Method and system for storage control of multi-component gas parameters

Abstract

The application discloses a kind of multi-component gas parameter's storage control method and system, belong to automation control technical field, comprising: collection initial environmental parameter, obtain the stacking form image of storage material, identify stacking characteristic parameter, construct gradient pre-judgment model, to obtain gradient pre-judgment interval, and mark high-risk area;Construct hierarchical node monitoring network, and real-time acquisition each monitoring node's local parameter, combine total environmental mean value and identify abnormal deviation area;Coupling association is constructed, and predictive gradient intervention is carried out, the optimal airflow path is calculated and adjusted gas supply strategy, utilize dynamic impedance adaptation strategy, identify high impedance area, utilize multi-objective optimization, carry out differentiation regulation, to generate targeted regulation strategy;Set update time interval, calculate pre-judgment deviation rate, to carry out deviation correction, ensure that local parameter is stable in safety range, avoid the problem of out of control due to space heterogeneity and dynamic coupling, guarantee the stability of overall storage effect.

Description

Technical Field

[0001] This invention relates to a storage control method and system for multi-component gas parameters, belonging to the field of automation control technology. Background Technology

[0002] In the industrial production sector, especially in material storage, driven by the increasing demands for stable storage quality, control technologies have been continuously iterating and upgrading. Early technologies focused on static temperature and humidity control, relying on fixed parameter settings and single equipment to maintain a basic environment. This could only meet the simple storage requirements of preventing materials from becoming moldy or rotting, and could not adapt to the dynamic physiological changes of materials during the storage period. However, with the integration of sensor technology and automatic control concepts, the control dimension has expanded from single temperature and humidity to gas composition, forming a multi-parameter linkage mode of temperature, humidity, and gas. This has broken through the adaptation limitations of early technologies. At the same time, with the further advancement of intelligent technologies, by monitoring environmental parameters in real time and dynamically adjusting equipment operation, the impact of parameter fluctuations on material quality can be effectively reduced, ultimately providing reliable technical support for the long-term stable storage of various materials.

[0003] However, existing technologies do not consider the spatial heterogeneity and dynamic coupling characteristics of parameters in local fine-tuning environments, leading to frequent parameter runaway problems in local areas. Specifically, although existing solutions achieve zonal parameter setting, they still use the overall regional average as the basis for control, failing to deeply capture the subtle differences in local microenvironments. During material stacking and storage, such as multi-layer pallet stacking and bulk material stacking, the stack's internal structure is easily affected by airflow obstruction and the difficulty in rapid diffusion of local metabolic products, naturally forming a significant microenvironment gradient that deviates significantly from the overall regional average. In addition, parameter gradients caused by the superposition of multiple factors in local microenvironments, these objectively existing variables in actual production, will exacerbate the deviation of local parameters from preset thresholds. Control modes that rely on regional averages cannot intervene in a targeted manner, ultimately leading to continuous runaway of local parameters and affecting the stability of overall storage performance. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a storage control method and system for multi-component gas parameters. By constructing a gradient prediction model to predict the stacked microenvironment gradient and high-risk areas, then capturing local differences through layered node monitoring, followed by targeted regulation to adapt to dynamic coupling characteristics, and finally closed-loop correction of prediction and strategy, the invention solves the problems of relying on regional mean to ignore spatial heterogeneity and dynamic coupling leading to local loss of control.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for controlling the storage parameters of a multi-component gas, comprising:

[0007] Initial environmental parameters are collected, images of the stacked morphology of stored materials are obtained, stacking feature parameters are identified, a gradient prediction model is constructed to obtain the gradient prediction interval, and high-risk areas are marked.

[0008] A hierarchical node monitoring network is constructed, and local parameters of each monitoring node are collected in real time. The abnormal deviation areas are identified by combining the total environmental mean.

[0009] Construct coupling correlations and perform predictive gradient interventions to calculate the optimal airflow path and adjust the gas supply strategy. Utilize dynamic impedance matching strategies to identify high impedance regions and employ multi-objective optimization for differentiated regulation to generate targeted regulation strategies.

[0010] Set the update interval and calculate the prediction deviation rate to correct the deviation.

[0011] Specifically, the steps for identifying stacking feature parameters include:

[0012] Images of the stacked region from different angles are acquired and preprocessed to generate a stereo image set;

[0013] Feature points are extracted from each image in the stereo image set, feature point matching is performed, disparity map is calculated, and a three-dimensional point cloud model of the stacked area is generated by combining the calibration parameters of the camera device.

