Method for adjusting nitrogen and oxygen concentration of cold-chain air conditioner based on deep neural network

Abstract

The invention discloses a cold-chain air conditioner nitrogen and oxygen concentration adjusting method based on a deep neural network. The cold-chain air conditioner nitrogen and oxygen concentration adjusting method comprises the following steps that 1, potential feature representation is constructed; 2, inputting the potential feature representation into a multi-target variational inversion network, and reasoning to obtain an oxygen potential disturbance field; 3, generating an oxygen content prediction curve based on the oxygen potential disturbance field in combination with a historical adjustment behavior response relationship; 4, a decision calculation module generates an adjustment strategy sequence; 5, synchronously issuing the adjustment strategy sequence to each device; step 6, after each adjustment period is finished, writing the feedback sample into an incremental training cache; step 7, when the incremental training cache reaches a set capacity threshold value, performing weight updating on the multi-target variational inversion network by using an adaptive learning rate mechanism; and 8, when the disturbance reasoning abnormity of the oxygen potential disturbance field is identified, starting a field pattern tracking recovery mechanism. According to the method, a cold chain oxygen control strategy is intelligently generated in combination with multi-target variational inversion and a graph tracking recovery mechanism.

Description

Technical Field

[0001] The present invention relates to the field of cold chain logistics gas control technology, and in particular to a method for regulating nitrogen and oxygen concentrations in a cold chain gas conditioner based on a deep neural network. Background Art

[0002] With the continuous expansion of cross-regional distribution of fresh agricultural products and the increasing requirements for the quality of goods preservation in cold chain transportation, cold chain gas conditioning technology based on gas composition regulation has become an important means to ensure the quality of goods storage and transportation. Existing cold chain gas conditioning machines mostly use fixed control thresholds or simplified feedback adjustment algorithms to control nitrogen injection and fan circulation based on oxygen concentration monitoring values ​​to maintain a set low-oxygen environment. However, the following problems are common in complex practical applications: The collected oxygen concentration, temperature, humidity and pressure difference sensor data have the problems of sparse spatial distribution and inconsistent dynamic response, which makes it difficult to fully reflect the real spatiotemporal evolution process of gas exchange inside the cold chain cabinet, resulting in insufficient information for the control strategy and delayed response. Traditional control methods mostly use linear regression, proportional-integral algorithms or empirical logic rules for modeling, but lack the ability to deeply model the metabolic behavior of goods, nitrogen disturbance effects and changes in system air tightness. In actual operation, they cannot accurately predict the comprehensive impact of adjustment actions, and are prone to under- or over-adjustment. The adjustment system lacks effective memory and utilization of the causal relationship between historical behavior and disturbance results, making it difficult for the adjustment strategy to adaptively update with changes in cargo type, load and seasonality. In long-term operation, the model performance is prone to degradation and control accuracy is reduced. When sensor drift, model inaccuracy or execution unit abnormality occurs, existing equipment mostly operates with fixed parameters as a backup, lacks a recovery mechanism based on reasoning of historical similar scenarios, and cannot maintain optimal control capabilities under abnormal conditions, resulting in adjustment blind spots and preservation risks.

[0003] Therefore, how to provide a method for regulating nitrogen and oxygen concentration in cold chain gas conditioners based on deep neural networks is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0004] One purpose of the present invention is to propose a method for regulating nitrogen and oxygen concentrations in cold chain gas conditioners based on deep neural networks. The present invention combines a multi-objective variational inversion network with an oxygen content prediction network to intelligently construct an oxygen potential disturbance field and generate a cold chain gas conditioning control strategy, thereby solving the problems of inaccurate oxygen concentration prediction, delayed response of the adjustment strategy, and unstable control under complex disturbances. The method has the advantages of precise adjustment, strong adaptability, stable operation, and high abnormal recovery capability, and is suitable for the intelligent gas conditioning control needs of various types of goods and under multiple environmental conditions.

[0005] According to an embodiment of the present invention, a method for adjusting nitrogen and oxygen concentrations in a cold chain gas conditioner based on a deep neural network includes the following steps: Step 1: Collect multi-dimensional historical operation data and perform cross-dimensional embedding coding to construct latent feature representation; Step 2: Input the latent feature representation into a multi-objective variational inversion network to generate latent variables of the current metabolic state and infer the oxygen potential disturbance field; Step 3: Based on the oxygen potential disturbance field and in combination with the historical regulatory behavior response relationship, generate an oxygen content prediction curve within a future target time window; Step 4: The decision calculation module generates an adjustment strategy sequence based on the oxygen content prediction curve and the upper and lower oxygen concentration thresholds corresponding to the cargo, combined with the oxygen potential disturbance field. The adjustment strategy sequence includes the nitrogen production set value, fan frequency, air supply valve opening, and oxygen reduction trigger judgment value; Step 5: The control module executes the adjustment strategy sequence to be synchronously sent to the nitrogen generator, blower and air supply valve to achieve dynamic coordination of nitrogen injection, oxygen replacement, gas circulation and pressure difference stabilization; Step 6: After each adjustment cycle, the real-time sampling sequence, adjustment strategy sequence, and final oxygen content are packaged as feedback samples and written into the incremental training cache; Step 7: When the incremental training cache reaches a set capacity threshold, the weights of the multi-objective variational inversion network are updated using an adaptive learning rate mechanism; Step 8: When it is identified that the oxygen potential disturbance field has a disturbance inference anomaly, the field map tracking and recovery mechanism is started, the oxygen potential disturbance field state with the best similarity is matched from the historical operation database, and the emergency recovery control parameters are quickly generated and immediately issued for execution.

[0006] Optionally, the multidimensional historical operation data includes cargo type, loading capacity, cold chain equipment ambient temperature, humidity, air pressure, membrane separation nitrogen generator power level, nitrogen flow, valve opening, fan frequency, oxygen concentration in the cabinet, nitrogen purity and pressure difference.

