding 10% are selected as core grid parameters. For example, if the virtual medium transport coefficient contributes 25% and the virtual pressure regulation threshold contributes 15%, both are determined as core grid parameters.

[0136] Step S140: Design a parameter adaptive optimization rule system based on the influence characteristics of the core grid parameters, adjust the grid node configuration parameters of the virtual grid nitrogen storage and oxygen control model based on the parameter adaptive optimization rule system, and generate an optimized parameter set. The parameter adaptive optimization rule system is constructed based on the multi-objective influence characteristics of the core grid parameters on oxygen control accuracy, nitrogen consumption, pressure stability and energy consumption level, and combined with the set multi-objective constraints. It includes an interactive influence compensation mechanism for handling the coupling influence between parameters, and rules for collaborative adjustment and comprehensive evaluation of the core grid parameters based on the parameter influence evaluation model.

[0137] This step designs optimization rules and adjusts parameters based on the core grid parameters determined in step S139 to reduce deviations and improve oxygen control stability. The optimization process is described in detail below using the virtual medium transport coefficient and virtual pressure regulation threshold as examples.

[0138] Step S141: Analyze the influence characteristics of each core grid parameter, including the influence of parameter changes on oxygen concentration control accuracy, nitrogen consumption, system pressure stability, and energy consumption level.

[0139] The influence characteristics of core mesh parameters were analyzed using the controlled variable method: In the virtual model, keeping other parameters constant, a single core mesh parameter was changed (e.g., increasing the virtual medium transport coefficient by 5%), and simulations were run, recording changes in oxygen concentration control accuracy (standard deviation), nitrogen consumption, pressure fluctuation amplitude (standard deviation), and energy consumption. The parameter values ​​were changed repeatedly to obtain the relationship curves between parameter changes and changes in each performance index. For example, the virtual medium transport coefficient showed a negative correlation with oxygen concentration control accuracy (increased coefficient, improved accuracy) and a positive correlation with nitrogen consumption (increased coefficient, increased consumption); the virtual pressure regulation threshold showed a negative correlation with pressure fluctuation amplitude (increased threshold, decreased fluctuation) and a positive correlation with energy consumption (increased threshold, increased energy consumption). These influence patterns were fitted into a mathematical model, such as a linear or quadratic model, as the basis for parameter optimization.

[0140] Step S142: Construct a parameter impact assessment model based on the influence law, input different parameter values, and output the corresponding oxygen control accuracy assessment value, nitrogen consumption assessment value, pressure stability assessment value, and energy consumption assessment value.

[0141] The parameter impact assessment model employs a feedforward neural network architecture. The input layer consists of the value vectors of the core grid parameters, the hidden layers contain two fully connected layers (each with twice the number of neurons as the input layer), and the output layer contains four evaluation values ​​(oxygen control accuracy, nitrogen consumption, pressure stability, and energy consumption). The training dataset for the neural network is obtained through Latin hypercube sampling of the virtual model, covering the value range of the core grid parameters, with a sample size 1000 times the number of parameters. The training process uses the Adam optimizer, with the loss function being the mean squared error between the evaluation values ​​and the actual output of the virtual model. The training iterations are 1000 times until the loss function converges. After the model is trained, its generalization ability is verified using a test set to ensure that the relative error of the evaluation values ​​is less than 5%.

[0142] Step S143: Set multi-objective constraints for parameter optimization, including oxygen control accuracy constraint range, nitrogen consumption constraint range, pressure fluctuation constraint range, and energy consumption constraint range.

[0143] The multi-objective constraints are set based on the actual needs of a cold chain logistics center in a certain region: the oxygen control accuracy constraint range is ≤0.5% of the standard deviation of the system's average oxygen concentration; the nitrogen consumption constraint range is ≤105% of the total nitrogen consumption of the design value; the pressure fluctuation constraint range is ≤50Pa of the system pressure fluctuation amplitude; and the energy consumption constraint range is ≤110% of the total energy consumption of the system of the design value. The upper and lower limits of the constraints are determined according to industry standards and project requirements.

[0144] Step S144: Design a parameter adaptive optimization rule system based on multi-objective constraints. The optimization rule system includes the priority order of parameter adjustment, the method for determining the step size of parameter changes, the termination condition of parameter adjustment, and the compensation mechanism for parameter interaction effects.

[0145] The priority of parameter adjustments is determined based on the contribution of core grid parameters to oxygen control stability. Parameters with higher contributions are adjusted first; for example, the virtual medium transport coefficient (25% contribution) takes precedence over the virtual pressure regulation threshold (15% contribution). An adaptive step-size strategy is used to determine the step size of parameter changes. The initial step size is set to 5% of the parameter value range. If the objective function improves after adjustment, the step size is increased (multiplied by 1.2); if it deteriorates, the step size is decreased (multiplied by 0.5). Termination conditions for parameter adjustments include: reaching the upper limit of iterations (100 times), the change in the objective function being less than the threshold (0.1%), or all evaluation values ​​meeting the constraints. A parameter interaction compensation mechanism is used to handle the coupling relationship between parameters. For example, when adjusting the virtual medium transport coefficient, the virtual pressure regulation threshold is adjusted synchronously according to the correlation model between the two. The compensation coefficient is determined through the influence law in step S141.

[0146] Step S145: Adjust the core grid parameters that have the most significant impact on oxygen control accuracy according to the priority order of parameter adjustment, and set the initial adjustment step size according to the parameter change step size determination method.

[0147] First, select the highest priority core mesh parameter, such as the virtual medium transport coefficient. Based on its current value and range (e.g., 0.8-1.2 m² / s), set the initial adjustment step size to 5% of the range (0.02 m² / s). The adjustment direction is determined by the deviation direction: if the oxygen concentration deviation is positive (simulated value > actual value), decrease the virtual medium transport coefficient; if the deviation is negative, increase the virtual medium transport coefficient. The initial adjustment step size is pre-validated using the parameter influence evaluation model to ensure that the adjustment does not lead to a significant deterioration in other evaluation indicators.

[0148] Step S146: Calculate the impact of the current core mesh parameter change on other core mesh parameters, and preset the compensation adjustment range of other core mesh parameters according to the impact.