[0014] Set an interlayer gap threshold, and perform layer segmentation on the three-dimensional point cloud model to obtain the number of stacked layers;

[0015] For point clouds of two adjacent layers, calculate the height difference of each sampling point and obtain the vertical gap between layers;

[0016] Calculate the effective space volume of the tray. Based on the material information, calculate the theoretical volume occupied. Combined with the void ratio between fruits, calculate the filling density.

[0017] Specifically, the steps for obtaining the gradient prediction interval include:

[0018] Based on the priority matching order, metabolic characteristic data matching the currently stored material are screened from the material database and divided into three storage stages according to storage time to construct a stage metabolic model;

[0019] A metabolic stage trigger is embedded in the stage metabolic model so that when the storage time enters a new stage, the curve data of the corresponding stage is automatically called.

[0020] Set historical filtering conditions to filter matching samples from the historical database and generate matching association groups;

[0021] The initial number of adoptions, stacked feature parameters, and stage metabolic model parameters are used as input variables, and feature engineering is performed to generate the input feature set.

[0022] Specifically, the steps for obtaining the gradient prediction interval also include:

[0023] A gradient prediction model is constructed using a hybrid architecture that combines physical mechanisms and data-driven approaches. This model includes a physical layer, a metabolic layer, and a prediction layer to output gradient prediction values ​​for each region.

[0024] Based on the gradient prediction value and combined with the prediction error distribution of historical samples, Bayesian estimation is used to calculate the 95% confidence interval of the gradient for each region in order to generate the gradient prediction interval.

[0025] The stacked region is divided into grids, and the gradient prediction model is used to output the gradient prediction interval for each grid.

[0026] Set a safety deviation threshold. If the upper limit of any deviation interval is greater than the redundancy ratio of the safety deviation threshold, the corresponding grid is identified as a high-risk area and the risk type is marked.

[0027] Risk scores are calculated based on the ratio of the upper limit of the deviation range to the safety deviation threshold, and monitoring priorities are assigned based on the set thresholds.

[0028] Specifically, the steps for identifying abnormal deviation areas include:

[0029] Based on the coordinates of high-risk areas, monitoring priorities, and stacking feature parameters, the node density is determined by region to generate a hierarchical node monitoring network.

[0030] The regional data collection frequency is set based on monitoring priority, and a sudden change data collection mechanism is set to collect data synchronously based on each monitoring node.

[0031] The collected data is preprocessed to generate a local parameter set;

[0032] Set an overall update interval, collect overall environmental parameters inside the storage chamber, and combine them with local parameter sets to calculate local deviations and standardized deviation indices.

[0033] Specifically, the steps for identifying abnormal deviation areas also include:

[0034] Based on the gradient prediction interval, a reserved coefficient is introduced, and an anomaly detection threshold is set.

[0035] For each monitoring node, a single-node deviation exceeding the standard is determined. If the local deviation exceeds the anomaly determination threshold within two consecutive collection cycles, it is determined that the deviation exceeds the standard.

[0036] For any adjacent There are at least [number] monitoring nodes, if there are at least [number] monitoring nodes. If multiple monitoring nodes simultaneously exceed the deviation limit, then adjacent nodes will be... The area covered by each monitoring node is identified as an abnormal deviation area, and the area coordinates and deviation type are recorded.

[0037] Specifically, the steps of predictive gradient intervention include:

[0038] For each region of abnormal deviation, multi-dimensional features are extracted, and hierarchical normalization coefficients are calculated.

[0039] Call the weight coefficient group under different risk types, calculate the contribution weight of each risk type, and select the risk type with the largest contribution weight as the dominant factor;

[0040] The regulatory measures are classified according to their mechanisms of action, and a correlation mapping table is established based on the historical database.

[0041] Using the parameters of the regulatory means as input and the local deviation value as output, a coupling relationship model is constructed. Partial least squares regression is used to train the model, and metabolic stage triggers are embedded to automatically update the model.

[0042] Based on the safety deviation threshold, three levels of intervention targets are set, and the optimal control parameters are solved in stages using a dynamic programming algorithm.

[0043] During the optimization process, the deviation change curves of different parameter combinations are predicted by the coupling relationship model, and the optimal control parameter sequence that meets the three-level intervention objectives is selected.

[0044] Specifically, the steps of predictive gradient intervention also include:

[0045] Based on the aforementioned three-dimensional point cloud model and real-time impedance monitoring, the optimal airflow path is planned.

[0046] The latest impedance data is collected at the impedance update interval to update the optimal airflow path;

[0047] Collect the inlet and outlet pressures of the airflow in the optimal airflow path and calculate the real-time impedance;

[0048] Obtain the initial impedance, combine it with the impedance threshold coefficient, set the impedance division threshold, and once the real-time impedance is greater than the impedance division threshold, it is determined to be a high impedance region, and the region coordinates and impedance increment are recorded.