[0007] Optionally, the cross-dimensional embedding coding is performed to construct a potential feature representation, specifically: Extracting features of the multidimensional historical operation data, normalizing them, and inputting them into corresponding feature embedding sub-networks; The feature embedding subnetwork is an encoding module including a fully connected layer and a ReLU activation function, which is used to map the original features of each dimension into an embedding vector of a set length; The embedding vector is input into the crisscross attention mechanism module, which consists of a query network, a key-value network and a weighted fusion module: The query network takes the embedding vectors of cargo types and load quantities as input and generates an attention query matrix; The key-value network takes the remaining dimensional feature embedding vectors as input to generate an attention key matrix and an attention value matrix; The remaining dimensional features include the cold chain equipment ambient temperature, humidity, air pressure, membrane separation nitrogen generator power level, nitrogen flow rate, valve opening, fan frequency, cabinet oxygen concentration, nitrogen purity, and pressure difference characteristics; The weighted fusion module calculates attention weights based on the dot product similarity between the attention query matrix and the attention key matrix, and applies the attention weights to the attention value matrix to obtain a context-aware representation; The context-aware representation is concatenated with the original embedding vector to form a latent feature representation, which serves as the input of a multi-objective variational inversion network.

[0008] Optionally, the step 2 is specifically as follows: Inputting the latent feature representation into a multi-objective variational inversion network, wherein the target variational inversion network includes an encoder subnetwork, a variational sampling module, a decoder subnetwork and a perturbation mapping layer; The encoder subnetwork includes a plurality of sequentially connected fully connected layers and ReLU activation functions for encoding the latent feature representation into a mean vector and a logarithmic variance vector of a latent distribution; The variational sampling module generates the current metabolic state latent variable through a reparameterization method according to the mean vector and the logarithmic variance vector; The decoder subnetwork includes an oxygen metabolic rate estimation branch, a nitrogen displacement intensity inversion branch, and a pressure difference adaptation branch; The outputs of the three branches are merged and input into the perturbation mapping layer, which includes a spatial position encoding unit, a temporal recursive generation unit, and a tensor combination module; The spatial position encoding unit is used to receive the spatial layout parameters of the goods in the cold chain cabinet, the air duct distribution map and the sensor layout map, discretize the spatial structure information into grids, and generate a spatial position matrix; The time recursive generation unit is a group of gated recurrent units with shared weights, which receives the outputs of the three branches, performs recursive iterative calculations based on the historical adjustment behavior sequence, and outputs a disturbance state vector with time series characteristics; The tensor combination module receives the spatial position matrix and the disturbance state vector, and constructs an oxygen potential disturbance field with a three-dimensional structure through point-by-point mapping and channel fusion. The channel dimension of each position point represents the oxygen concentration disturbance trend of the position point in the next several time steps.

[0009] Optionally, the oxygen metabolic rate estimation branch includes a time convolution layer and a temperature modulation unit, wherein the time convolution layer is used to extract the time characteristics of the latent variable of the current metabolic state, and the temperature modulation unit adjusts the amplitude of the convolution result according to the ambient temperature to generate the oxygen metabolic consumption per unit time in the current state; The nitrogen displacement intensity inversion branch includes two fully connected layers connected in series and an attention weighting mechanism. The attention weighting mechanism assigns feature weights based on the coupling relationship between the nitrogen injection rate and the oxygen concentration change rate to generate an oxygen concentration change intensity factor caused by nitrogen injection in the current state. The pressure difference adaptation branch includes two residual convolution units and an airtightness adjustment factor. The residual convolution unit is used to extract multi-scale flow structure characteristics. The airtightness adjustment factor corrects the pressure difference estimation value according to the equipment sealing level and the structural parameters in the cabinet to output the pressure difference compensation value.

[0010] Optionally, the step three is specifically as follows: Inputting the oxygen potential disturbance field into an oxygen content prediction network, wherein the oxygen content prediction network includes a disturbance feature extraction module, a historical response memory module, and a sequence curve generation module; The disturbance feature extraction module includes multiple stacked two-dimensional convolutional layers and residual connection structures, receives the oxygen potential disturbance field, extracts the spatial features of the oxygen potential disturbance field, and outputs a spatial disturbance feature map; The historical response memory module includes a gated recurrent unit network, which receives a historical adjustment behavior parameter sequence and a corresponding historical oxygen concentration measurement sequence, extracts the temporal response characteristics between the historical adjustment behavior and the oxygen concentration, and outputs a historical response feature vector; The sequence curve generation module includes an attention fusion network and a cyclic decoding network. The attention fusion network is used to fuse the spatial disturbance feature map and the historical response feature vector to form a disturbance response feature. The cyclic decoding network decodes the disturbance response feature time-step by time step and outputs the oxygen concentration prediction value for each time step in the future target time window. The oxygen concentration prediction values ​​of each time step are sequentially combined into a complete sequence to generate an oxygen content prediction curve within a future target time window.

[0011] Optionally, the step 4 is specifically as follows: The decision calculation module identifies all time segments where the predicted oxygen concentration exceeds the upper and lower thresholds of the corresponding oxygen concentration for the cargo within the target time window based on the oxygen content prediction curve, and records the maximum deviation value of each exceeding limit segment. and duration of overrun ; According to the change in oxygen concentration caused by each unit input of the regulation behavior in the oxygen potential disturbance field, the disturbance response coefficient is defined as: ; in, represents the disturbance response coefficient of the i-th type regulatory behavior unit, Represents the unit adjustment amount of the i-th type of adjustment behavior The change in oxygen concentration caused by represents the unit adjustment amount of the i-th type of adjustment behavior; In each oxygen concentration exceeding limit range, the control input that meets the following optimization objectives is calculated: ; If multiple If the optimization goal is met, the one with the lowest energy consumption per unit input is selected; When the maximum deviation When the deviation from the preset early intervention threshold is less than the disturbance response coefficient, The maximum regulation behavior is to perform low-amplitude compensation operations to suppress the upward trend of oxygen concentration. That is, when the oxygen concentration is predicted to deviate slightly but not seriously, the control method with the highest regulation efficiency is selected and only light intervention is performed to prevent the problem from escalating and avoid excessive control. If there is a continuous over-limit duration in the oxygen content prediction curve Greater than the preset overtime threshold or maximum deviation value If the maximum deviation is greater than the preset target upper limit, the oxygen reduction trigger state is set, the oxygen reduction trigger judgment value is set to 1, the nitrogen injection rate and fan frequency in the corresponding time period are specified to be the maximum adjustment values, and the air supply valve is closed; when the oxygen reduction trigger judgment value is set to 0, the oxygen reduction trigger state is not started.

[0012] Finally, the control parameter set is combined into an adjustment strategy sequence in chronological order, and the adjustment strategy sequence includes a nitrogen production set value, a fan frequency, an air supply valve opening, and an oxygen reduction trigger judgment value.