[0149] Based on the parameter influence law established in step S141, calculate the influence coefficient of the current parameter change on other core mesh parameters. For example, if the virtual medium transport coefficient changes by 0.1 m² / s and the virtual pressure regulation threshold changes by 0.05 kPa, the influence coefficient is 0.5. According to the influence coefficient and the constraints of other core mesh parameters, calculate the compensation adjustment range: Lower compensation limit = Current value - Influence coefficient × Current parameter adjustment amount; Upper compensation limit = Current value + Influence coefficient × Current parameter adjustment amount. The compensation adjustment range must not exceed the parameter's value range; if it does, the boundary value of the value range is used.

[0150] Step S147: Input the parameter value corresponding to the initial adjustment step size into the parameter influence evaluation model. If the output evaluation value satisfies the multi-objective constraint, record the parameter value as a candidate optimization parameter.

[0151] The adjusted core grid parameter values ​​(current value ± initial step size) are input into the parameter influence evaluation model to obtain evaluation values ​​for oxygen control accuracy, nitrogen consumption, pressure stability, and energy consumption. Each evaluation value is checked to see if it falls within the constraints. If all constraints are met, the parameter value is recorded as a candidate optimization parameter, and adjustment of that parameter is stopped. If any evaluation value does not meet the constraints, the step size needs further adjustment.

[0152] Step S148: If the output evaluation value does not meet the multi-objective constraint conditions, adjust the parameter adjustment step size, and simultaneously perform synchronous compensation adjustment on other core grid parameters within the compensation adjustment range. Re-input the parameters to influence the evaluation model for evaluation until candidate optimization parameters that meet the constraint conditions are obtained.

[0153] If the evaluated values ​​do not meet the constraints, the step size is adjusted according to the type of non-met requirement: if the oxygen control accuracy is not met (standard deviation > 0.5%), the step size is increased (e.g., from 0.02 m² / s to 0.03 m² / s) to enhance the regulation; if the nitrogen consumption is not met (> 105% of the design value), the step size is decreased (e.g., reduced to 0.01 m² / s) to reduce consumption. Simultaneously, based on the compensation adjustment range calculated in step S146, other core grid parameters are adjusted synchronously. For example, when the virtual medium transport coefficient increases by 0.03 m² / s, the virtual pressure regulation threshold increases by 0.015 kPa. The adjusted parameter combination is input into the evaluation model for re-evaluation. This process is repeated until all evaluated values ​​meet the constraints, yielding candidate optimized parameters for the current core grid parameters.

[0154] Step S149: Thus, all core grid parameters are adjusted sequentially to obtain candidate optimization parameters corresponding to each core grid parameter.

[0155] Taking the virtual medium transmission coefficient (core mesh parameter 1) and virtual pressure regulation threshold (core mesh parameter 2) as examples, the adjustment process is achieved through the following sub-steps:

[0156] For example, step S1491: Select the first core mesh parameter from the core mesh parameter list according to the parameter adjustment priority, and determine the current value and parameter adjustment range of the core mesh parameter.

[0157] The core grid parameter list is sorted by priority as follows: Virtual medium transfer coefficient (priority 1), Virtual pressure regulation threshold (priority 2), and Virtual energy consumption calculation coefficient (priority 3). The first parameter is selected as the virtual medium transfer coefficient, which currently has a value of 0.9 m² / s and can be adjusted within the range of 0.7-1.1 m² / s (determined according to the upper and lower limits of the medium transfer coefficient allowed by the system design).

[0158] Step S1492: Based on the method for determining the step size of parameter changes, and combined with the influence characteristics of the core grid parameters, calculate the initial adjustment step size. The size of the initial adjustment step size is positively correlated with the sensitivity of the parameters to the oxygen control accuracy.

[0159] The sensitivity of parameters to oxygen control accuracy was determined through the influence characteristic analysis in step S141. The sensitivity coefficient of the virtual medium transmission coefficient was 0.8 (out of 1). The step size was determined as follows: initial adjustment step size = parameter adjustment range × sensitivity coefficient × 0.1, i.e., (1.1-0.7) × 0.8 × 0.1 = 0.032 m² / s. 0.03 m² / s was chosen as the initial step size. Higher sensitivity requires a larger step size to ensure rapid convergence of parameters that significantly affect oxygen control accuracy.

[0160] Step S1493: Analyze the correlation between the core mesh parameter and other core mesh parameters, determine the list of related parameters, and calculate the influence coefficient of each unit change in the core mesh parameter on each parameter in the list of related parameters.

[0161] Through parameter correlation matrix analysis, a strong correlation was found between the virtual medium transmission coefficient and the virtual pressure regulation threshold (correlation coefficient 0.6), and the [virtual pressure regulation threshold] was identified as the associated parameter. The influence coefficient was calculated as follows: for every 1 m² / s change in the virtual medium transmission coefficient, the virtual pressure regulation threshold changes by 0.5 kPa (obtained based on regression analysis of historical data).

[0162] Step S1494: Based on the current values ​​of the influence coefficient and the correlation parameter, determine the compensation adjustment range for each correlation parameter. The upper and lower limits of the compensation adjustment range are calculated based on the allowable variation range of the influence coefficient and the correlation parameter.

[0163] The current value of the associated parameter virtual pressure adjustment threshold is 2.0 kPa, with an allowable range of 1.8-2.2 kPa. The compensation adjustment range = current value ± (influence coefficient × core parameter adjustment step size), i.e., 2.0 ± (0.5 × 0.03) = 2.0 ± 0.015 kPa. Therefore, the compensation adjustment range is 1.985-2.015 kPa, which is within the allowable range and does not require truncation. If the calculated compensation range exceeds the allowable range, the boundary value of the allowable range is taken as the upper or lower limit of the compensation.

[0164] Step S1495: Add the current value to the initial adjustment step size to obtain the first test value. At the same time, calculate the compensation adjustment value of each parameter in the related parameter list according to the influence coefficient, and adjust the related parameters synchronously within the compensation adjustment range.

[0165] Current value + initial adjustment step size = 0.9 + 0.03 = 0.93 m² / s (first test value). Correlation parameter compensation adjustment value = influence coefficient × initial adjustment step size = 0.5 × 0.03 = 0.015 kPa. Therefore, the virtual pressure regulation threshold is adjusted to 2.0 + 0.015 = 2.015 kPa (upper limit of the compensation adjustment range).

[0166] Step S1496: Input the first test value and the adjusted associated parameter value into the parameter influence assessment model to obtain the corresponding oxygen control accuracy assessment value, nitrogen consumption assessment value, pressure stability assessment value and energy consumption assessment value.

