Natural gas hydrogen doping proportion dynamic control and safety evaluation system and method

Abstract

The invention discloses a natural gas hydrogen doping proportion dynamic control and safety evaluation system and method, and belongs to the technical field of clean energy transmission and distribution and safety control, and the method comprises the steps: collecting constraint data and real-time multi-modal sensing data, and carrying out the fusion processing to generate a fusion initial data set; performing domain adaptation initialization and parameter optimization to generate an initial symbiosis decision framework; performing state synchronization and risk path exploration to generate a predictive state space; executing hierarchical reinforcement learning self-evolution training to generate an evolution control strategy group; calculating a real difference residual error as a feedback signal, and generating a self-healing optimization model; and performing multi-dimensional risk quantification and dynamic proportion regulation and control, and outputting a safety assessment report. According to the method, a self-evolution symbiotic architecture of multi-mode digital twinning and hierarchical reinforcement learning is adopted, dynamic optimization of the hydrogen doping proportion and real-time assessment of risks can be achieved, and the control stability, safety and reliability of a natural gas pipeline network under complex working conditions are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of clean energy transmission and distribution and safety control, and in particular to a system and method for dynamic control and safety assessment of the hydrogen blending ratio of natural gas. Background Art

[0002] With the advancement of the clean energy transition, natural gas hydrogen blending technology, as an emerging energy utilization method, has been widely applied in scenarios such as urban pipeline transportation, industrial fuel supply, and distributed energy systems. Through computer systems, digital processing and algorithm optimization of the hydrogen blending ratio enable digital management and process control of energy transportation. This technology relies on multimodal data fusion, model building, and a computational framework to support real-time data processing and strategy generation of pipeline network operating status, playing a vital role in the energy infrastructure sector.

[0003] However, existing natural gas hydrogen blending control systems suffer from widespread technical flaws. Under complex operating conditions, data processing struggles to adapt to multi-source, heterogeneous inputs, leading to unstable control strategy generation. Traditional methods often rely on fixed-parameter models, which are unable to effectively adapt to real-time operating environments and affect the overall system's responsiveness. Furthermore, safety assessments often overlook the computational integration of multi-dimensional risks, limiting the flexibility and reliability of proportional control. These issues hinder the system's performance in practical applications. Summary of the Invention

[0004] To solve the above problems, the present invention provides a dynamic control and safety assessment system and method for the hydrogen blending ratio of natural gas. It adopts a self-evolving symbiotic architecture of multimodal digital twins and hierarchical reinforcement learning, which can realize dynamic optimization of the hydrogen blending ratio and real-time assessment of risks, significantly improving the control stability and safety reliability of the natural gas pipeline network under complex working conditions.

[0005] The above objectives can be achieved through the following solutions: A natural gas hydrogen blending ratio dynamic control and safety assessment system includes a data acquisition and fusion module for collecting constraint data of the physical characteristics of the natural gas pipeline network and multimodal sensor data of the real-time operating status of the natural gas pipeline network, and fusing the constraint data and the multimodal sensor data to generate a fused initial data set; a model construction module for performing domain adaptation initialization and parameter optimization based on the fused initial data set, building a multimodal digital twin model and a hierarchical reinforcement learning network, and generating an initial symbiotic decision framework; a state synchronization exploration module for performing state synchronization and risk path exploration based on the initial symbiotic decision framework Explore and generate a predictive state space; a strategy evolution module is used to perform hierarchical reinforcement learning self-evolution training based on the predictive state space, evolve the optimal strategy through a virtual trial and error mechanism, and generate an evolutionary control strategy group; a self-healing optimization module is used to calculate the reality gap residual as a meta-level feedback signal based on the evolutionary control strategy group, reversely adjust the multimodal digital twin model and the hierarchical reinforcement learning network, and generate a self-healing optimization model; a risk control output module is used to perform multi-dimensional risk quantification and dynamic proportion control based on the self-healing optimization model, output a safety assessment report and perform adaptive control of the hydrogen blending ratio.

[0006] Optionally, the data acquisition and fusion module includes: a data acquisition unit, used to collect the constraint data of the physical characteristics of the natural gas pipeline network and the multimodal sensor data of the real-time operation status of the natural gas pipeline network to generate an initial data group; an adaptive synchronization unit, used to perform edge computing adaptive data synchronization based on the initial data group, dynamically adjust the synchronization frequency to match data heterogeneity, and generate a synchronization data group; a fusion optimization unit, used to perform variational autoencoder fusion optimization based on the synchronization data group, quantify uncertainty and refine features, and generate a fusion feature group; an output generation unit, used to perform standardization processing based on the fusion feature group to generate a fusion initial data set.

[0007] Optionally, the model construction module includes: a data input unit, used for fusing the initial data set, performing data preprocessing and feature mapping, and generating a preprocessed data set; a domain adaptation initialization unit, used for the preprocessed data set, performing variational inference domain adaptation initialization, quantifying data uncertainty and adapting to changes in working conditions, and generating an initialized model parameter group; a parameter optimization unit, used for the initialized model parameter group, performing multi-objective search optimization parameters, constructing the multimodal digital twin model and the hierarchical reinforcement learning network, and generating an optimized network structure; a framework output unit, used for the optimized network structure, performing integrated verification and framework assembly, and generating an initial symbiotic decision framework.

[0008] Optionally, the state synchronization exploration module includes: a state synchronization unit, which is used to perform adaptive state alignment optimization based on the initial symbiotic decision framework, dynamically calibrate multi-source operating state matching risk scenarios, and generate a synchronized state group; a risk exploration unit, which is used for the synchronized state group to perform probabilistic risk path mining, simulate evolution trajectories and quantify boundary uncertainties, and generate a predictive state space.

[0009] Optionally, the system also includes: based on the synchronization data group and the synchronization state group, performing heterogeneous data state coupling analysis, fusing multimodal data characteristics and operating state associations, and generating a preliminary calibration parameter group; based on the preliminary calibration parameter group, performing dynamic calibration optimization, iteratively adjusting the data state matching weight and verifying the calibration stability, and generating an optimized calibration parameter group; based on the optimized calibration parameter group, performing comprehensive indicator integration, quantifying the impact of multi-source consistency, and generating a comprehensive dynamic calibration indicator.