[0013] Optionally, the gas supplied by the air supply valve is air, which is used to increase the oxygen concentration.

[0014] Optionally, the step seven is specifically as follows: All feedback samples in the incremental training buffer are organized into several batches of training data in the order of their collection time, where each feedback sample contains a data sequence sampled in real time during a complete regulation cycle, a corresponding sequence of actually executed regulation strategies, and a final oxygen concentration measurement result; Using the existing weight parameters in the current multi-objective variational inversion network as initial parameters, inputting the training data batches into the multi-objective variational inversion network one by one, and determining the current loss function size by calculating the error between the output and the actual measurement result; An adaptive learning rate mechanism is used during training. After each training batch, the loss function is evaluated compared to the loss function of the previous batch. If the loss function continuously decreases at a rate less than the set deceleration threshold or even increases, the current learning rate is reduced to avoid fluctuations in parameter updates or degradation of model performance. If the loss function continues to decrease at a rate greater than the set decrease threshold, the current learning rate is increased to speed up the convergence of the weight parameters. If the rate of decrease of the loss function is equal to the set rate-decreasing threshold, the current learning rate is kept unchanged to maintain a stable training effect; An early stopping condition is set during the training process. When the loss function of the multi-objective variational inversion network model changes less than the set lower limit after multiple batches of training, indicating that the performance improvement is no longer significant, the current round of fine-tuning is terminated. The latest multi-objective variational inversion network model weight parameters obtained after fine-tuning are used to replace the original weight parameters. At the same time, the incremental training cache is cleared and new feedback samples are accumulated again to maintain the dynamic adaptability of the multi-objective variational inversion network model to the cold chain system environment and equipment status.

[0015] Optionally, the field image tracking and recovery mechanism is specifically: If the deviation between the predicted output of the oxygen potential disturbance field and the measured oxygen concentration exceeds the dynamic tolerance threshold over multiple consecutive sampling periods, or if the output of the oxygen potential disturbance field is missing a value or decoding fails, it is identified as a disturbance inference anomaly. After identifying the disturbance inference anomaly, data samples containing oxygen potential disturbance field state labels are retrieved from the historical operation database. A joint similarity measurement function is constructed based on the spatial distribution structure of the oxygen potential disturbance field and the temporal characteristics of the corresponding regulation behavior. The current abnormal state is matched with the historical oxygen potential disturbance field state, and multiple historical oxygen potential disturbance field state samples with the best similarity are determined as candidate sets. Performing a weighted combination of the corresponding control parameters in the candidate set to generate emergency recovery control parameters for the current regulation cycle, wherein the emergency recovery control parameters include the nitrogen production set value, the fan frequency, and the air supply valve opening; Send the emergency recovery control parameters to each corresponding control object to replace the original adjustment strategy sequence for execution; During the execution process, the disturbance inference state and actual response data are collected. When the deviation between the predicted output of the oxygen potential disturbance field and the actual oxygen concentration measured value is less than or equal to the dynamic tolerance threshold, and there is no value gap or decoding failure in the output of the oxygen potential disturbance field, the original control path is restored and the field map tracking recovery process is ended.

[0016] The beneficial effects of the present invention are: The present invention uses the collaborative modeling of a multi-objective variational inversion network and an oxygen content prediction network to address the problems of large disturbance uncertainty, complex regulation response mechanism and poor model adaptability in the oxygen concentration control of cold chain gas conditioning systems. It proposes a regulation strategy generation method jointly driven by cross-dimensional potential feature expression and disturbance response inversion. It uses an embedded context representation based on cargo variety, load capacity and environmental parameters, and realizes the nonlinear interactive fusion of multi-dimensional features through the attention mechanism to obtain a metabolic latent variable representation that adapts to various operating states. In the inversion stage, oxygen metabolism, nitrogen injection disturbance and pressure difference compensation are modeled as independent branches, and their dynamic causal relationships are extracted respectively and integrated into the oxygen potential disturbance field with spatiotemporal structure to further drive the target. The generation of time period oxygen content prediction curves introduces a disturbance response coefficient at the control level to optimize the adjustment input amplitude, matching the minimum control change rate path according to the degree of deviation from the target oxygen concentration to achieve precise adjustment of the oxygen content. In terms of model self-update, by constructing a feedback sample cache with target labels and combining it with a dynamic loss trend to adjust the learning rate mechanism, online fine-tuning and updating of the disturbance inversion network are achieved, improving the model's adaptability to seasonal changes and cargo characteristics. In terms of abnormal state processing, a field map tracking and recovery mechanism for the oxygen potential disturbance field is introduced for the first time. When reasoning fails or abnormal fluctuations occur, similar field states are quickly retrieved from historical samples to construct restorative control parameters, ensuring the system's adjustment stability and operational safety in complex situations. Ultimately, the stable prediction of oxygen content in cold chain gas conditioning cabinets, low-energy control, and anomaly recovery capabilities are comprehensively improved, significantly surpassing the adjustment effects of traditional feedback control and linear modeling methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is an overall flow chart of a method for adjusting nitrogen and oxygen concentrations in a cold chain gas conditioner based on a deep neural network proposed in the present invention; Figure 2 This is a schematic diagram of the multi-objective variational inversion network structure of a cold chain gas conditioner nitrogen and oxygen concentration adjustment method based on a deep neural network proposed in the present invention; Figure 3 This is a matching and emergency control flow chart of the field graph tracking recovery mechanism of the cold chain gas conditioner nitrogen and oxygen concentration adjustment method based on deep neural network proposed in the present invention. DETAILED DESCRIPTION

[0018] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0019] refer to Figure 1-Figure 3 A method for adjusting nitrogen and oxygen concentration in a cold chain gas conditioner based on a deep neural network comprises the following steps: Step 1: Collect multi-dimensional historical operation data and perform cross-dimensional embedding coding to construct latent feature representation; Step 2: Input the latent feature representation into a multi-objective variational inversion network to generate latent variables of the current metabolic state and infer the oxygen potential disturbance field; Step 3: Based on the oxygen potential disturbance field and in combination with the historical regulatory behavior response relationship, generate an oxygen content prediction curve within a future target time window; Step 4: The decision calculation module generates an adjustment strategy sequence based on the oxygen content prediction curve and the upper and lower oxygen concentration thresholds corresponding to the cargo, combined with the oxygen potential disturbance field. The adjustment strategy sequence includes the nitrogen production set value, fan frequency, air supply valve opening, and oxygen reduction trigger judgment value; Step 5: The control module executes the adjustment strategy sequence to be synchronously sent to the nitrogen generator, blower and air supply valve to achieve dynamic coordination of nitrogen injection, oxygen replacement, gas circulation and pressure difference stabilization; Step 6: After each adjustment cycle, the real-time sampling sequence, adjustment strategy sequence, and final oxygen content are packaged as feedback samples and written into the incremental training cache; Step 7: When the incremental training cache reaches a set capacity threshold, the weights of the multi-objective variational inversion network are updated using an adaptive learning rate mechanism; Step 8: When it is identified that the oxygen potential disturbance field has a disturbance inference anomaly, the field map tracking and recovery mechanism is started, the oxygen potential disturbance field state with the best similarity is matched from the historical operation database, and the emergency recovery control parameters are quickly generated and immediately issued for execution.