[0167] The parameter impact assessment model inputs are (virtual medium transmission coefficient = 0.93 m² / s, virtual pressure regulation threshold = 2.015 kPa), and the output assessment values ​​are: oxygen control accuracy = 0.45% (standard deviation), nitrogen consumption = 102% (relative to design value), pressure stability = 0.3 kPa (standard deviation), and energy consumption = 105% (relative to design value).

[0168] Step S1497: Determine whether the oxygen control accuracy assessment value, nitrogen consumption assessment value, pressure stability assessment value and energy consumption assessment value all meet the multi-objective constraint conditions. If all are met, the first test value is used as the candidate optimization parameter of the core grid parameter, and the adjustment value of the associated parameter is recorded.

[0169] The multi-objective constraints are: oxygen control accuracy ≤ 0.5%, nitrogen consumption ≤ 105%, pressure stability ≤ 0.4 kPa, and energy consumption ≤ 110%. The current evaluation values ​​all meet the constraints; therefore, 0.93 m² / s is selected as a candidate optimization parameter for the virtual medium transport coefficient, and the adjusted value of the associated parameter is recorded as 2.015 kPa.

[0170] Step S1498: If at least one evaluation value does not meet the constraint conditions, determine whether the evaluation value exceeds the upper limit of the constraint or is lower than the lower limit of the constraint. If it exceeds the upper limit of the constraint, subtract the initial adjustment step size from the current value to obtain the second test value, and recalculate the compensation adjustment value of the associated parameter.

[0171] Assuming the nitrogen consumption assessment value for the first test is 106% (exceeding the constraint limit of 105%), the virtual medium transport coefficient needs to be reduced. Current value - initial adjustment step size = 0.9 - 0.03 = 0.87 m² / s (second test value). Corresponding parameter compensation adjustment value = 0.5 × (-0.03) = -0.015 kPa, and the virtual pressure regulation threshold is adjusted to 2.0 - 0.015 = 1.985 kPa (lower limit of the compensation adjustment range).

[0172] Step S1499: If the value is below the lower limit of the constraint, add the initial adjustment step size to the first test value to obtain the second test value. Similarly, recalculate the compensation adjustment value of the associated parameter, and input the second test value and the new associated parameter adjustment value into the parameter impact assessment model for reassessment.

[0173] If the oxygen control accuracy assessment value is 0.55% (exceeding the upper limit of the constraint by 0.5%), the virtual medium transmission coefficient needs to be further increased. The second test value is 0.93 + 0.03 = 0.96 m² / s, the related parameter adjustment value is 0.5 × 0.03 = 0.015 kPa, and the pressure regulation threshold is 2.015 + 0.015 = 2.03 kPa (exceeding the compensation adjustment range, the upper limit of allowable 2.2 kPa is taken).

[0174] Step S14910: Repeat the above process of generating test values, adjusting and evaluating related parameters. Each time, adjust the direction and step size of the test values ​​based on the previous evaluation results. The step size adjustment range is determined based on the deviation between the evaluation value and the constraint conditions. The larger the deviation, the larger the step size adjustment range. At the same time, dynamically update the compensation adjustment range of the related parameters until a test value that satisfies all multi-objective constraints is obtained. Determine the test value as a candidate optimization parameter for the current core mesh parameters and record the corresponding related parameter adjustment value.

[0175] If the second test value still does not meet the constraints, such as nitrogen consumption 104% (met), oxygen control accuracy 0.48% (met), pressure stability 0.35 kPa (met), and energy consumption 108% (met), then 0.87 m² / s becomes a candidate optimization parameter. If it still does not meet the constraints, adjust the step size to 50% of the initial step size (0.015 m² / s) and continue testing values ​​such as 0.885 m² / s until a value that meets all constraints is found.

[0176] Step S14911: Record the candidate optimization parameters and related parameter adjustment values ​​of the current core mesh parameters into the optimization parameter record table, and select the next core mesh parameter from the core mesh parameter list.

[0177] Add the following entry to the optimization parameter record table: Parameter Name = Virtual Medium Transmission Coefficient, Candidate Optimization Parameter = 0.93 m² / s, Correlated Parameter = Virtual Pressure Adjustment Threshold, Correlated Adjustment Value = 2.015 kPa. Select the next core grid parameter as the virtual pressure adjustment threshold and repeat steps S1491-S14910 for adjustment.

[0178] Step S14912: Following the same adjustment process, test values ​​for the next core grid parameter, generate associated parameter compensation adjustments and filters, until all core grid parameters have corresponding candidate optimization parameters determined, and complete the adjustment of all core grid parameters.

[0179] For the virtual pressure adjustment threshold (currently set at 2.0 kPa, with an adjustment range of 1.8-2.2 kPa), the initial step size and compensation adjustment range (with the virtual energy consumption calculation coefficient as the associated parameter) are calculated according to the above procedure. Values ​​are tested and evaluated, and the final candidate optimization parameter is determined to be 2.05 kPa, with an associated adjustment value of 0.92 (virtual energy consumption calculation coefficient). After all core parameters are adjusted sequentially, the optimization parameter record table contains the candidate value and associated adjustment value for each parameter.

[0180] Step S1410: Combine all candidate optimization parameters to generate multiple parameter combination schemes. Input each parameter combination scheme into the parameter influence assessment model to obtain the corresponding oxygen control accuracy assessment value, nitrogen consumption assessment value, pressure stability assessment value, and energy consumption assessment value.

[0181] If there are n core grid parameters, and each parameter has m candidate optimization parameters, then the number of parameter combinations is mⁿ. To reduce the number of combinations, an orthogonal experimental design method is used to select representative combinations, such as L9(3 4 The orthogonal array can generate 9 sets of schemes, covering 3 levels for each of the 4 parameters. Each set of parameter combinations is input into the parameter impact assessment model to obtain the corresponding 4 evaluation values, which are stored in the scheme evaluation matrix.

[0182] Step S1411: Based on the multi-objective constraints, normalize each evaluation value into a dimensionless score value, and perform a weighted summation of the oxygen control accuracy score value, nitrogen consumption score value, pressure stability score value, and energy consumption score value according to the preset weights to calculate the comprehensive evaluation score of each scheme; select the parameter combination scheme with the highest comprehensive evaluation score as the optimization parameter set according to the ranking of the comprehensive evaluation scores.