[0010] Optionally, the strategy evolution module includes: a state input unit, used to perform initial state mapping and calibration injection on the predictive state space and the comprehensive dynamic calibration index to generate a calibration state space; a cross-layer distillation unit, used for the calibration state space, to perform cross-layer knowledge distillation, to transfer the macro safety goal to the hydrogen blending ratio action, and to generate a distillation strategy prototype; a feedback coupling unit, used for dynamic feedback coupling of the distillation strategy prototype and the optimized network structure, to optimize the evolution path through virtual trial and error injection into the network structure, and to generate a coupled evolution strategy group; an output refining unit, used for the coupled evolution strategy group, to perform strategy verification and fusion to generate an evolution control strategy group.

[0011] Optionally, the system also includes: based on the synchronization data group and the coupling evolution strategy group, performing multi-dimensional coupling analysis of data strategies, integrating data synchronization characteristics with strategy evolution associations, and generating a preliminary coupling parameter group; based on the preliminary coupling parameter group, performing adaptive coupling optimization, iteratively adjusting matching weights and verifying stability, and generating an optimized coupling parameter group; based on the optimized coupling parameter group, performing comprehensive indicator integration, quantifying the impact of synchronization strategy consistency, and feeding back to the strategy evolution module.

[0012] Optionally, the self-healing optimization module includes: a residual calculation unit, which is used for the evolution control strategy group to perform multi-source residual pre-aggregation calculation, fuse multi-modal gap signals and filter noise interference to generate an initial residual signal group; a scale decomposition unit, which is used for the initial residual signal group and the coupled evolution strategy group to perform multi-scale residual decomposition hierarchy to generate a decomposed residual group; a reconstruction output unit, which is used for the decomposed residual group and the initialization model parameter group to perform adaptive network reconstruction, reversely adjust the multimodal digital twin model and the hierarchical reinforcement learning network and perform verification fusion to generate a self-healing optimization model.

[0013] Optionally, the risk control output module includes: a risk quantification unit, which is used for the self-healing optimization model to perform dynamic risk vector mapping, quantify multi-dimensional risk associations and calculate probability boundaries, and generate a risk quantification indicator group; a proportion control unit, which is used for the risk quantification indicator group to perform adaptive proportion control, dynamically adjust the hydrogen blending ratio and output a safety assessment report.

[0014] Based on the same inventive concept, the present invention also provides a method for dynamic control and safety assessment of the hydrogen blending ratio of natural gas, which includes: collecting constraint data of the physical characteristics of the natural gas pipeline network and multimodal sensor data of the real-time operating status of the natural gas pipeline network, and fusing the constraint data and the multimodal sensor data to generate a fused initial data set; based on the fused initial data set, performing domain adaptation initialization and parameter optimization, constructing a multimodal digital twin model and a hierarchical reinforcement learning network, and generating an initial symbiotic decision framework; based on the initial symbiotic decision framework, performing state synchronization and risk path exploration to generate a predictive state space; based on the predictive state space, executing hierarchical reinforcement learning self-evolution training, evolving the optimal strategy through a virtual trial and error mechanism, and generating an evolutionary control strategy group; based on the evolutionary control strategy group, calculating the actual gap residual as a meta-level feedback signal, reversely adjusting the multimodal digital twin model and the hierarchical reinforcement learning network, and generating a self-healing optimization model; based on the self-healing optimization model, performing multi-dimensional risk quantification and dynamic ratio regulation, outputting a safety assessment report and performing adaptive control of the hydrogen blending ratio.

[0015] Compared with the prior art, the present invention has the following advantages: 1. This invention achieves dynamic optimization control of the hydrogen blending ratio in natural gas through a self-evolving symbiotic architecture combining multimodal digital twins and hierarchical reinforcement learning. Compared to traditional fixed-parameter models, this system dynamically adapts to complex operating conditions based on real-time multimodal data fusion and virtual trial-and-error mechanisms, significantly improving the flexibility and stability of the control process, ensuring the optimal hydrogen blending ratio under pressure fluctuations, and optimizing the response efficiency of computer-aided control. 2. This invention utilizes a multi-dimensional risk quantification coupled with dynamic feedback to enhance the computational accuracy and real-time performance of safety assessments. Compared to existing single-risk assessment methods, this system, through data strategy coupling and self-healing optimization calculations, comprehensively analyzes multi-source risk correlations and dynamically adjusts proportional control strategies, significantly improving the safety management capabilities of the pipeline network system and avoiding the neglect of hidden risks often encountered in traditional methods.

[0016] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a framework diagram of a natural gas hydrogen blending ratio dynamic control and safety assessment system according to an embodiment of the present invention.

[0019] Figure 2 It is a structural schematic diagram of a natural gas hydrogen blending ratio dynamic control and safety assessment system according to an embodiment of the present invention.

[0020] Figure 3 It is a latent space distribution diagram of multimodal data fusion in an embodiment of the present invention.

[0021] Figure 4 3. It is a Pareto front diagram of multi-objective search optimization according to an embodiment of the present invention.

[0022] Figure 5 Schematic diagram of the dynamic hydrogen blending ratio control process within the safety envelope of an embodiment of the present invention. DETAILED DESCRIPTION

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0024] Reference Figure 1 One embodiment of the present invention proposes a system and method for dynamic control and safety assessment of the hydrogen blending ratio in natural gas. By adopting a self-evolving symbiotic architecture of multimodal digital twins and hierarchical reinforcement learning, the system can achieve dynamic optimization of the hydrogen blending ratio and real-time risk assessment, significantly improving the control stability and safety reliability of the natural gas pipeline network under complex working conditions.

[0025] like Figure 2As shown, the system of this embodiment specifically includes: A data acquisition and fusion module is used to collect constraint data of the physical characteristics of the natural gas pipeline network and multimodal sensor data of the real-time operating status of the natural gas pipeline network, and fuse the constraint data and the multimodal sensor data to generate a fused initial data set; A model building module is used to perform domain adaptation initialization and parameter optimization based on the fused initial data set, build a multimodal digital twin model and a hierarchical reinforcement learning network, and generate an initial symbiotic decision framework; A state synchronization exploration module, configured to perform state synchronization and risk path exploration based on the initial symbiotic decision framework to generate a predictive state space; A strategy evolution module is used to perform hierarchical reinforcement learning self-evolution training based on the predictive state space, evolve the optimal strategy through a virtual trial and error mechanism, and generate an evolutionary control strategy group; A self-healing optimization module, configured to calculate the reality gap residual as a meta-level feedback signal based on the evolutionary control strategy group, reversely adjust the multimodal digital twin model and the hierarchical reinforcement learning network, and generate a self-healing optimization model; The risk control output module is used to perform multi-dimensional risk quantification and dynamic ratio control based on the self-healing optimization model, output a safety assessment report and perform adaptive control of the hydrogen blending ratio.