[0049] Set an incremental division threshold to classify the high impedance region into different levels and configure a hierarchical adaptation strategy.

[0050] Based on the deviation index and the contribution weight of the dominant factors, a comprehensive priority index is calculated, and the abnormal deviation areas are adjusted and prioritized to allocate equipment resources.

[0051] The optimized control parameters are converted into standardized execution instructions.

[0052] Specifically, the steps for deviation correction include:

[0053] Based on the three-level intervention objectives, an initial update interval is set, and dynamic triggering conditions are set to generate a dynamic update plan table to collect environmental parameters across the entire domain, perform preprocessing, and generate an actual gradient dataset.

[0054] The actual parameter values ​​are compared with the corresponding gradient prediction intervals to distinguish between out-of-target nodes and normal nodes;

[0055] Calculate the actual interlayer gradient difference, obtain the interlayer prediction range, and evaluate whether the gradient distribution meets expectations;

[0056] At the special level for abnormal areas, the target achievement rate is calculated for abnormal deviation areas after regulation.

[0057] Calculate the node exceedance rate, gradient deviation, and target non-achievement rate, and calculate the prediction deviation rate by weighted summation, and perform deviation classification and classification correction;

[0058] Based on the initial update interval, repeat the multi-dimensional verification and prediction deviation rate calculation. If the new prediction deviation rate is a slight deviation, the correction is deemed effective. If it is still a moderate or severe deviation, repeat the graded correction until the target is met or the emergency mechanism is triggered.

[0059] Calculate the stability index of global parameters, and combine it with the initial quality parameters to generate a storage effect evaluation report.

[0060] A storage control system for multi-component gas parameters includes: a feature acquisition module, a gradient prediction module, a stratified monitoring module, a control module, and a dynamic correction module;

[0061] The feature acquisition module is used to acquire initial environmental parameters, obtain images of the stacked morphology of stored materials, and identify stacked feature parameters.

[0062] The gradient prediction module is used to construct a gradient prediction model to obtain the gradient prediction interval and mark high-risk areas;

[0063] The hierarchical monitoring module is used to construct a hierarchical node monitoring network and collect local parameters of each monitoring node in real time to identify abnormal deviation areas.

[0064] The regulation module is used to construct coupling correlations and perform predictive gradient interventions to generate targeted regulation strategies.

[0065] The dynamic correction module is used to set the update time interval and calculate the prediction deviation rate in order to correct the deviation.

[0066] The beneficial effects of this invention are:

[0067] By recognizing stacking features from multiple angles and constructing a gradient prediction model based on material metabolism characteristics, the microenvironment gradient within the stack is captured in advance, and high-risk areas are marked. This solves the problem of insufficient prediction of spatial heterogeneity in existing technologies, providing targeted basis for subsequent monitoring and control, avoiding blindness. A hierarchical node monitoring network is constructed based on risk levels, and local parameters are collected in a targeted manner and compared with the overall mean to identify anomalies. This breaks through the limitation of relying on the overall mean of the region, accurately capturing subtle local differences. By quantifying the dominant factors, constructing a coupling model, dynamically planning the optimal path, and multi-objective optimization, predictive gradient intervention and differentiated control are achieved. This adapts to the dynamic coupling characteristics of material metabolism and airflow, solving the defect of relying on overall control that cannot provide targeted intervention. Through dynamic update verification and hierarchical correction, a closed-loop adaptation mechanism is formed to continuously eliminate the risk of local runaway, ensure that local parameters are stable within a safe range, avoid runaway problems caused by spatial heterogeneity and dynamic coupling, and ensure the stability of the overall storage effect. Attached Figure Description

[0068] Figure 1 This is a schematic diagram of a storage control method for multi-component gas parameters;

[0069] Figure 2 This is a flowchart of obtaining the gradient prediction interval in this invention;

[0070] Figure 3 This is a flowchart illustrating the process of identifying abnormal deviation regions in this invention;

[0071] Figure 4 This is a flowchart of the predictive gradient intervention in this invention. Detailed Implementation

[0072] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0073] Example 1

[0074] refer to Figures 1 to 4 As shown in the figure, this embodiment introduces a storage control method for multi-component gas parameters, including the following steps:

[0075] Step S1: Using basic sensors deployed in the storage chamber, the initial environmental parameters of the current scene are collected in real time, including the overall concentration, temperature and airflow rate of various gases in the chamber. At the same time, the image acquisition equipment in the storage chamber is used to obtain images of the stacking morphology of the stored materials. Combined with the basic information of the initially stored materials, such as the single fruit size and typical stacking characteristics of fresh walnuts, the stacking characteristic parameters are identified, such as the number of stacking layers, the size of the gap between layers and the stacking density. Combined with historical storage data and basic metabolic characteristics, a gradient prediction model is constructed to fit the superimposed influence of the respiratory metabolism and stacking morphology of the stored materials on local parameters. Based on the real-time stacking characteristic parameters and initial environmental parameters, the gradient prediction intervals of different levels and regions within the stacking area are obtained, and high-risk areas are marked. This provides a clear basis for the subsequent deployment of monitoring nodes for stratified monitoring and the direction of targeted regulation, avoiding blind regulation due to a lack of local gradient awareness.