[0183] The normalization of evaluation values ​​uses a linear transformation method: for positive indicators (such as oxygen control accuracy and pressure stability), the score = (evaluation value - minimum value) / (maximum value - minimum value); for negative indicators (such as nitrogen consumption and energy consumption), the score = (maximum value - evaluation value) / (maximum value - minimum value). Preset weights are set according to system priority: oxygen control accuracy weight 0.4, pressure stability weight 0.3, nitrogen consumption weight 0.2, and energy consumption weight 0.1. The comprehensive evaluation score is the sum of the products of each score value and its corresponding weight, with a maximum score of 1. The parameter combination scheme with the highest comprehensive evaluation score is selected as the optimization parameter set; this scheme has the optimal comprehensive performance while satisfying all constraints.

[0184] Step S150: Load the optimized parameter set into the virtual grid nitrogen storage and oxygen control model to perform closed-loop simulation verification, generate a simulation optimization scheme that meets the requirements of oxygen control accuracy and energy consumption coordination, generate execution control commands for the physical nitrogen storage and oxygen control system based on the simulation optimization scheme, and transmit them to the corresponding execution components.

[0185] This step verifies the effectiveness of the optimized parameter set through closed-loop simulation, and then applies the verified optimization scheme to the physical system to achieve oxygen control optimization. The process is described in detail below.

[0186] Step S151: Load the optimized parameter set into the virtual grid nitrogen storage and oxygen control model, and update the configuration parameters of each grid node in the virtual grid nitrogen storage and oxygen control model. The configuration parameters include the virtual medium transmission coefficient, the virtual pressure regulation threshold, and the virtual energy consumption calculation coefficient.

[0187] The optimized parameter set is stored in an XML file, containing the name, grid node number, and optimized value of each core grid parameter. The virtual grid nitrogen storage and oxygen control model provides a parameter update interface, which writes the optimized parameters to the model's grid node configuration file via API calls. During the parameter update process, the model automatically verifies the validity of the parameters (e.g., whether they are within the range of values). If illegal parameters are found, the update is rejected and an error message is returned. After the parameter update is complete, the model reinitializes the grid node state to ensure that the new parameters take effect. The update process takes less than 30 seconds, without affecting the continuity of the simulation.

[0188] Step S152: Set the operating parameters for closed-loop simulation verification. The operating parameters include simulation verification duration, data sampling interval, and verification operating condition type. The verification operating condition type includes normal operating condition, peak load operating condition, low load operating condition, sudden disturbance operating condition, and long-term operating condition.

[0189] The simulation verification duration is set according to the operating condition type: 24 hours for normal operating conditions, peak load operating conditions, low load operating conditions, and sudden disturbance operating conditions; and 7 days for long-term operation conditions. The data sampling interval is consistent with the data interaction frequency in the coupled simulation control command, set to 1 second. The verification operating condition types cover typical operating scenarios of a cold chain logistics center in a certain region. The setting parameters for each operating condition (such as the number of door openings, cargo volume, and ambient temperature) are the same as those in the dynamic response consistency analysis in step S117, ensuring the comprehensiveness of the verification.

[0190] Step S153: Start the closed-loop simulation verification of the virtual grid nitrogen storage and oxygen control model, execute the simulation under different verification conditions, and collect grid node status data and overall system performance data in real time during the simulation process.

[0191] Closed-loop simulation verification employs an automated script execution mechanism, sequentially initiating simulation runs for each operating condition. During the simulation, the model receives real-time data from the physical system (such as actual oxygen concentration and nitrogen flow rate) through dynamic coupling and applies optimized parameter sets to the grid nodes, outputting simulated grid node status data (medium transmission volume, pressure transmission volume, etc.) and overall system performance data (average oxygen concentration, total energy consumption, etc.). Data acquisition and storage methods are the same as in steps S123-S128, ensuring data integrity and traceability.

[0192] Step S154: Analyze the collected simulation data to determine whether the oxygen control accuracy of the model meets the preset oxygen control accuracy requirements and whether the energy consumption level meets the preset energy consumption requirements under different verification conditions.

[0193] Simulation data analysis metrics include oxygen control accuracy and energy consumption level: oxygen control accuracy is evaluated using the standard deviation of the system's average oxygen concentration, with a preset requirement of ≤0.5%; energy consumption level is evaluated using the ratio of total energy consumption to the design value, with a preset requirement of ≤110%. For each verification condition, the oxygen control accuracy and energy consumption level are calculated throughout the entire simulation cycle to determine whether the preset requirements are met. In addition, pressure stability and nitrogen consumption are analyzed as auxiliary evaluation metrics to ensure optimal overall system performance.

[0194] Step S155: If the oxygen control accuracy or energy consumption level does not meet the requirements under any verification condition, return to the parameter adaptive optimization rule system design step, adjust the parameter adjustment step size, multi-objective constraint conditions or parameter interaction effect compensation mechanism in the optimization rule system, regenerate the optimization parameter set and perform closed-loop simulation verification.

[0195] If the oxygen control accuracy for a certain operating condition is 0.6% (>0.5%), return to step S144 and adjust the parameter adjustment step size (e.g., increase the step size of the virtual medium transmission coefficient) or the parameter interaction compensation mechanism (e.g., increase the compensation strength for the virtual pressure regulation threshold). If the energy consumption level is 115% (>110%), adjust the multi-objective constraints (e.g., tighten the energy consumption constraint range to ≤105%) or reduce the energy consumption weight. After adjustment, re-execute steps S145-S1411 to generate a new set of optimized parameters, and then repeat the closed-loop simulation verification of steps S151-S154 until the oxygen control accuracy and energy consumption level for all operating conditions meet the requirements.

[0196] Step S156: If the oxygen control accuracy and energy consumption level meet the requirements under all verification conditions, stop the closed-loop simulation verification, and organize all data in the simulation process, including optimized parameter values, oxygen control accuracy data, energy consumption data, pressure stability data, and anti-disturbance capability data under each condition.

[0197] The closed-loop simulation verification passed when all validation conditions met the oxygen control accuracy and energy consumption requirements. The compiled data includes: specific values ​​for the optimized parameter set (e.g., virtual medium transmission coefficient = 1.05 m² / s, virtual pressure regulation threshold = 2.3 kPa); oxygen control accuracy curves (24-hour standard deviation variation), energy consumption curves (hourly energy consumption), and pressure stability curves (pressure fluctuation amplitude variation) under each condition; and data on disturbance resistance under sudden disturbance conditions (recovery time after disturbance, maximum deviation). The data was compiled into a standardized report, including data charts, statistical indicators (average, maximum, minimum values), and analytical conclusions.