[0026] The present invention adopts a self-evolving symbiotic architecture of multimodal digital twins and hierarchical reinforcement learning, which can achieve dynamic optimization of hydrogen blending ratio and real-time assessment of risks, significantly improving the control stability and safety and reliability of the natural gas pipeline network under complex working conditions.

[0027] Optionally, the data acquisition and fusion module includes: a data acquisition unit, configured to acquire the constraint data of the physical characteristics of the natural gas pipeline network and the multimodal sensing data of the real-time operating status of the natural gas pipeline network to generate an initial data set; Specifically, this step aims to provide all the raw information required for decision-making and analysis. Constraint data refers to engineering data describing the static physical properties of the pipeline network, such as pipeline topology, diameter, wall thickness, and the rated yield strength determined by the material grade. Multimodal sensor data refers to real-time measurements describing the dynamic operating status of the pipeline network. These include time series data collected by pressure sensors, temperature sensors, turbine flowmeters, electrochemical hydrogen concentration sensors, and acoustic emission (AE) sensors deployed along the pipeline for monitoring microcrack development. The data acquisition unit aggregates these two types of data to form an initial data set.

[0028] An adaptive synchronization unit, configured to perform edge computing adaptive data synchronization based on the initial data group, dynamically adjust the synchronization frequency to match the data heterogeneity, and generate a synchronized data group; Specifically, this step aims to solve the time alignment problem caused by the diverse sources, sampling frequencies and importance of multimodal sensor data. This unit is deployed in the edge computing gateway close to the sensor and implements an adaptive data synchronization strategy. This strategy dynamically adjusts the synchronization and upload frequency of data packets according to the information change rate of each data stream. For example, a method for determining the synchronization frequency The calculation of can be defined by the following formula: , in, is a basic synchronization frequency, It is The normalized signal value of each sensor data stream, is the rate of change of its signal, is the preset importance weight of the data stream, is a sensitivity adjustment coefficient. When the pipeline network operates smoothly, the rate of change of each signal is low, and the synchronization frequency remains at a basic level. However, when transient events such as pressure fluctuations occur, the rate of change surges, and the synchronization frequency increases dynamically. This greatly optimizes the efficiency of data transmission and processing while ensuring that critical information is not lost, ultimately generating synchronized data sets that are precisely aligned in time.

[0029] A fusion optimization unit, configured to perform variational autoencoder fusion optimization based on the synchronized data group, quantify uncertainty and refine features, and generate a fusion feature group; Specifically, this step aims to fuse and compress heterogeneous, multimodal synchronized data sets into a unified, low-dimensional feature space. In this process, a variational autoencoder (VAE) is used for fusion optimization. The encoder part of the network receives the synchronized data set as input and maps it into a probability distribution in the latent space. The mean vector of the probability distribution is extracted as a highly refined, information-concentrated fusion feature set that can characterize the current state of the pipeline network; and the variance vector of the probability distribution is used as a direct quantification of the "uncertainty" of the current data fusion result, which can be used by downstream modules, such as Figure 3 As shown in the figure, the two-dimensional latent space distribution learned by the variational autoencoder is displayed in the form of a scatter plot, where the marked points of different shapes represent the fusion feature groups under different pipeline network operating conditions. It can be seen that the fusion optimization unit successfully separates different states effectively, and the ellipse around each point intuitively represents the uncertainty quantification result of the data point.

[0030] The output generation unit is used to perform standardization processing based on the fused feature group to generate a fused initial data set.

[0031] Specifically, this step aims to provide a standardized, dimensionally uniform input for the subsequent model building modules. During this process, the mean and standard deviation of each feature dimension in the fused feature set are calculated over a specific historical time window. Z-score normalization is then performed by subtracting the corresponding mean from each feature value and dividing it by its standard deviation. This step ultimately generates a fused initial dataset, whose features in each dimension follow a standard normal distribution, as the final output.

[0032] Optionally, the model building module includes: A data input unit, used for fusing the initial data set, performing data preprocessing and feature mapping, and generating a preprocessed data set; Specifically, this unit serves as an interface to the model building module, responsible for receiving the fused initial dataset generated by the data acquisition and fusion module. Within this unit, final data cleaning and transformation operations are performed on this dataset, such as detecting and removing potential anomalous data points through algorithms such as isolation forests. Furthermore, to improve the efficiency of subsequent neural network training, this unit also performs feature mapping, projecting the input data into a feature space that is more easily learned by the model, ultimately generating a preprocessed dataset.

[0033] A domain adaptation initialization unit, used for the preprocessed data set, performs variational inference domain adaptation initialization, quantifies data uncertainty and adapts to working condition changes, and generates an initialization model parameter group; Specifically, this unit aims to provide a high-quality, information-rich "starting point" for subsequent parameter optimization, rather than traditional random initialization. In this process, a variational inference domain adaptation initialization method is adopted. This method does not directly solve a fixed set of initial model parameters, but instead learns a posterior probability distribution that can characterize the model parameters by training an inference network. By training on a preprocessed data set containing a variety of historical working conditions, the posterior probability distribution itself contains the ability to adapt to different working conditions. At the same time, the variance of the distribution directly quantifies the parameter uncertainty caused by data noise or incompleteness. Finally, the set of initialized model parameters output by the unit is the set of hyperparameters that describe the posterior probability distribution of the parameters.