[0020] In this embodiment, the multidimensional historical operation data includes cargo type, loading capacity, cold chain equipment ambient temperature, humidity, air pressure, membrane separation nitrogen generator power level, nitrogen flow, valve opening, fan frequency, cabinet oxygen concentration, nitrogen purity and pressure difference.

[0021] Using a combined membrane separation and molecular sieve nitrogen generation technology, the generated nitrogen is primarily supplied to the gas conditioning tank. Once the nitrogen concentration in the gas conditioning tank reaches the set value, the nitrogen generation system continues to operate, and the excess nitrogen is converted into liquid nitrogen for storage through a condensation system. Liquid nitrogen is used for rapid replenishment during subsequent emergencies or peak periods, ensuring the continuity and stability of system operation.

[0022] In this embodiment, the cross-dimensional embedding coding is performed to construct the potential feature representation, specifically: Extracting features of the multidimensional historical operation data, normalizing them, and inputting them into corresponding feature embedding sub-networks; The feature embedding subnetwork is an encoding module including a fully connected layer and a ReLU activation function, which is used to map the original features of each dimension into an embedding vector of a set length; The embedding vector is input into the crisscross attention mechanism module, which consists of a query network, a key-value network and a weighted fusion module: The query network takes the embedding vectors of cargo types and load quantities as input and generates an attention query matrix; The key-value network takes the remaining dimensional feature embedding vectors as input to generate an attention key matrix and an attention value matrix; The remaining dimensional features include the cold chain equipment ambient temperature, humidity, air pressure, membrane separation nitrogen generator power level, nitrogen flow rate, valve opening, fan frequency, cabinet oxygen concentration, nitrogen purity, and pressure difference characteristics; The weighted fusion module calculates attention weights based on the dot product similarity between the attention query matrix and the attention key matrix, and applies the attention weights to the attention value matrix to obtain a context-aware representation; The context-aware representation is concatenated with the original embedding vector to form a latent feature representation, which serves as the input of a multi-objective variational inversion network.

[0023] In this embodiment, the step 2 is specifically as follows: Inputting the latent feature representation into a multi-objective variational inversion network, wherein the target variational inversion network includes an encoder subnetwork, a variational sampling module, a decoder subnetwork and a perturbation mapping layer; The encoder subnetwork includes a plurality of sequentially connected fully connected layers and ReLU activation functions for encoding the latent feature representation into a mean vector and a logarithmic variance vector of a latent distribution; The variational sampling module generates the current metabolic state latent variable through a reparameterization method according to the mean vector and the logarithmic variance vector; The decoder subnetwork includes an oxygen metabolic rate estimation branch, a nitrogen displacement intensity inversion branch, and a pressure difference adaptation branch; The outputs of the three branches are merged and input into the perturbation mapping layer, which includes a spatial position encoding unit, a temporal recursive generation unit, and a tensor combination module; The spatial position encoding unit is used to receive the spatial layout parameters of the goods in the cold chain cabinet, the air duct distribution map and the sensor layout map, discretize the spatial structure information into grids, and generate a spatial position matrix; The time recursive generation unit is a group of gated recurrent units with shared weights, which receives the outputs of the three branches, performs recursive iterative calculations based on the historical adjustment behavior sequence, and outputs a disturbance state vector with time series characteristics; The tensor combination module receives the spatial position matrix and the disturbance state vector, and constructs an oxygen potential disturbance field with a three-dimensional structure through point-by-point mapping and channel fusion. The channel dimension of each position point represents the oxygen concentration disturbance trend of the position point in the next several time steps.

[0024] In this embodiment, the oxygen metabolism rate estimation branch includes a time convolution layer and a temperature modulation unit. The time convolution layer is used to extract the time characteristics of the latent variable of the current metabolic state. The temperature modulation unit adjusts the amplitude of the convolution result according to the ambient temperature to generate the oxygen metabolism consumption per unit time in the current state. The nitrogen displacement intensity inversion branch includes two fully connected layers connected in series and an attention weighting mechanism. The attention weighting mechanism assigns feature weights based on the coupling relationship between the nitrogen injection rate and the oxygen concentration change rate to generate an oxygen concentration change intensity factor caused by nitrogen injection in the current state. The pressure difference adaptation branch includes two residual convolution units and an airtightness adjustment factor. The residual convolution unit is used to extract multi-scale flow structure characteristics. The airtightness adjustment factor corrects the pressure difference estimation value according to the equipment sealing level and the structural parameters in the cabinet to output the pressure difference compensation value.

[0025] In this embodiment, the step three is specifically as follows: Inputting the oxygen potential disturbance field into an oxygen content prediction network, wherein the oxygen content prediction network includes a disturbance feature extraction module, a historical response memory module, and a sequence curve generation module; The disturbance feature extraction module includes multiple stacked two-dimensional convolutional layers and residual connection structures, receives the oxygen potential disturbance field, extracts the spatial features of the oxygen potential disturbance field, and outputs a spatial disturbance feature map; The historical response memory module includes a gated recurrent unit network, which receives a historical adjustment behavior parameter sequence and a corresponding historical oxygen concentration measurement sequence, extracts the temporal response characteristics between the historical adjustment behavior and the oxygen concentration, and outputs a historical response feature vector; The sequence curve generation module includes an attention fusion network and a cyclic decoding network. The attention fusion network is used to fuse the spatial disturbance feature map and the historical response feature vector to form a disturbance response feature. The cyclic decoding network decodes the disturbance response feature time-step by time step and outputs the oxygen concentration prediction value for each time step in the future target time window. The oxygen concentration prediction values ​​of each time step are sequentially combined into a complete sequence to generate an oxygen content prediction curve within a future target time window.