[0198] Step S157: Generate a simulation optimization scheme based on the processed simulation data. The simulation optimization scheme includes an optimization parameter configuration table, operation control strategies under various operating conditions, emergency plans for parameter adjustment, and operating condition adaptation suggestions.

[0199] The optimized parameter configuration table lists the names, grid node numbers, optimized values, value ranges, and adjustment dates of all core grid parameters in tabular form. The operational control strategies for each operating condition specify specific control parameters for different conditions, such as a target oxygen concentration of 15.5% under normal operating conditions, increased to 15.8% under peak load conditions to address oxygen intrusion caused by door opening. The parameter adjustment contingency plan specifies temporary replacement values ​​and switching procedures when a core grid parameter fails. Operating condition adaptation suggestions are provided based on simulation data, such as reducing the nitrogen generator's operating frequency to save energy under low load conditions. The simulation optimization scheme is compiled in PDF format and serves as the basis for implementing the physical system optimization.

[0200] Step S158: Extract the control parameters corresponding to the physical nitrogen storage and oxygen control system from the simulation optimization scheme. The control parameters include nitrogen input rate adjustment parameters, oxygen concentration feedback adjustment parameters, pressure control parameters, energy consumption optimization parameters, and anti-disturbance adjustment parameters.

[0201] The extraction of control parameters is based on the mapping relationship between virtual mesh parameters and physical system control parameters: nitrogen input rate regulation parameters correspond to the virtual nitrogen flow rate and the PID controller parameters (proportional coefficient, integral time, derivative time) of the nitrogen generator; oxygen concentration feedback regulation parameters correspond to the virtual oxygen concentration and the opening curve of the nitrogen release valve; pressure control parameters correspond to the virtual pressure regulation threshold and the set value of the pressure regulating valve; energy consumption optimization parameters correspond to the virtual energy consumption calculation coefficient and equipment operation scheduling strategy (such as the start-up and shutdown sequence of the nitrogen generator); disturbance rejection regulation parameters correspond to the disturbance rejection capability data of the virtual model and the parameters of the disturbance compensation algorithm (such as feedforward control coefficients). The extracted control parameters are categorized by functional subsystems to form a control parameter table for the physical system.

[0202] Step S159: Generate execution control instructions based on the extracted control parameters. The execution control instructions include instruction identifier, execution component identifier, parameter setting value, execution time requirement, and instruction priority.

[0203] The execution of control commands adopts the standard format in the industrial control field, with each command identified by a unique UUID. The execution component identifier corresponds to the specific equipment in the physical system (e.g., nitrogen generator 1, valve group A). ​​Parameter settings are extracted control parameter values ​​(e.g., PID proportional coefficient = 2.5). Execution time requirements are the command's effective time (e.g., immediate execution or timed execution). Command priorities are divided into emergency (level 1), high (level 2), medium (level 3), and low (level 4), with safety-related commands set to emergency priority. After command generation, verification ensures that parameter settings are within the equipment's allowable range to prevent equipment malfunction.

[0204] Step S1510: Establish communication links with each execution component of the physical nitrogen storage and oxygen control system, transmit execution control commands to the corresponding execution components through a preset communication protocol, and receive command execution confirmation information returned by the execution components to confirm that the execution components have adjusted the operating parameters according to the execution control commands.

[0205] The communication link employs a redundant design, with the primary link being Industrial Ethernet (Profinet) and the backup link being a wireless LoRa network to ensure communication reliability. The communication protocol is selected based on the type of actuator; for example, PLCs use Modbus TCP / IP, while smart valves use HART. Control commands are transmitted to the actuator via the communication link. Upon receiving the command, the actuator sets its parameters and returns confirmation information including the command identifier, execution status (success / failure), and current parameter values. If reception fails or the execution status is negative, the system automatically retransmits the command (up to 3 times); if retransmission fails, an alarm is triggered. After confirming that the actuator has correctly adjusted its operating parameters, the physical implementation of the entire simulation optimization scheme is completed.

[0206] In one exemplary embodiment, a simulation and optimization system for a virtual grid nitrogen storage and oxygen control system incorporating digital twins is provided. This simulation and optimization system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2As shown, the virtual grid nitrogen storage and oxygen control system simulation and optimization system combining digital twins includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for information exchange between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by a processor, this computer program implements a simulation optimization method for a virtual grid nitrogen storage and oxygen control system incorporating digital twins. The display unit of this simulation optimization system, used to form a visually visible image, can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device for this simulation optimization system can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the system's casing, or an external keyboard, touchpad, or mouse.

[0207] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A simulation optimization method for a virtual grid nitrogen storage and oxygen control system combining digital twins, characterized in that, The method includes: A cross-domain dynamic coupling relationship is constructed between the physical nitrogen storage and oxygen control system and the virtual grid nitrogen storage and oxygen control model. Multi-dimensional operation data of the physical nitrogen storage and oxygen control system and grid node feature data of the virtual grid nitrogen storage and oxygen control model are collected. Coupled simulation control commands are generated by aligning cross-domain data features. Based on the coupled simulation control command, the virtual grid nitrogen storage and oxygen control model is driven to perform grid node collaborative simulation operation, and the status interaction data of grid nodes and the overall system operation data are collected in real time during the simulation process to generate a multi-dimensional simulation dataset. A hierarchical deviation analysis was performed on the multi-dimensional simulation dataset and the actual operation dataset of the physical nitrogen storage and oxygen control system to construct a multi-dimensional deviation data matrix. Based on the deviation source tracing method, the core grid parameters affecting the stability of oxygen control were located. Based on the influence characteristics of the core grid parameters, a parameter adaptive optimization rule system is designed. Based on the parameter adaptive optimization rule system, the grid node configuration parameters of the virtual grid nitrogen storage and oxygen control model are adjusted to generate an optimized parameter set. The parameter adaptive optimization rule system is constructed based on the multi-objective influence characteristics of the core grid parameters on oxygen control accuracy, nitrogen consumption, pressure stability and energy consumption level, and combined with the set multi-objective constraints. It includes an interactive influence compensation mechanism for handling the coupling influence between parameters, as well as rules for collaborative adjustment and comprehensive evaluation of the core grid parameters based on the parameter influence evaluation model. The optimized parameter set is loaded into the virtual grid nitrogen storage and oxygen control model to perform closed-loop simulation verification, generating a simulation optimization scheme that meets the requirements of oxygen control accuracy and energy consumption coordination. Based on the simulation optimization scheme, the execution control command of the physical nitrogen storage and oxygen control system is generated and transmitted to the corresponding execution component.