[0034] A parameter optimization unit, configured to initialize the model parameter group, perform multi-objective search and optimization of the parameters, construct the multimodal digital twin model and the hierarchical reinforcement learning network, and generate an optimized network structure; Specifically, this unit aims to find a specific parameter instance with the best comprehensive performance in multiple dimensions from the parameter probability distribution obtained in the previous step. In this process, a multi-objective search algorithm is executed, such as a search strategy based on an evolutionary algorithm. The algorithm samples from the probability distribution defined by the initialization model parameter group to generate a large number of candidate network structures and parameter configurations. Each candidate configuration will be evaluated in a virtual environment based on multiple conflicting performance objectives, such as: the simulation accuracy of the multimodal digital twin model, the initial learning efficiency of the hierarchical reinforcement learning network, and the computational complexity of the overall model. Through multiple generations of iterative optimization, the unit finally finds a network configuration that is on the "Pareto frontier" and achieves the best balance between the above-mentioned multiple objectives as the optimized network structure, such as Figure 4 As shown in the figure, the performance distribution of a large number of candidate network structures in the multi-objective search optimization process is displayed in the form of a scatter plot, where the horizontal axis represents the model calculation complexity and the vertical axis represents the model simulation accuracy. The figure clearly shows the Pareto front curve that achieves the optimal trade-off between the two conflicting objectives, as well as the optimal network structure solution that is finally selected to build the symbiotic decision-making framework.

[0035] The framework output unit is used for optimizing the network structure, performing integrated verification and framework assembly, and generating an initial symbiotic decision-making framework.

[0036] Specifically, this unit is the final step in model construction, aiming to assemble and verify two independently optimized network models into a collaborative, organic whole. During this process, the multimodal digital twin model and hierarchical reinforcement learning network defined in the optimized network structure are instantiated. This unit performs an integrated verification, placing the two models in a simulation environment for joint debugging to ensure the correctness and stability of their interface calls, data transmission, and interaction logic. After verification, the unit performs the final framework assembly of the two models, encapsulating them into a unified, initial symbiotic decision-making framework that can be called by subsequent modules and serves as the final output.

[0037] Optionally, the state synchronization exploration module includes: A state synchronization unit is used to perform adaptive state alignment optimization based on the initial symbiotic decision framework, dynamically calibrate multi-source operating state matching risk scenarios, and generate a synchronized state group; Specifically, this unit is designed to ensure that the internal state of the multimodal digital twin model can accurately and in real time reflect the real health status of the physical pipeline network. During this process, the unit continuously inputs the multimodal sensor data collected in real time into the multimodal digital twin model in the initial symbiotic decision-making framework, and compares the deviation between the simulation output of the model and the real sensor data to form a state alignment error. Through an online optimization algorithm, such as a particle swarm optimization algorithm, with the goal of minimizing the state alignment error, iteratively adjusts the internal state parameters that cannot be directly measured in the multimodal digital twin model, such as the virtual fatigue damage degree or hydrogen embrittlement sensitivity coefficient of the pipeline. After dynamic calibration, this group of internal parameters that can represent the current real state of the physical pipeline network constitutes a synchronized state group.

[0038] The risk exploration unit is used for the synchronous state group to perform probabilistic risk path mining, simulate evolution trajectory and quantify boundary uncertainty to generate a predictive state space.

[0039] Specifically, this unit aims to use a multimodal digital twin model that is synchronized with reality to conduct a forward-looking, probabilistic exploration of possible future failure paths. During this process, the unit uses the synchronized state group as the initial condition and performs thousands of Monte Carlo accelerated simulations. In each simulation, random perturbations that conform to the uncertainty distribution are applied to the key parameters in the model, thereby simulating a unique future evolution trajectory. In order to accurately quantify potential extreme risks, the unit further uses the Conditional Value at Risk (CVaR) method to assess boundary uncertainty, which can be calculated as follows: , in, is a loss function that represents the state close to failure. At a confidence level of The risk value at By calculating the conditional value at risk, the unit can accurately assess the worst case scenario. Finally, all simulated evolutionary trajectories are combined with the calculated boundary uncertainty indicators to construct a predictive state space that includes both future trends and extreme risk quantification as the final output.

[0040] Optionally, the system further comprises: Based on the synchronization data group and the synchronization state group, performing heterogeneous data state coupling analysis, fusing multimodal data characteristics and operating state associations, and generating a preliminary calibration parameter group; Specifically, this step aims to construct a parallel verification path for assessing self-consistency—that is, verifying the consistency between the underlying sensor data and the higher-level twin state. During this process, a heterogeneous data-state coupling analysis is performed. This analysis quantifies the strength of the correlation between the underlying statistical features in the synchronized data set and the higher-level states inferred by the digital twin model in the synchronized state set by calculating the mutual information or correlation coefficient between the two. All these quantified correlation strength metrics together constitute the initial calibration parameter set.

[0041] Based on the preliminary calibration parameter set, dynamic calibration optimization is performed, data state matching weights are iteratively adjusted, and calibration stability is verified to generate an optimized calibration parameter set; Specifically, this step aims to refine the preliminary coupling correlation indicators to eliminate spurious correlations caused by random noise or transient disturbances. During this process, a sliding time window-based verification mechanism is used to check the stability of each parameter in the preliminary calibration parameter set. Only those parameters that show stable, sustained strong correlations across multiple consecutive time windows will have their matching weights retained or enhanced; while the weights of weak correlations that appear intermittently and fluctuate violently are suppressed or eliminated. Through this iterative verification and adjustment, an optimized calibration parameter set is ultimately generated that more robustly reflects the true correlation between data and state.

[0042] Based on the optimized calibration parameter set, comprehensive index integration is performed to quantify the impact of multi-source consistency and generate a comprehensive dynamic calibration index.

[0043] Specifically, this step aims to integrate the optimized, multi-dimensional calibration parameters obtained in the previous step into a single, comprehensive indicator that can intuitively reflect the current overall self-consistency. In this process, all parameters in the optimized calibration parameter group can be fused into a standardized single value through a pre-trained lightweight neural network or a weighted sum model. This value is the comprehensive dynamic calibration index. A higher index value indicates that the underlying sensor data is highly consistent with the high-level model state and the cognition is reliable; conversely, a lower index value indicates that there is a deviation between the two, which may be due to sensor anomalies or model mismatches, requiring downstream modules to make more prudent decisions.