[0026] In this embodiment, the step 4 is specifically as follows: The decision calculation module identifies all time segments where the predicted oxygen concentration exceeds the upper and lower thresholds of the corresponding oxygen concentration for the cargo within the target time window based on the oxygen content prediction curve, and records the maximum deviation value of each exceeding limit segment. and duration of overrun ; According to the change in oxygen concentration caused by each unit input of the regulation behavior in the oxygen potential disturbance field, the disturbance response coefficient is defined as: ; in, represents the disturbance response coefficient of the i-th type regulatory behavior unit, Represents the unit adjustment amount of the i-th type of adjustment behavior The change in oxygen concentration caused by represents the unit adjustment amount of the i-th type of adjustment behavior; In each oxygen concentration exceeding limit range, the control input that meets the following optimization objectives is calculated: ; If multiple If the optimization goal is met, the one with the lowest energy consumption per unit input is selected; When the maximum deviation When the deviation from the preset early intervention threshold is less than the disturbance response coefficient, The maximum regulation behavior is to perform low-amplitude compensation operations to suppress the upward trend of oxygen concentration. That is, when the oxygen concentration is predicted to deviate slightly but not seriously, the control method with the highest regulation efficiency is selected and only light intervention is performed to prevent the problem from escalating and avoid excessive control. If there is a continuous over-limit duration in the oxygen content prediction curve Greater than the preset overtime threshold or maximum deviation value If the maximum deviation is greater than the preset target upper limit, the oxygen reduction trigger state is set, the oxygen reduction trigger judgment value is set to 1, the nitrogen injection rate and fan frequency in the corresponding time period are specified to be the maximum adjustment values, and the air supply valve is closed; when the oxygen reduction trigger judgment value is set to 0, the oxygen reduction trigger state is not started.

[0027] Finally, the control parameter set is combined into an adjustment strategy sequence in chronological order, and the adjustment strategy sequence includes a nitrogen production set value, a fan frequency, an air supply valve opening, and an oxygen reduction trigger judgment value.

[0028] In this embodiment, the gas supplied by the gas supply valve is air, which is used to increase the oxygen concentration.

[0029] In this embodiment, the step seven is specifically as follows: All feedback samples in the incremental training buffer are organized into several batches of training data in the order of their collection time, where each feedback sample contains a data sequence sampled in real time during a complete regulation cycle, a corresponding sequence of actually executed regulation strategies, and a final oxygen concentration measurement result; Using the existing weight parameters in the current multi-objective variational inversion network as initial parameters, inputting the training data batches into the multi-objective variational inversion network one by one, and determining the current loss function size by calculating the error between the output and the actual measurement result; The loss function of the multi-objective variational inversion network consists of multiple sub-objectives. It aims to simultaneously optimize the prediction accuracy of oxygen metabolism rate estimation, nitrogen displacement intensity inversion, and pressure difference compensation tasks, while ensuring the stability of the latent space expression. The loss function mainly consists of two parts: The first is the reconstruction error term, which is used to measure the deviation between the model output and the actual sampled data, ensuring that the prediction results of each output branch can accurately reflect the actual system state; The second is the variational regularization term, which is used to constrain the gap between the distribution of latent variables and the standard normal distribution, thereby improving the generalization ability and stability of the model.

[0030] An adaptive learning rate mechanism is used during training. After each training batch, the loss function is evaluated compared to the loss function of the previous batch. If the loss function continuously decreases at a rate less than the set deceleration threshold or even increases, the current learning rate is reduced to avoid fluctuations in parameter updates or degradation of model performance. If the loss function continues to decrease at a rate greater than the set decrease threshold, the current learning rate is increased to speed up the convergence of the weight parameters. If the rate of decrease of the loss function is equal to the set rate-decreasing threshold, the current learning rate is kept unchanged to maintain a stable training effect; An early stopping condition is set during the training process. When the loss function of the multi-objective variational inversion network model changes less than the set lower limit after multiple batches of training, indicating that the performance improvement is no longer significant, the current round of fine-tuning is terminated. The latest multi-objective variational inversion network model weight parameters obtained after fine-tuning are used to replace the original weight parameters. At the same time, the incremental training cache is cleared and new feedback samples are accumulated again to maintain the dynamic adaptability of the multi-objective variational inversion network model to the cold chain system environment and equipment status.

[0031] In this implementation, the field pattern tracking and recovery mechanism is specifically as follows: If the deviation between the predicted output of the oxygen potential disturbance field and the measured oxygen concentration exceeds the dynamic tolerance threshold over multiple consecutive sampling periods, or if the output of the oxygen potential disturbance field is missing a value or decoding fails, it is identified as a disturbance inference anomaly. After identifying the disturbance inference anomaly, data samples containing oxygen potential disturbance field state labels are retrieved from the historical operation database. A joint similarity measurement function is constructed based on the spatial distribution structure of the oxygen potential disturbance field and the temporal characteristics of the corresponding regulation behavior. The current abnormal state is matched with the historical oxygen potential disturbance field state, and multiple historical oxygen potential disturbance field state samples with the best similarity are determined as candidate sets. Assume that the oxygen potential disturbance field of the current abnormal state is a three-dimensional tensor , where X, Y are spatial dimensions, representing the horizontal and vertical grid coordinates of the cold chain cabinet, and T is the time dimension, representing the future prediction time step.

[0032] Assume that the oxygen potential disturbance field of the kth candidate sample in the historical operation database is , and the corresponding regulatory behavior sequence is ,in represents the control parameter vector of the kth candidate sample at time step T; The current regulation behavior sequence is ,in Represents the control parameter vector of the current regulation behavior sequence at time step t; Joint similarity measure function The definition is as follows: ; in, represents the weight coefficient, Represents the difference measure of the spatial disturbance field: ; in, represents the Frobenius norm; : A measure of the difference between the sequences of regulatory behaviors: ; in, represents the inner product operation, represents the two-norm; The final candidate set is: ; Among them, TopK means selecting similarity scores from N historical samples The smallest K candidate samples constitute the candidate set.