2. The simulation optimization method for a virtual grid nitrogen storage and oxygen control system combined with digital twins according to claim 1, characterized in that, The construction of a cross-domain dynamic coupling relationship between the physical nitrogen storage and oxygen control system and the virtual grid nitrogen storage and oxygen control model involves collecting multi-dimensional operational data of the physical nitrogen storage and oxygen control system and grid node feature data of the virtual grid nitrogen storage and oxygen control model. Coupled simulation control commands are generated through cross-domain data feature alignment, including: The physical nitrogen storage and oxygen control system is divided into functional subsystems. Each functional subsystem is equipped with multiple types of sensors to collect multi-dimensional operating data of the functional subsystem. The multi-dimensional operating data includes oxygen concentration distribution data, nitrogen flow dynamic data, system pressure gradient data, temperature field distribution data, and real-time energy consumption data. Extract the grid node feature data of the virtual grid nitrogen storage and oxygen control model. The grid node feature data includes the virtual functional attributes, virtual spatial location parameters, virtual medium transmission coefficient, virtual pressure regulation threshold and virtual energy consumption calculation coefficient of each grid node. The multi-dimensional operational data and the grid node feature data are subjected to data standardization processing to generate standardized physical data and standardized virtual data. A cross-domain data feature mapping model is constructed to correlate the oxygen concentration distribution data in the standardized physical data with the virtual medium transmission coefficient in the standardized virtual data, and to correlate the nitrogen flow dynamic data in the standardized physical data with the virtual pressure regulation threshold in the standardized virtual data. Calculate the correlation matching degree of cross-domain data features. If the correlation matching degree does not reach the preset matching threshold, adjust the feature weight allocation of the cross-domain data feature mapping model and re-associate the features until the correlation matching degree reaches the preset matching threshold. Based on the cross-domain data feature association results that reach the preset matching threshold, a dynamic coupling association between the physical nitrogen storage and oxygen control system and the virtual grid nitrogen storage and oxygen control model is constructed to determine the collaborative parameters between the two in terms of data interaction, state feedback, and parameter adjustment. The dynamic response consistency between the two was analyzed by collecting multi-dimensional operational data trends of the physical nitrogen storage and oxygen control system under different operating conditions and the grid node characteristic data trends of the virtual grid nitrogen storage and oxygen control model. The collaborative parameters of the dynamic coupling association are adjusted according to the dynamic response consistency results, so that the response delay between changes in physical system data and changes in virtual model data is controlled within a preset range. Based on the adjusted dynamic coupling correlation and coordination parameters, a coupled simulation control command is generated, which includes data interaction frequency, state feedback cycle, and parameter adjustment trigger conditions. The parameter settings of the coupled simulation control command are adapted to the actual operating conditions of the physical nitrogen storage and oxygen control system.

3. The simulation optimization method for a virtual grid nitrogen storage and oxygen control system combined with digital twins according to claim 1, characterized in that, The virtual grid nitrogen storage and oxygen control model driven by the coupled simulation control commands executes grid node collaborative simulation operation, and collects grid node status interaction data and overall system operation data in real time during the simulation process to generate a multi-dimensional simulation dataset, including: The data interaction frequency, state feedback cycle, and parameter adjustment trigger conditions in the coupled simulation control command are analyzed and converted into simulation operation parameters that can be recognized by the virtual grid nitrogen storage and oxygen control model. The state of the grid nodes of the virtual grid nitrogen storage and oxygen control model is initialized according to the simulation operation parameters, and the initial virtual oxygen concentration, initial virtual nitrogen flow rate, initial virtual pressure and initial virtual energy consumption value are assigned to each grid node; The grid nodes of the virtual grid nitrogen storage and oxygen control model are launched for collaborative simulation. During the simulation, the state interaction data between each grid node and its neighboring grid nodes are collected in real time. The state interaction data includes medium transmission data, pressure transmission data, temperature exchange data and energy consumption sharing data. Collect overall system operation data of the virtual grid nitrogen storage and oxygen control model. The overall system operation data includes the system average oxygen concentration, the system total nitrogen consumption, the system pressure fluctuation amplitude, the system temperature distribution uniformity, and the system total energy consumption data. According to the state feedback cycle in the coupled simulation control command, the collected state interaction data and the overall system operation data are time-marked to form time-series interaction data and time-series overall data. Data correlation analysis is performed on the time-series interactive data to identify strong and weak correlation paths of state interaction between grid nodes, and the interaction frequency and interaction amount of strong correlation paths are recorded. Perform trend analysis on the time-series overall data, extract the upward trend segment, downward trend segment and stable trend segment from the overall system operation data, and determine the simulation time interval corresponding to each trend segment; By integrating time-series interactive data, time-series overall data, strongly correlated path information, and trend segment time interval information, a multi-dimensional simulation dataset is constructed according to the simulation time sequence.