[0044] Optionally, the strategy evolution module includes: A state input unit, configured to perform initial state mapping and calibration injection on the predictive state space and the comprehensive dynamic calibration index to generate a calibration state space; Specifically, this unit is designed to provide an information-rich and confidence-calibrated input for the hierarchical reinforcement learning network. In this process, "initial state mapping" refers to mapping the complex high-dimensional probabilistic data in the predictive state space into a fixed-length feature vector that can represent the current state of the environment through an encoder network. "Calibration injection" is a key creative step that uses the comprehensive dynamic calibration indicator as an additional input channel and performs splicing or gated fusion with the state feature vector. In this way, "self-consistency" or "confidence level" is directly injected into the state representation, allowing subsequent reinforcement learning agents to perceive the quality of information on which the current decision is based, thereby generating a calibrated state space.

[0045] A cross-layer distillation unit is used for the calibration state space, performs cross-layer knowledge distillation, transfers the macro safety goal to the hydrogen doping ratio action, and generates a distillation strategy prototype; Specifically, this unit is the core training link of the hierarchical reinforcement learning network, which aims to efficiently transform abstract macro goals into specific micro actions. In this process, a cross-layer knowledge distillation technology is adopted. This technology includes a "teacher network" responsible for evaluating macro safety goals, and a "student network" responsible for outputting specific hydrogen doping ratio actions. The training of the "student network" not only relies on traditional reinforcement learning rewards, but also learns to fit the soft label probability distribution output by the "teacher network" through a knowledge distillation loss function. The total loss function of the student network is It can be expressed as: , in, is the standard reinforcement learning loss function, It is used to measure the output distribution of the teacher network Output distribution of the student network The knowledge distillation loss of the difference between is a hyperparameter that balances the two losses. By minimizing this total loss function, macroscopic security knowledge is efficiently “distilled” and transferred into specific action strategies, generating a distilled strategy prototype.

[0046] A feedback coupling unit is used to perform dynamic feedback coupling between the distillation strategy prototype and the optimized network structure, inject virtual trial and error into the network structure to optimize the evolution path, and generate a coupled evolution strategy group; Specifically, this unit aims to deeply optimize strategies through virtual trial and error, creatively introducing the evolution of the network structure itself. During this process, the distilled strategy prototype serves as the initial strategy, performing extensive trial-and-error exploration within the virtual environment provided by the multimodal digital twin model. Furthermore, "dynamic feedback coupling" is reflected in not only optimizing the weight parameters of the strategy network, but also fine-tuning the optimized network structure itself based on specific challenging scenarios encountered during virtual trial and error. For example, through Neural Architecture Search (NAS) technology, neuronal connections are dynamically added or pruned to optimize its ability to model specific evolutionary paths, thereby generating a coupled evolutionary strategy group.

[0047] The output refining unit is used for the coupled evolution strategy group to perform strategy verification and fusion to generate an evolution control strategy group.

[0048] Specifically, this unit is the final step in strategy evolution, ensuring the robustness and stability of the output strategy. During this process, "strategy verification" involves testing the optimal strategy from the coupled evolution strategy group under a series of preset, rigorous virtual boundary conditions to verify its safety and generalization capabilities. "Fusion" is an optional enhancement step that integrates multiple high-performing strategies that emerged during the evolution process, for example through weighted averaging or voting mechanisms, to form a final strategy with enhanced overall performance and greater stability. This final, verified and fused strategy becomes the evolutionary control strategy group.

[0049] Optionally, the system further comprises: Based on the synchronization data group and the coupling evolution strategy group, perform data strategy multi-dimensional coupling analysis, integrate data synchronization characteristics and strategy evolution association, and generate a preliminary coupling parameter group; Specifically, this step aims to build an analysis module to assess the interplay between data quality and policy behavior. During this process, a multidimensional data-policy coupling analysis is performed. This analysis aims to answer a core question: whether the performance of the control policy learned by the policy evolution module fluctuates dramatically due to minor changes in the input data characteristics. This analysis quantifies this potential coupling relationship by calculating a multidimensional correlation matrix between the statistical properties of the synchronized data set and the performance indicators of the coupled evolved policy set in the virtual environment. This correlation matrix serves as the preliminary coupling parameter set.

[0050] Based on the preliminary coupling parameter set, adaptive coupling optimization is performed, matching weights are iteratively adjusted and stability is verified, and an optimized coupling parameter set is generated; Specifically, this step aims to refine the initially quantified coupling relationships to ensure their stability and reliability. During this process, a time-series cross-validation approach is employed to repeatedly perform multi-dimensional data-strategy coupling analysis across different historical data time windows. Only those coupling relationships that are consistently reproducible and statistically significant across multiple time windows are confirmed and strengthened in terms of matching weights. This iterative validation and optimization process filters out accidental spurious couplings caused by specific data fragments, resulting in an optimized set of coupling parameters that more consistently and fundamentally reflects the relationship between data characteristics and strategy performance.

[0051] Based on the optimized coupling parameter group, comprehensive indicator integration is performed to quantify the impact of synchronization strategy consistency and feed back to the strategy evolution module.

[0052] Specifically, this step is to achieve a closed-loop link for another higher-dimensional feedback loop. In this process, the optimized coupling parameter group is integrated into one or more feedback adjustment signals and fed back to the policy evolution module. The feedback signal is used to dynamically and adaptively adjust the learning process of the policy evolution module itself. For example, when the optimized coupling parameter group shows that the current control strategy is highly sensitive to the noise of the input data, the feedback signal will instruct the cross-layer distillation unit in the policy evolution module to increase the weight of the regularization term in its loss function to suppress overfitting, or instruct the feedback coupling unit to increase the proportion of noisy training samples in virtual trial and error. In this way, direct feedback from data quality to the policy learning process is achieved, so that the evolution of the strategy can actively adapt to the current data environment.

[0053] Optionally, the self-healing optimization module includes: A residual calculation unit is used for the evolutionary control strategy group to perform multi-source residual pre-aggregation calculation, fuse multi-modal difference signals and filter noise interference to generate an initial residual signal group; Specifically, this unit aims to accurately quantify the deviation between the "virtual world" and the "physical world" of the symbiotic decision-making framework, namely the "reality gap." In this process, the unit first uses the evolutionary control strategy group to drive the multimodal digital twin model for simulation, obtaining a set of simulation outputs corresponding to the multimodal sensor data. Subsequently, this simulation output is compared one by one with the synchronized, real multimodal sensor data, and the difference between the two is calculated to form a multi-dimensional residual vector. The unit further filters this residual vector to remove the interference of random measurement noise, ultimately generating an initial residual signal group that can truly reflect the model deviation.