[0076] Step S2: Based on the gradient prediction interval and high-risk area, layered monitoring nodes are arranged in the stacked structure of the stored materials to construct a layered node monitoring network. Local parameters of each monitoring node, such as local gas concentration and temperature, are collected in real time and compared with the total environmental mean. The deviation between the local parameters and the total environmental mean is calculated. At the same time, combined with the gradient prediction interval, abnormal deviation areas are identified, and the deviation location, deviation magnitude and corresponding influencing factors are clarified to make up for the deficiency of local differences being ignored due to the reliance on regional mean control.

[0077] Step S3: Based on the abnormal deviation areas and corresponding influencing factors, construct the coupling relationship between local parameters and control measures, and perform predictive gradient intervention. Through the optimal airflow path and coupling relationship, direct airflow is sent to the abnormal deviation areas, while adjusting the gas supply strategy of the corresponding areas. Gas composition compensation is performed upstream of the abnormal deviation areas in advance to prevent the expansion of parameter deviation from the source. At the same time, a dynamic impedance adaptation strategy is introduced to identify high impedance areas. In addition, the changes in the metabolic characteristics of the stored materials during the storage process are updated in real time, and a multi-objective optimization algorithm is used to coordinate the control priority of environmental factors. Differentiated control is carried out for the core deviations of different abnormal deviation areas to generate the final targeted control strategy. This ensures that the parameters of each abnormal deviation area are stable within the preset range, realizes differentiated control by region, and thus achieves the optimization of storage effect at the global level. This avoids the failure of local control due to the dynamic coupling characteristics of parameters and realizes the dynamic adaptation of local parameters and material requirements.

[0078] Step S4: Set the update time interval, compare the actual microenvironment gradient data collected by the hierarchical node monitoring network with the gradient prediction interval, calculate the prediction deviation rate, and correct the deviation to ensure that the microenvironment gradient prediction and control strategy always adapts to the material state and environmental changes, continuously eliminate the risk of local parameter runaway, and ensure the stability of the overall storage effect.

[0079] Specifically, the steps for identifying stacking feature parameters include:

[0080] Multiple cameras at different angles are deployed inside the storage compartment to simultaneously capture images of the stacked area of ​​the stored materials, obtaining multiple sets of images. Each set contains multiple consecutive frames, and each set of images is obtained through a camera at the same angle, ensuring the stability of the captured stacking pattern. The original images are preprocessed by using adaptive histogram equalization to enhance the contrast between the materials and the background, and Gaussian filtering to remove noise caused by reflections from the equipment inside the compartment, ensuring that the stacking edges and layers are clearly distinguishable. The inter-frame difference method is used to filter clear frames by the pixel difference between adjacent frames and remove blurry frames caused by slight shaking of the materials, generating a stereoscopic image set.

[0081] Using a scale-invariant feature transformation algorithm, feature points are extracted from each image in the stereo image set, such as the edge contour lines of stacked pallets and the boundary lines between adjacent layers of materials. Feature points from images at different angles are matched, and mismatched points are eliminated using a random sampling consensus algorithm, retaining valid matching pairs. Based on the valid matching pairs, a disparity map between images at different angles is calculated. Combined with the calibration parameters of the camera device, the three-dimensional coordinates of each feature point are calculated using the triangulation principle, and the coordinates are spatially stitched to generate a three-dimensional point cloud model of the stacked area. Here, the pallet is a container for storing materials, such as a plastic pallet, turnover box, or grid frame. Its specific shape does not affect the recognition logic and is only used as a reference benchmark for the stacking layer.

[0082] A region growing algorithm is used to segment the 3D point cloud model into layers. Using the z-axis as a reference, and based on typical interlayer gaps in the material's basic information, an interlayer gap threshold is set. Starting from the bottom of the point cloud, points within a continuous height range are divided into the same layer, and the process is gradually grown upwards until all point clouds are covered, thus obtaining the number of stacked layers. And number each layer in order from bottom to top, such as the first layer. layer, For point clouds of two adjacent layers, multiple sampling points are taken at each corner and center of each layer. The height difference of the corresponding sampling points on the z-axis is calculated, and the vertical gap between layers is obtained by averaging the coordinate differences.