[0033] Performing a weighted combination of the corresponding control parameters in the candidate set to generate emergency recovery control parameters for the current regulation cycle, wherein the emergency recovery control parameters include the nitrogen production set value, the fan frequency, and the air supply valve opening; Send the emergency recovery control parameters to each corresponding control object to replace the original adjustment strategy sequence for execution; During the execution process, the disturbance inference state and actual response data are collected. When the deviation between the predicted output of the oxygen potential disturbance field and the actual oxygen concentration measured value is less than or equal to the dynamic tolerance threshold, and there is no value gap or decoding failure in the output of the oxygen potential disturbance field, the original control path is restored and the field map tracking recovery process is ended. Example

[0034] In order to verify the feasibility of the present invention in practice, the present invention was applied to the cold chain logistics environmental control system of a certain food storage enterprise. Faced with the problem of oxygen concentration fluctuations during the transportation of high-value perishable goods (such as fresh fruits and seafood), the conventional PID control strategy has a delayed response in a complex dynamic environment and cannot achieve real-time coupling of the respiratory metabolism and displacement regulation of the goods, which can easily cause the oxygen concentration to be too high or to decrease too quickly, affecting the preservation effect. In order to verify the effectiveness of the nitrogen and oxygen concentration adjustment method based on deep neural network proposed in the present invention in the cold chain scenario, a batch of highly active tropical fruits were selected as target goods, and a continuous adjustment experiment was carried out. During the experiment, the multi-dimensional operating status was sampled in real time, and the method of the present invention was used for online adjustment decision-making and execution control.

[0035] During the actual deployment process, a latent feature representation is first constructed based on multi-dimensional historical operating data such as cargo type, load volume, cold chain equipment ambient temperature, humidity, air pressure, membrane separation nitrogen generator power level, nitrogen flow rate, valve opening, fan frequency, cabinet oxygen concentration, nitrogen purity, and pressure difference. This is then fed into a multi-objective variational inversion network to generate an oxygen potential disturbance field. Combined with historical regulation behavior response relationships, the oxygen concentration trend over the next 30 minutes is predicted. If the prediction results indicate that the oxygen concentration will exceed the target upper limit within the next 5 minutes, the optimal regulation behavior unit is calculated based on the disturbance response coefficient, and an intelligent regulation strategy sequence is generated and issued for execution.

[0036] The system executed a total of 11 continuous control cycles (each cycle lasted about 2 minutes). Table 1 shows the cold chain regulation experiment data table, which records the oxygen concentration, regulation parameters and disturbance status in each cycle.

[0037] Table 1 Cold chain regulation experiment data Control cycle Initial oxygen concentration (%) Nitrogen production setting value (L / min) Fan frequency (Hz) Air supply valve opening (%) Predicted oxygen concentration (%) Measured oxygen concentration (%) Pressure difference (Pa) T1 20.9 25 40 0 18.2 18.3 120 T2 18.2 28 45 0 15.6 15.5 125 T3 15.6 30 50 0 13.4 13.5 130 T4 13.4 32 55 0 11.2 11.1 135 T5 11.2 35 60 0 10.1 10.0 138 T6 10.1 30 55 5 10.4 10.3 135 T7 10.4 28 50 10 10.2 10.2 132 T8 10.2 27 48 12 10.3 10.2 130 T9 10.3 26 46 14 10.1 10.2 128 T10 10.1 25 45 15 10.2 10.1 126 T11 10.2 25 45 15 10.2 10.2 125 As shown in Table 1, the initial oxygen concentration was 20.9%. Over three consecutive cycles, increasing nitrogen production and fan frequency rapidly reduced the oxygen concentration to approximately 10.1%. The strategy then automatically switched to maintaining a stable state. Through small adjustments to air supply and frequency, the oxygen concentration remained within the target range, and the overall differential pressure remained stable between 125 and 135 Pa. The system achieved a smooth transition from a high-oxygen environment to the target range without significant overshoot or oscillation.

[0038] To test the present invention's ability to recover from abnormal state identification, some oxygen concentration sensors were intentionally disconnected after control period T7. The system identified a disturbance field prediction anomaly and triggered the field pattern tracking recovery mechanism. During this recovery mechanism, the system retrieved samples containing complete disturbance state labels from the historical operation database and matched the four most similar field states using a similarity metric function.

[0039] Table 2. Field map tracking recovery mechanism candidate sample data table Sample No. Disturbance spatial difference score Behavioral sequence difference score Joint similarity score Sample weight coefficient Control parameters (nitrogen / blower / valve) H001 0.183 0.209 0.196 0.30 85 / 42 / 25 H004 0.166 0.241 0.202 0.26 83 / 40 / 28 H007 0.191 0.214 0.202 0.24 86 / 43 / 27 H011 0.176 0.232 0.204 0.20 84 / 41 / 29 As shown in Table 2, after identifying the disturbance inference anomaly, the system successfully triggered the field map tracking and recovery mechanism. It then retrieved and matched four historical disturbance field state samples with the best joint similarity scores from the historical operation database: H001, H004, H007, and H011. Each sample had a low value for both the disturbance spatial difference score and the behavior sequence difference score, with an overall joint similarity score of no more than 0.205, indicating a high degree of similarity with the current abnormal state in terms of both spatial disturbance structure and control behavior characteristics.

[0040] Among them, sample H001 has the lowest joint similarity score (0.196), indicating the strongest coupling with the current state. Therefore, it is assigned the highest sample weight (0.30). This is followed by H004, H007, and H011, which are assigned weight coefficients of 0.26, 0.24, and 0.20, respectively. The system performs a weighted fusion of the control parameters of these four candidate samples according to their weight ratios. The resulting emergency recovery control parameters are a nitrogen production of 84.52 L / min, a fan frequency of 41.52 Hz, and an air supply valve opening of 27.06%. Both the fan frequency and air supply valve opening are within acceptable ranges. While nitrogen production, expressed in L / min, represents a relatively large total flow rate, because emergency recovery is triggered under abnormal conditions, a conservative high flow rate is permitted for a short period of time to ensure stable oxygen concentration. After the emergency recovery control parameters were immediately issued and put into execution, the oxygen concentration in the subsequent cycles was successfully maintained within the target range, and the pressure difference changed stably without abnormal fluctuations or system jitter, fully verifying the response efficiency and control accuracy of the field pattern tracking recovery mechanism in actual operation.

[0041] This round of emergency response experiments demonstrates that the proposed anomaly recognition mechanism can rapidly detect disturbance inference anomalies, while the field graph tracking and recovery mechanism boasts high-precision historical sample matching capabilities. This allows for the automatic generation of highly adaptable control parameters in a very short time, ensuring the continuity and stability of oxygen concentration regulation in the cold chain gas conditioning system even under abnormal conditions. The overall system demonstrates excellent robustness and adaptability, significantly outperforming traditional rule-based preset recovery strategies and possessing broad practical application value.