[0054] A scale decomposition unit, configured to perform a multi-scale residual decomposition hierarchy on the initial residual signal group and the coupled evolution strategy group to generate a decomposed residual group; Specifically, this unit aims to perform an in-depth analysis of the initial residual signal group to determine whether its root cause is short-term strategic deviation or long-term model drift. In this process, a multi-scale decomposition technique, such as the Continuous Wavelet Transform (CWT), is used to decompose the initial residual signal group. The calculation of the continuous wavelet transform can be defined by the following formula: , in, is the time series of the initial residual signal group, is the mother wavelet function, is the scale factor, is the translation factor, Denotes conjugate. By using different scale factors By transforming the residuals on the time scale, this unit can decompose the total residual into different time scales. For example, residuals associated with high-frequency, transient behavior will be concentrated on smaller scales, while residuals associated with slow, drifting behavior will be concentrated on larger scales. All these residual components at different scales together constitute the decomposed residual group.

[0055] A reconstruction output unit is used for the decomposition residual group and the initialization model parameter group to perform adaptive network reconstruction, reversely adjust the multimodal digital twin model and the hierarchical reinforcement learning network and perform verification fusion to generate a self-healing optimization model.

[0056] Specifically, this unit is the core of the "self-healing" operation, and its purpose is to intelligently and differentially feed back the residuals of different scales decomposed in the previous step to the corresponding models. In this process, the unit performs an adaptive network reconstruction. Its basic principle is: the residual components concentrated on smaller time scales obtained by the scale decomposition unit are mainly attributed to the short-term policy deviation of the hierarchical reinforcement learning network, and are therefore used to generate correction signals for reverse adjustment of the hierarchical reinforcement learning network; while the residual components concentrated on larger time scales are mainly attributed to the physical mechanism mismatch or parameter drift of the multimodal digital twin model, and are therefore used to generate correction signals for reverse adjustment of the multimodal digital twin model. Through this "divide and conquer" feedback mechanism, accurate and efficient collaborative optimization of the two core models is achieved, and the final generated, doubly calibrated model set is the self-healing optimization model.

[0057] Optionally, the risk control output module includes: A risk quantification unit, used in the self-healing optimization model, performs dynamic risk vector mapping, quantifies multi-dimensional risk associations, calculates probability boundaries, and generates a risk quantification indicator group; Specifically, this unit is the final diagnostic and quantification step, aiming to transform the complex internal state of the self-healing optimization model into multi-dimensional risk insights that provide clear guidance to operations and maintenance personnel. During this process, the unit performs a dynamic risk vector mapping. This mapping process includes: first, extracting the final failure probability directly from the self-healing optimization model as the core "probabilistic risk" indicator; second, using a feature attribution algorithm to analyze and calculate the contribution of each input feature to this failure probability, forming a "causal risk" indicator; and third, analyzing the variance of the probability distribution output by the self-healing optimization model to form an "uncertainty risk" indicator. This unit integrates the risk indicators from these three dimensions to generate a set of risk quantification indicators.

[0058] The ratio control unit is used for the risk quantification indicator group to perform adaptive ratio control, dynamically adjust the hydrogen blending ratio and output a safety assessment report.

[0059] Specifically, this unit is the final decision-making and output link, responsible for generating two types of key outputs for people and equipment. On the one hand, the unit generates a visual, multi-level safety assessment report based on the risk quantification indicator group. On the other hand, the unit performs adaptive ratio control. In this process, it first receives the optimal hydrogen blending ratio recommendation value output by the hierarchical reinforcement learning network, and then uses a "safety guard strategy" to verify and smooth the recommendation value using the risk quantification indicator group. For example, when the "uncertainty risk" indicator is high, the guard strategy will moderately lower the recommendation value to increase the safety margin, and finally output a dynamic hydrogen blending ratio control instruction that is both intelligent and safe, such as Figure 5 As shown in the figure, the dynamic process of adaptive ratio control is demonstrated, in which the theoretical optimal hydrogen blending ratio generated by the hierarchical reinforcement learning network is smoothed and constrained by a safety control envelope that is dynamically tightened based on uncertainty when the risk is predicted to increase, and finally an actual controlled hydrogen blending ratio that is both efficient and absolutely safe is output.

[0060] Based on the same inventive concept, the present invention also provides a method for dynamic control and safety assessment of the hydrogen blending ratio of natural gas, the method comprising: Collecting constraint data of the physical characteristics of the natural gas pipeline network and multimodal sensor data of the real-time operating status of the natural gas pipeline network, and fusing the constraint data and the multimodal sensor data to generate a fused initial data set; Based on the fused initial dataset, domain adaptation initialization and parameter optimization are performed to construct a multimodal digital twin model and a hierarchical reinforcement learning network to generate an initial symbiotic decision framework; Based on the initial symbiotic decision framework, state synchronization and risk path exploration are performed to generate a predictive state space; Based on the predictive state space, performing hierarchical reinforcement learning self-evolution training, evolving the optimal strategy through a virtual trial and error mechanism, and generating an evolutionary control strategy group; Based on the evolutionary control strategy group, the reality gap residual is calculated as a meta-level feedback signal, and the multimodal digital twin model and the hierarchical reinforcement learning network are reversely adjusted to generate a self-healing optimization model; Based on the self-healing optimization model, multi-dimensional risk quantification and dynamic ratio control are carried out, a safety assessment report is output, and adaptive control of the hydrogen blending ratio is performed.

[0061] To demonstrate the feasibility and advancement of this invention, a regional hydrogen blending demonstration project was applied to a city-level natural gas pipeline network. This project aims to gradually increase the proportion of clean energy in the natural gas supply of a large industrial park and surrounding residential areas by injecting green hydrogen produced by photovoltaic electrolysis. The regional pipeline network, with its mixed pipe materials and complex topology, presents a key challenge: maximizing the hydrogen blending ratio while ensuring absolute safety, thereby achieving both economic and environmental benefits.

[0062] In this example, a six-month dynamic control and safety assessment of the demonstration area's pipeline network was conducted. Data was collected from multimodal sensors deployed along the pipeline, including pressure, flow, temperature, hydrogen concentration, and acoustic emissions. Constraints such as the network's GIS topology and material properties were also integrated.