4. The simulation optimization method for a virtual grid nitrogen storage and oxygen control system combined with digital twins according to claim 1, characterized in that, The process involves performing a hierarchical deviation analysis on the multi-dimensional simulation dataset and the actual operation dataset of the physical nitrogen storage and oxygen control system, constructing a multi-dimensional deviation data matrix, and locating the core grid parameters affecting oxygen control stability based on the deviation source tracing method, including: The actual operating dataset of the physical nitrogen storage and oxygen control system under the same operating conditions is collected. The actual operating dataset includes actual oxygen concentration distribution data, actual nitrogen flow rate data, actual pressure gradient data, actual temperature field data, and actual energy consumption data, and is marked with time stamps corresponding to the multi-dimensional simulation dataset. The multi-dimensional simulation dataset and the actual running dataset are divided into three levels according to data type: basic parameter layer, interaction relationship layer and system performance layer. Each level contains a corresponding subset of simulation data and a subset of actual data. At the basic parameter layer, the absolute deviation between the oxygen concentration data of the grid nodes in the simulation data subset and the oxygen concentration data of the physical region in the actual data subset is calculated, and the relative deviation between the nitrogen flow rate data of the grid nodes in the simulation data subset and the nitrogen flow rate data of the physical system in the actual data subset is calculated. In the interaction layer, the deviation of medium transmission between grid nodes in the simulation data subset and the deviation of medium transmission between physical subsystems in the actual data subset are calculated, as well as the deviation of pressure transmission between grid nodes in the simulation data subset and the deviation of pressure transmission between physical systems in the actual data subset are calculated. At the system performance level, the deviation of the total system energy consumption in the simulation data subset is compared with the deviation of the actual total system energy consumption in the actual data subset, and the deviation of the system oxygen control accuracy in the simulation data subset is compared with the deviation of the system oxygen control accuracy in the actual data subset. Based on the deviation calculation results at three levels, a multi-dimensional deviation data matrix is ​​constructed. The rows of the multi-dimensional deviation data matrix represent different data types, the columns of the multi-dimensional deviation data matrix represent different time points, and the elements of the multi-dimensional deviation data matrix represent the deviation value of the corresponding data type at the corresponding time point. Perform deviation intensity analysis on the multi-dimensional deviation data matrix, calculate the mean deviation, maximum deviation and deviation fluctuation frequency for each data type, and determine the data types whose deviation intensity exceeds the preset deviation intensity threshold; For data types where the deviation intensity exceeds the preset deviation intensity threshold, the correlation between the virtual mesh node characteristic parameters and the physical system operating parameters corresponding to the data type is analyzed based on the deviation tracing method. By using the path tracing function of the deviation source tracing method, the characteristic parameters of the grid nodes that cause the deviation are located, the contribution of each parameter to the deviation is calculated, and the grid node characteristic parameters whose contribution exceeds the preset contribution threshold are selected and identified as the core grid parameters affecting the stability of oxygen control.

5. The simulation optimization method for a virtual grid nitrogen storage and oxygen control system combined with digital twins according to claim 1, characterized in that, The step involves designing an adaptive optimization rule system based on the influence characteristics of the core grid parameters, adjusting the grid node configuration parameters of the virtual grid nitrogen storage and oxygen control model based on the adaptive optimization rule system, and generating an optimized parameter set, including: The influence characteristics of each core grid parameter were analyzed, including the influence of parameter changes on oxygen concentration control accuracy, nitrogen consumption, system pressure stability, and energy consumption level. Based on the influence patterns, a parameter impact assessment model is constructed. Different parameter values ​​are input, and the corresponding oxygen control accuracy assessment value, nitrogen consumption assessment value, pressure stability assessment value, and energy consumption assessment value are output. Set multi-objective constraints for parameter optimization, including oxygen control accuracy constraint range, nitrogen consumption constraint range, pressure fluctuation constraint range, and energy consumption constraint range. A parameter adaptive optimization rule system is designed based on multi-objective constraints. The optimization rule system includes the priority order of parameter adjustment, the method for determining the step size of parameter changes, the termination condition of parameter adjustment, and the compensation mechanism for parameter interaction effects. According to the priority order of parameter adjustment, the core grid parameters that have the most significant impact on oxygen control accuracy are adjusted, and the initial adjustment step size is set according to the method for determining the step size of parameter changes. Calculate the impact of the current adjustment of core mesh parameters on other core mesh parameters, and preset the compensation adjustment range of other core mesh parameters based on the impact. Input the parameter value corresponding to the initial adjustment step size into the parameter influence evaluation model. If the output evaluation value satisfies the multi-objective constraint, record the parameter value as a candidate optimization parameter. If the output evaluation value does not meet the multi-objective constraint conditions, the parameter adjustment step size is adjusted, and other core grid parameters are simultaneously compensated and adjusted within the compensation adjustment range. The parameters are re-inputted to affect the evaluation model for evaluation until candidate optimization parameters that meet the constraint conditions are obtained. In this way, all core grid parameters are adjusted in turn to obtain the candidate optimization parameters corresponding to each core grid parameter. All candidate optimization parameters are combined to generate multiple parameter combination schemes. Each parameter combination scheme is input into the parameter impact assessment model to obtain the corresponding oxygen control accuracy assessment value, nitrogen consumption assessment value, pressure stability assessment value and energy consumption assessment value. Based on multi-objective constraints, each evaluation value is normalized into a dimensionless score value. The oxygen control accuracy score value, nitrogen consumption score value, pressure stability score value, and energy consumption score value are weighted and summed according to preset weights to calculate the comprehensive evaluation score of each scheme. Based on the ranking of comprehensive evaluation scores, the parameter combination scheme with the highest comprehensive evaluation score is selected as the optimization parameter set.

6. The simulation optimization method for a virtual grid nitrogen storage and oxygen control system combined with digital twins according to claim 1, characterized in that, The process involves loading the optimized parameter set into a virtual mesh nitrogen storage and oxygen control model, performing closed-loop simulation verification, generating a simulation optimization scheme that meets the requirements of oxygen control accuracy and energy consumption coordination, and generating execution control commands for the physical nitrogen storage and oxygen control system based on the simulation optimization scheme and transmitting them to the corresponding execution components, including: The optimized parameter set is fully loaded into the virtual grid nitrogen storage and oxygen control model, and the configuration parameters of each grid node in the virtual grid nitrogen storage and oxygen control model are updated. The configuration parameters include the virtual medium transmission coefficient, the virtual pressure regulation threshold, and the virtual energy consumption calculation coefficient. Set the operating parameters for closed-loop simulation verification. The operating parameters include simulation verification duration, data sampling interval, and verification operating condition type. The verification operating condition type includes normal operating condition, peak load operating condition, low load operating condition, sudden disturbance operating condition, and long-term operating condition. Initiate closed-loop simulation verification of the virtual grid nitrogen storage and oxygen control model, execute simulation runs under different verification conditions, and collect grid node status data and overall system performance data in real time during the simulation process; The collected simulation data is analyzed to determine whether the oxygen control accuracy of the model meets the preset oxygen control accuracy requirements and whether the energy consumption level meets the preset energy consumption requirements under different verification conditions. If the oxygen control accuracy or energy consumption level does not meet the requirements under any verification condition, return to the parameter adaptive optimization rule system design step, adjust the parameter adjustment step size, multi-objective constraint conditions or parameter interaction effect compensation mechanism in the optimization rule system, regenerate the optimization parameter set and perform closed-loop simulation verification. If the oxygen control accuracy and energy consumption level meet the requirements under all verification conditions, then stop the closed-loop simulation verification, and organize all data in the simulation process, including optimized parameter values, oxygen control accuracy data, energy consumption data, pressure stability data, and anti-disturbance capability data under each condition. Based on the processed simulation data, a simulation optimization scheme is generated. The simulation optimization scheme includes an optimization parameter configuration table, operation control strategies under various working conditions, emergency plans for parameter adjustment, and working condition adaptation suggestions. The control parameters corresponding to the physical nitrogen storage and oxygen control system are extracted from the simulation optimization scheme. The control parameters include nitrogen input rate adjustment parameters, oxygen concentration feedback adjustment parameters, pressure control parameters, energy consumption optimization parameters, and anti-disturbance adjustment parameters. Execution control instructions are generated based on the extracted control parameters. The execution control instructions include an instruction identifier, an execution component identifier, parameter setting values, execution time requirements, and instruction priority. Establish communication links with each execution component of the physical nitrogen storage and oxygen control system, transmit execution control commands to the corresponding execution components through a preset communication protocol, and receive command execution confirmation information returned by the execution components to confirm that the execution components have adjusted the operating parameters according to the execution control commands.