[0042] This embodiment demonstrates the present invention's real-time analysis capability for multi-target nonlinear disturbances in a complex cold chain environment and its robustness in online response to abnormal conditions. It effectively solves the pain points of traditional methods in terms of inaccurate prediction, unstable control, and emergency failure, and is particularly suitable for high-precision fresh food transportation, vaccine storage and transportation, and precision equipment cold chain transportation scenarios.

[0043] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for regulating nitrogen and oxygen concentration in cold chain gas conditioners based on deep neural networks, characterized in that: The steps include: Step 1: Collect multi-dimensional historical operation data and perform cross-dimensional embedding coding to construct latent feature representation; Step 2: Input the latent feature representation into a multi-objective variational inversion network to generate latent variables of the current metabolic state and infer the oxygen potential disturbance field; Step 3: Based on the oxygen potential disturbance field and in combination with the historical regulatory behavior response relationship, generate an oxygen content prediction curve within a future target time window; Step 4: The decision calculation module generates an adjustment strategy sequence based on the oxygen content prediction curve and the upper and lower oxygen concentration thresholds corresponding to the cargo, combined with the oxygen potential disturbance field. The adjustment strategy sequence includes the nitrogen production set value, fan frequency, air supply valve opening, and oxygen reduction trigger judgment value; Step 5: The adjustment strategy sequence is synchronously sent to the nitrogen generator, the blower and the air supply valve through the execution control module; Step 6: After each adjustment cycle, the real-time sampling sequence, adjustment strategy sequence, and final oxygen content are packaged as feedback samples and written into the incremental training cache; Step 7: When the incremental training cache reaches a set capacity threshold, the weights of the multi-objective variational inversion network are updated using an adaptive learning rate mechanism; Step 8: When it is identified that the oxygen potential disturbance field has a disturbance inference anomaly, the field map tracking and recovery mechanism is started, the oxygen potential disturbance field state with the best similarity is matched from the historical operation database, and the emergency recovery control parameters are quickly generated and immediately issued for execution.

2. The method for adjusting nitrogen and oxygen concentration of cold chain gas conditioner based on deep neural network according to claim 1 is characterized in that: The multi-dimensional historical operation data includes cargo type, loading capacity, cold chain equipment ambient temperature, humidity, air pressure, membrane separation nitrogen generator power level, nitrogen flow, valve opening, fan frequency, oxygen concentration in the cabinet, nitrogen purity and pressure difference.

3. The method for adjusting nitrogen and oxygen concentration of cold chain gas conditioner based on deep neural network according to claim 1, characterized in that: The cross-dimensional embedding coding is performed to construct a potential feature representation, specifically: Extracting features of the multidimensional historical operation data, normalizing them, and inputting them into corresponding feature embedding sub-networks; The feature embedding subnetwork is an encoding module including a fully connected layer and a ReLU activation function, which is used to map the original features of each dimension into an embedding vector of a set length; The embedding vector is input into the crisscross attention mechanism module, which consists of a query network, a key-value network and a weighted fusion module: The query network takes the embedding vectors of cargo types and load quantities as input and generates an attention query matrix; The key-value network takes the remaining dimensional feature embedding vectors as input to generate an attention key matrix and an attention value matrix; The weighted fusion module calculates attention weights based on the dot product similarity between the attention query matrix and the attention key matrix, and applies the attention weights to the attention value matrix to obtain a context-aware representation; The context-aware representation is concatenated with the original embedding vector to form a latent feature representation, which serves as the input of a multi-objective variational inversion network.

4. The method for adjusting nitrogen and oxygen concentration of cold chain gas conditioner based on deep neural network according to claim 1, characterized in that: The step 2 is specifically as follows: Inputting the latent feature representation into a multi-objective variational inversion network, wherein the target variational inversion network includes an encoder subnetwork, a variational sampling module, a decoder subnetwork and a perturbation mapping layer; The encoder subnetwork includes a plurality of sequentially connected fully connected layers and ReLU activation functions for encoding the latent feature representation into a mean vector and a logarithmic variance vector of a latent distribution; The variational sampling module generates the current metabolic state latent variable through a reparameterization method according to the mean vector and the logarithmic variance vector; The decoder subnetwork includes an oxygen metabolic rate estimation branch, a nitrogen displacement intensity inversion branch, and a pressure difference adaptation branch; The outputs of the three branches are merged and input into the perturbation mapping layer, which includes a spatial position encoding unit, a temporal recursive generation unit, and a tensor combination module; The spatial position encoding unit is used to receive the spatial layout parameters of the goods in the cold chain cabinet, the air duct distribution map and the sensor layout map, discretize the spatial structure information into grids, and generate a spatial position matrix; The time recursive generation unit is a group of gated recurrent units with shared weights, which receives the outputs of the three branches, performs recursive iterative calculations based on the historical adjustment behavior sequence, and outputs a disturbance state vector with time series characteristics; The tensor combination module receives the spatial position matrix and the disturbance state vector, and constructs an oxygen potential disturbance field with a three-dimensional structure through point-by-point mapping and channel fusion. The channel dimension of each position point represents the oxygen concentration disturbance trend of the position point in the next several time steps.

5. The method for adjusting nitrogen and oxygen concentration of cold chain gas conditioner based on deep neural network according to claim 4 is characterized in that: The oxygen metabolic rate estimation branch includes a time convolution layer and a temperature modulation unit. The time convolution layer is used to extract the time characteristics of the latent variable of the current metabolic state. The temperature modulation unit adjusts the amplitude of the convolution result according to the ambient temperature to generate the oxygen metabolic consumption per unit time in the current state. The nitrogen displacement intensity inversion branch includes two fully connected layers connected in series and an attention weighting mechanism. The attention weighting mechanism assigns feature weights based on the coupling relationship between the nitrogen injection rate and the oxygen concentration change rate to generate an oxygen concentration change intensity factor caused by nitrogen injection in the current state. The pressure difference adaptation branch includes two residual convolution units and an airtightness adjustment factor. The residual convolution unit is used to extract multi-scale flow structure characteristics. The airtightness adjustment factor corrects the pressure difference estimation value according to the equipment sealing level and the structural parameters in the cabinet to output the pressure difference compensation value.