[0063] First, the data acquisition and fusion module processes heterogeneous data. For example, during a transient upstream pressure fluctuation on January 20, 2025, the adaptive synchronization unit deployed at the edge gateway automatically increased the synchronization frequency of the pressure and acoustic emission sensors from the conventional 1Hz to 50Hz, generating a high-fidelity synchronized data set. The fusion optimization unit then transforms this data set into a fused initial dataset that includes uncertainty quantification using a variational autoencoder.

[0064] The model building module constructs a multimodal digital twin model and a hierarchical reinforcement learning network that include the physical mechanism of the pipeline network through the domain adaptation initialization unit and the parameter optimization unit, forming an initial symbiotic decision-making framework.

[0065] During real-time operation, the state synchronization exploration module continues to operate. The state synchronization unit aligns real-time sensor data with the digital twin model, dynamically calibrating implicit states such as the virtual hydrogen embrittlement damage within the pipeline. Based on this calibrated state, the risk exploration unit conducts probabilistic risk path mining, generating a predictive state space on February 5th. This space indicates that if the hydrogen blending ratio is increased from 8% to 15%, the failure probability of a certain old pipeline section will exceed the safety threshold within the next 72 hours.

[0066] At the same time, a comprehensive dynamic calibration index generated by a parallel verification module remained high at above 0.9 during this period, indicating that the underlying data was highly consistent with the high-level twin state and cognitively reliable.

[0067] The strategy evolution module receives the aforementioned predictive state space and high-confidence calibration metrics. The cross-layer distillation unit efficiently translates the macro-goal of "maintaining pipe safety margin" into a specific hydrogen blending ratio adjustment strategy. The feedback coupling unit conducts trial and error within the virtual environment provided by the digital twin, optimizing the network structure of the intelligent agent and ultimately generating an evolved control strategy set.

[0068] The self-healing optimization module demonstrated its powerful adaptive capabilities. In early April, seasonal temperature fluctuations caused minor drifts in the pipeline's physical properties, and the calculated "reality gap residual" began to slowly increase. The scale decomposition unit attributed this residual to long-term model drift. The reconstruction output unit then fine-tuned the material thermal expansion coefficients of the multimodal digital twin model, restoring high model accuracy and generating a self-healing optimization model.

[0069] Ultimately, the risk control output module produced highly intelligent results. The generated safety assessment report not only quantified the risks of each pipeline section but also, through dynamic risk vector mapping, identified the dominant risk factor as the localized hydrogen concentration downstream of valve 2. Based on this report and implementing a "safety protection strategy," the ratio control unit adaptively and smoothly lowered the hydrogen blending ratio from 12% to 10.5% during peak gas consumption periods, ensuring a safety margin.

[0070] Data demonstrates that this invention achieves performance far exceeding that of traditional fixed-ratio or passive response control methods. Through multi-level intelligent optimization and feedback, the average hydrogen blending ratio was increased from a conservative fixed value of 5% to a dynamically variable 11.8% over a six-month operation period, without triggering any safety alarms.

[0071] Table 1 Example of dynamic hydrogen blending ratio and pipeline network status monitoring data

[0072] Table 2 Comparison of safety assessment and dynamic control performance

[0073] Table 3 Example of risk event prediction and verification table

[0074] As can be seen from the data recorded in Tables 1-3 above, the present invention performs well in practical applications. Table 1 demonstrates the ability to dynamically adjust the hydrogen blending ratio based on operating conditions and data quality. Table 2, by comparison with traditional methods, clearly demonstrates that the present invention significantly improves the efficiency of hydrogen blending while also ensuring better safety. Table 3 verifies the core predictive safety assessment capability, and its deduction results for potential risks are strongly confirmed by virtual simulation, providing strong technical support for proactive and preventative pipeline safety management.

[0075] It should be noted that the formulas appearing above can translate physical quantities of different properties into unitless standard values ​​or superimposable parameters of the same dimension through the principle of dimensional consistency and mathematical standardization means (such as normalization, dimensionless parameter conversion or unit system unification), thereby eliminating the interference of different dimensions on the operation logic, so that the formulas have mathematical operation rationality and objective law adaptability while retaining the distribution characteristics of the original data. It is a conventional technical means and will not be elaborated here. The electrical connection between the above-mentioned units does not necessarily mean a direct connection of the circuit. The indirect connection method can be applied to the embodiments of the present invention as long as the purpose of the present invention is achieved. The above is only an exemplary embodiment of the present invention and the scope of the present invention cannot be limited thereto.

[0076] That is, any equivalent changes and modifications made according to the teachings of the present invention are still within the scope of the present invention. Those skilled in the art will readily conceive of other embodiments of the present invention after considering the disclosure of the specification and practical truths. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary technical means in the art not described herein.

Claims

1. A natural gas hydrogen blending ratio dynamic control and safety assessment system, characterized in that: The system comprises: A data acquisition and fusion module is used to collect constraint data of the physical characteristics of the natural gas pipeline network and multimodal sensor data of the real-time operating status of the natural gas pipeline network, and fuse the constraint data and the multimodal sensor data to generate a fused initial data set; A model building module is used to perform domain adaptation initialization and parameter optimization based on the fused initial data set, build a multimodal digital twin model and a hierarchical reinforcement learning network, and generate an initial symbiotic decision framework; A state synchronization exploration module, configured to perform state synchronization and risk path exploration based on the initial symbiotic decision framework to generate a predictive state space; A strategy evolution module is used to perform hierarchical reinforcement learning self-evolution training based on the predictive state space, evolve the optimal strategy through a virtual trial and error mechanism, and generate an evolutionary control strategy group; A self-healing optimization module, configured to calculate the reality gap residual as a meta-level feedback signal based on the evolutionary control strategy group, reversely adjust the multimodal digital twin model and the hierarchical reinforcement learning network, and generate a self-healing optimization model; The risk control output module is used to perform multi-dimensional risk quantification and dynamic ratio control based on the self-healing optimization model, output a safety assessment report and perform adaptive control of the hydrogen blending ratio.