7. The simulation optimization method for a virtual grid nitrogen storage and oxygen control system combined with digital twins according to claim 2, characterized in that, The construction of the cross-domain data feature mapping model involves associating the oxygen concentration distribution data in the standardized physical data with the virtual medium transport coefficient in the standardized virtual data, and associating the nitrogen flow dynamic data in the standardized physical data with the virtual pressure regulation threshold in the standardized virtual data. This includes: Spatial interpolation is performed on the oxygen concentration distribution data in the standardized physical data to obtain the specific value of oxygen concentration at each spatial point in the physical system, forming an oxygen concentration spatial distribution matrix. The virtual medium transmission coefficients in the standardized virtual data are subjected to grid node matching processing, and the virtual medium transmission coefficients of each grid node are associated with the corresponding spatial points to form a virtual medium transmission coefficient space matrix. A correlation function between oxygen concentration and medium transport coefficient is constructed, using the values ​​in the spatial distribution matrix of oxygen concentration as input and the values ​​in the spatial matrix of virtual medium transport coefficient as output. The coefficients of the correlation function are determined by linear regression. Cross-validation was used to validate the correlation function. The oxygen concentration spatial distribution matrix was divided into a training set and a validation set. The correlation function was trained using the training set and the prediction accuracy of the correlation function was tested using the validation set. If the prediction accuracy does not reach the preset accuracy threshold, the type of the correlation function is adjusted, and the correlation function is reconstructed using a nonlinear regression method until the prediction accuracy reaches the preset accuracy threshold. Time series decomposition was performed on the nitrogen flow dynamic data in the standardized physical data to extract the trend term, periodic term and random term of nitrogen flow to form a nitrogen flow dynamic feature vector; Time response analysis was performed on the virtual pressure regulation threshold in the standardized virtual data to calculate the response change of the virtual pressure regulation threshold under different nitrogen flow inputs, and a pressure regulation response feature vector was formed. A correlation model between nitrogen flow rate and pressure regulation threshold is constructed. The dynamic feature vector of nitrogen flow rate and the feature vector of pressure regulation response are calculated to determine the correlation weight between the two. A vector mapping relationship is established based on the correlation weight, so that each element in the dynamic feature vector of nitrogen flow can correspond to a specific element in the pressure regulation response feature vector. By changing the dynamic data of nitrogen flow, the pressure regulation response feature vector is detected to see if it changes as expected according to the mapping relationship. If the expected change is not achieved, the correlation weights are adjusted and the mapping relationship is re-established.

8. The simulation optimization method for a virtual grid nitrogen storage and oxygen control system combined with digital twins according to claim 3, characterized in that, The step of performing data correlation analysis on the time-series interactive data to identify strong and weak correlation paths of state interactions between grid nodes, and recording the interaction frequency and interaction volume of strong correlation paths, includes: The media transmission data, pressure transmission data, temperature exchange data, and energy consumption sharing data in the time-series interactive data are classified according to grid node pairs to form an interactive data subset for each grid node pair; Statistical analysis is performed on the subset of interaction data for each grid node pair to calculate the total number of interactions, average interaction amount, maximum interaction amount, and interaction amount fluctuation coefficient during the simulation period. A correlation strength evaluation index is set, which includes an interaction frequency score, an interaction volume score, and an interaction stability score. The interaction frequency score, interaction volume score, and interaction stability score are calculated based on the corresponding statistical data according to a preset scoring standard. The overall score of the association strength of each grid node pair is calculated based on the association strength evaluation index. The overall score is obtained by adding the interaction frequency score, interaction quantity score and interaction stability score according to the preset weights. Grid node pairs whose overall association strength score exceeds the association strength threshold are identified as strongly associated node pairs, and the corresponding interaction paths are identified as strongly associated paths. Grid node pairs whose overall association strength score does not exceed the association strength threshold are identified as weakly associated node pairs, and the corresponding interaction paths are identified as weakly associated paths. Further analysis of strongly correlated paths was conducted, recording the interaction frequency of each strongly correlated path in different time intervals, and calculating the deviation rate between the interaction frequency and the average interaction frequency in each time interval. Record the interaction volume of each strongly associated path in different time intervals, and calculate the deviation rate between the interaction volume in each time interval and the average interaction volume; The identification information of strongly correlated paths, the interaction frequency and deviation rate of each time interval, and the interaction volume and deviation rate of each time interval are compiled into a strongly correlated path interaction report. Briefly record the weakly related paths, including the weakly related node pair identifiers, the total number of interactions, and the average number of interactions, to form a summary table of weakly related paths.

9. A simulation and optimization system for a virtual grid nitrogen storage and oxygen control system combining digital twins, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the simulation optimization method for a virtual grid nitrogen storage and oxygen control system incorporating digital twins as described in any one of claims 1 to 8 by executing the machine-executable instructions.

10. A computer program product, characterized in that, The computer program product includes machine-executable instructions stored in a computer-readable storage medium. The processor of the virtual grid nitrogen storage and oxygen control system simulation optimization system incorporating digital twins reads the machine-executable instructions from the computer-readable storage medium and executes the machine-executable instructions, causing the virtual grid nitrogen storage and oxygen control system simulation optimization system incorporating digital twins to perform the virtual grid nitrogen storage and oxygen control system simulation optimization method incorporating digital twins as described in any one of claims 1 to 8.

Images