6. The method for adjusting nitrogen and oxygen concentration of cold chain gas conditioner based on deep neural network according to claim 1, characterized in that: The step three is specifically as follows: Inputting the oxygen potential disturbance field into an oxygen content prediction network, wherein the oxygen content prediction network includes a disturbance feature extraction module, a historical response memory module, and a sequence curve generation module; The disturbance feature extraction module includes multiple stacked two-dimensional convolutional layers and residual connection structures, receives the oxygen potential disturbance field, extracts the spatial features of the oxygen potential disturbance field, and outputs a spatial disturbance feature map; The historical response memory module includes a gated recurrent unit network, which receives a historical adjustment behavior parameter sequence and a corresponding historical oxygen concentration measurement sequence, extracts the temporal response characteristics between the historical adjustment behavior and the oxygen concentration, and outputs a historical response feature vector; The sequence curve generation module includes an attention fusion network and a cyclic decoding network. The attention fusion network is used to fuse the spatial disturbance feature map and the historical response feature vector to form a disturbance response feature. The cyclic decoding network decodes the disturbance response feature time-step by time step and outputs the oxygen concentration prediction value for each time step in the future target time window. The oxygen concentration prediction values ​​of each time step are sequentially combined into a complete sequence to generate an oxygen content prediction curve within a future target time window.

7. The method for adjusting nitrogen and oxygen concentration of cold chain gas conditioner based on deep neural network according to claim 1, characterized in that: The step 4 is specifically as follows: The decision calculation module identifies all time segments where the predicted oxygen concentration exceeds the upper and lower thresholds of the corresponding oxygen concentration for the cargo within the target time window based on the oxygen content prediction curve, and records the maximum deviation value of each exceeding limit segment. and duration of overrun ; According to the change in oxygen concentration caused by each unit input of the regulation behavior in the oxygen potential disturbance field, the disturbance response coefficient is defined as: ; in, represents the disturbance response coefficient of the i-th type regulatory behavior unit, Represents the unit adjustment amount of the i-th type of adjustment behavior The change in oxygen concentration caused by represents the unit adjustment amount of the i-th type of adjustment behavior; In each oxygen concentration exceeding limit range, the control input that meets the following optimization objectives is calculated: ; If multiple If the optimization goal is met, the one with the lowest energy consumption per unit input is selected; When the maximum deviation When the deviation from the preset early intervention threshold is less than the disturbance response coefficient, The maximum regulation behavior performs low-amplitude compensation operations to suppress the upward trend of oxygen concentration; If there is a continuous over-limit duration in the oxygen content prediction curve Greater than the preset overtime threshold or maximum deviation value If the maximum deviation is greater than the preset target upper limit, the oxygen reduction trigger state is set, the oxygen reduction trigger judgment value is set to 1, the nitrogen injection rate and fan frequency in the corresponding time period are specified to be the maximum adjustment value, and the air supply valve is closed; when the oxygen reduction trigger judgment value is set to 0, the oxygen reduction trigger state is not activated; Finally, the control parameter set is combined into an adjustment strategy sequence in chronological order, and the adjustment strategy sequence includes a nitrogen production set value, a fan frequency, an air supply valve opening, and an oxygen reduction trigger judgment value.

8. The method for adjusting nitrogen and oxygen concentration of cold chain gas conditioner based on deep neural network according to claim 1, characterized in that: The gas replenished by the air replenishing valve is air, which is used to increase the oxygen concentration.

9. The method for adjusting nitrogen and oxygen concentration of cold chain gas conditioner based on deep neural network according to claim 1, characterized in that: The step seven is specifically as follows: All feedback samples in the incremental training buffer are organized into several batches of training data in the order of their collection time, where each feedback sample contains a data sequence sampled in real time during a complete regulation cycle, a corresponding sequence of actually executed regulation strategies, and a final oxygen concentration measurement result; Using the existing weight parameters in the current multi-objective variational inversion network as initial parameters, inputting the training data batches into the multi-objective variational inversion network one by one, and determining the current loss function size by calculating the error between the output and the actual measurement result; An adaptive learning rate mechanism is used during training. After each training batch, the loss function is evaluated compared to the loss function of the previous batch. If the loss function continuously decreases at a rate less than the set deceleration threshold or even shows an upward trend, the current learning rate is reduced. If the loss function continues to decrease at a rate greater than the set decrease threshold, increase the current learning rate; If the rate of decrease of the loss function is equal to the set rate reduction threshold, the current learning rate is maintained unchanged; An early stopping condition is set during the training process. When the loss function of the multi-objective variational inversion network model changes less than the set lower limit after multiple batches of training, indicating that the performance improvement is no longer significant, the current round of fine-tuning is terminated. The latest multi-objective variational inversion network model weight parameters obtained after fine-tuning are used to replace the original weight parameters, and the incremental training cache is cleared at the same time, waiting for the accumulation of new feedback samples.

10. The method for adjusting nitrogen and oxygen concentration of cold chain gas conditioner based on deep neural network according to claim 1, characterized in that: The field image tracking and recovery mechanism is specifically as follows: If the deviation between the predicted output of the oxygen potential disturbance field and the measured oxygen concentration exceeds the dynamic tolerance threshold over multiple consecutive sampling periods, or if the output of the oxygen potential disturbance field is missing a value or decoding fails, it is identified as a disturbance inference anomaly. After identifying the disturbance inference anomaly, data samples containing oxygen potential disturbance field state labels are retrieved from the historical operation database. A joint similarity measurement function is constructed based on the spatial distribution structure of the oxygen potential disturbance field and the temporal characteristics of the corresponding regulation behavior. The current abnormal state is matched with the historical oxygen potential disturbance field state, and multiple historical oxygen potential disturbance field state samples with the best similarity are determined as candidate sets. Performing a weighted combination of the corresponding control parameters in the candidate set to generate emergency recovery control parameters for the current regulation cycle, wherein the emergency recovery control parameters include the nitrogen production set value, the fan frequency, and the air supply valve opening; Send the emergency recovery control parameters to each corresponding control object to replace the original adjustment strategy sequence for execution; During the execution process, the disturbance inference state and actual response data are collected. When the deviation between the predicted output of the oxygen potential disturbance field and the actual oxygen concentration measured value is less than or equal to the dynamic tolerance threshold, and there is no value gap or decoding failure in the output of the oxygen potential disturbance field, the original control path is restored and the field map tracking recovery process is ended.

Images