2. A natural gas hydrogen blending ratio dynamic control and safety assessment system according to claim 1, characterized in that: The data acquisition and fusion module includes: a data acquisition unit, configured to acquire the constraint data of the physical characteristics of the natural gas pipeline network and the multimodal sensing data of the real-time operating status of the natural gas pipeline network to generate an initial data set; An adaptive synchronization unit, configured to perform edge computing adaptive data synchronization based on the initial data group, dynamically adjust the synchronization frequency to match the data heterogeneity, and generate a synchronized data group; A fusion optimization unit, configured to perform variational autoencoder fusion optimization based on the synchronized data group, quantify uncertainty and refine features, and generate a fusion feature group; The output generation unit is used to perform standardization processing based on the fused feature group to generate a fused initial data set.

3. A natural gas hydrogen blending ratio dynamic control and safety assessment system according to claim 2, characterized in that: The model building module includes: A data input unit, used for fusing the initial data set, performing data preprocessing and feature mapping, and generating a preprocessed data set; A domain adaptation initialization unit, used for the preprocessed data set, performs variational inference domain adaptation initialization, quantifies data uncertainty and adapts to working condition changes, and generates an initialization model parameter group; A parameter optimization unit, configured to initialize the model parameter group, perform multi-objective search and optimization of the parameters, construct the multimodal digital twin model and the hierarchical reinforcement learning network, and generate an optimized network structure; The framework output unit is used for optimizing the network structure, performing integrated verification and framework assembly, and generating an initial symbiotic decision-making framework.

4. A natural gas hydrogen blending ratio dynamic control and safety assessment system according to claim 3, characterized in that: The state synchronization exploration module includes: A state synchronization unit is used to perform adaptive state alignment optimization based on the initial symbiotic decision framework, dynamically calibrate multi-source operating state matching risk scenarios, and generate a synchronized state group; The risk exploration unit is used for the synchronous state group to perform probabilistic risk path mining, simulate evolution trajectory and quantify boundary uncertainty to generate a predictive state space.

5. A natural gas hydrogen blending ratio dynamic control and safety assessment system according to claim 4, characterized in that: The system further comprises: Based on the synchronization data group and the synchronization state group, performing heterogeneous data state coupling analysis, fusing multimodal data characteristics and operating state associations, and generating a preliminary calibration parameter group; Based on the preliminary calibration parameter set, dynamic calibration optimization is performed, data state matching weights are iteratively adjusted, and calibration stability is verified to generate an optimized calibration parameter set; Based on the optimized calibration parameter set, comprehensive index integration is performed to quantify the impact of multi-source consistency and generate a comprehensive dynamic calibration index.

6. A natural gas hydrogen blending ratio dynamic control and safety assessment system according to claim 5, characterized in that: The strategy evolution module includes: A state input unit, configured to perform initial state mapping and calibration injection on the predictive state space and the comprehensive dynamic calibration index to generate a calibration state space; A cross-layer distillation unit is used for the calibration state space, performs cross-layer knowledge distillation, transfers the macro safety goal to the hydrogen doping ratio action, and generates a distillation strategy prototype; A feedback coupling unit is used to perform dynamic feedback coupling between the distillation strategy prototype and the optimized network structure, inject virtual trial and error into the network structure to optimize the evolution path, and generate a coupled evolution strategy group; The output refining unit is used for the coupled evolution strategy group to perform strategy verification and fusion to generate an evolution control strategy group.

7. A natural gas hydrogen blending ratio dynamic control and safety assessment system according to claim 6, characterized in that: The system further comprises: Based on the synchronization data group and the coupling evolution strategy group, perform data strategy multi-dimensional coupling analysis, integrate data synchronization characteristics and strategy evolution association, and generate a preliminary coupling parameter group; Based on the preliminary coupling parameter set, adaptive coupling optimization is performed, matching weights are iteratively adjusted and stability is verified, and an optimized coupling parameter set is generated; Based on the optimized coupling parameter group, comprehensive indicator integration is performed to quantify the impact of synchronization strategy consistency and feed back to the strategy evolution module.

8. A natural gas hydrogen blending ratio dynamic control and safety assessment system according to claim 6, characterized in that: The self-healing optimization module includes: A residual calculation unit is used for the evolutionary control strategy group to perform multi-source residual pre-aggregation calculation, fuse multi-modal difference signals and filter noise interference to generate an initial residual signal group; A scale decomposition unit, configured to perform a multi-scale residual decomposition hierarchy on the initial residual signal group and the coupled evolution strategy group to generate a decomposed residual group; A reconstruction output unit is used for the decomposition residual group and the initialization model parameter group to perform adaptive network reconstruction, reversely adjust the multimodal digital twin model and the hierarchical reinforcement learning network and perform verification fusion to generate a self-healing optimization model.

9. A natural gas hydrogen blending ratio dynamic control and safety assessment system according to claim 1, characterized in that: The risk control output module includes: A risk quantification unit, used in the self-healing optimization model, performs dynamic risk vector mapping, quantifies multi-dimensional risk associations, calculates probability boundaries, and generates a risk quantification indicator group; The ratio control unit is used for the risk quantification indicator group to perform adaptive ratio control, dynamically adjust the hydrogen blending ratio and output a safety assessment report.

10. A method for dynamic control and safety assessment of hydrogen blending ratio in natural gas, applied to a system for dynamic control and safety assessment of hydrogen blending ratio in natural gas as claimed in any one of claims 1 to 9, characterized in that: The method comprises: Collecting constraint data of the physical characteristics of the natural gas pipeline network and multimodal sensor data of the real-time operating status of the natural gas pipeline network, and fusing the constraint data and the multimodal sensor data to generate a fused initial data set; Based on the fused initial dataset, domain adaptation initialization and parameter optimization are performed to construct a multimodal digital twin model and a hierarchical reinforcement learning network to generate an initial symbiotic decision framework; Based on the initial symbiotic decision framework, state synchronization and risk path exploration are performed to generate a predictive state space; Based on the predictive state space, performing hierarchical reinforcement learning self-evolution training, evolving the optimal strategy through a virtual trial and error mechanism, and generating an evolutionary control strategy group; Based on the evolutionary control strategy group, the reality gap residual is calculated as a meta-level feedback signal, and the multimodal digital twin model and the hierarchical reinforcement learning network are reversely adjusted to generate a self-healing optimization model; Based on the self-healing optimization model, multi-dimensional risk quantification and dynamic ratio control are carried out, a safety assessment report is output, and adaptive control of the hydrogen blending ratio is performed.

Images