[0083] Based on a 3D point cloud model, the boundary coordinates of each pallet layer are extracted. The length, width, and maximum height of the internal material are determined by point cloud clustering, and the effective space volume of the pallet is calculated. The maximum height of the internal material is the difference between the maximum z-value of the corresponding layer of point cloud and the z-value of the bottom of the pallet.

[0084] Based on the material information, the total weight and average weight of each fruit in each pallet layer are obtained to calculate the quantity of material in each layer. Combined with the average volume of each fruit in the material information, the theoretical volume occupied is calculated. Based on the porosity between fruits in typical stacking conditions in the material information, and combined with the effective space volume and theoretical volume occupied, the filling density is calculated. The stacking layer number, vertical gap between layers (including the vertical gap between the current layer and the layers above and below), effective space volume, and filling density are integrated to generate the stacking feature parameter set for the corresponding stacking layer, as shown in the following expression:

[0085]

[0086] In the formula, For the first Layer fill density, For the first The theoretical volume occupied by the layers For the first The effective spatial volume of the layer The porosity is determined through previous experiments and is the ratio of the volume of the voids between materials to the total volume of the stack.

[0087] Specifically, the steps for obtaining the gradient prediction interval include:

[0088] The system reads the material database stored in the storage terminal. The material database contains multiple data groups, which are associated with the material's unique identifier, material type, variety, initial quality parameters, and metabolic characteristic data. Based on a priority matching order, the system filters the material database for metabolic characteristic data that matches the currently stored material, including the generation rate curve of metabolic gases, oxygen consumption characteristic curve, and heat production rate curve. The priority matching order is as follows: the first priority is a complete match of the material's unique identifier. For example, if the current material identifier is "XL-01-Xiangling Walnut", the system directly matches the entries with the same identifier in the database. The second priority is if the identifier does not match, a weighted matching score is used to match step by step according to the material type, variety, and initial quality parameters. Only entries with a score greater than the preset matching lower limit are retained, and the entry with the highest score is used as the final matched entry.

[0089] The matched metabolic characteristic data were divided into three storage stages according to storage time: high metabolic period, stable period, and low metabolic period. Piecewise polynomial fitting was used for each type of curve to construct a stage metabolic model, which included staged expressions for metabolic gas generation, oxygen consumption, and heat production. At the same time, metabolic stage triggers were embedded in the model. When the storage time enters a new stage, the curve data of the corresponding stage is automatically called to ensure that the metabolic data is dynamically updated with the storage process.

[0090] Historical screening criteria are set, including material matching, environmental similarity, and stacking similarity. Based on these criteria, matching samples are selected from the historical storage database. Initial environmental parameters, stacking feature parameters, metabolic stages, and local parameter gradients are extracted from the matching samples to generate matching association groups. Material matching means that the material type and variety are consistent with the current material, such as both being Xiangling walnuts. Environmental similarity means that the environmental similarity between the historical initial environmental parameters and the current initial environmental parameters collected in real time is greater than the environmental similarity threshold, and the environmental similarity is calculated using cosine similarity. Stacking similarity means that the stacking similarity between the historical stacking feature parameters and the current stacking feature parameters is greater than the stacking similarity threshold, and the stacking similarity is calculated using Euclidean distance.

[0091] Using initial environmental adoption numbers, stacking feature parameters, and stage metabolic model parameters as input variables, feature engineering is performed to convert the stacking feature parameters into airflow drag coefficients and metabolic activity indices to generate spatial features. Stage identifier variables are generated through metabolic stage triggers as inputs to distinguish metabolic stages, such as 1 for high metabolic phase, 2 for stable phase, and 3 for low metabolic phase, to generate dynamic features. An interaction term between airflow drag and metabolic activity is constructed to quantify their coupling effect and generate interaction features, thereby generating the input feature set. Among them, the airflow drag coefficient is calculated by weighted summation of the average interlayer gap and average filling density. The metabolic activity index is calculated by multiplying the ratio of the initial generation rate of metabolic gas to the maximum initial generation rate of material in the stage metabolic model with the stage coefficient.

[0092] A gradient prediction model is constructed using a hybrid architecture combining physical mechanisms and data-driven approaches. This model comprises a physical layer, a metabolic layer, and a prediction layer. The physical layer simulates the internal airflow field and characteristic metabolic gas diffusion field based on computational fluid dynamics. Inputs include the airflow drag coefficient and initial airflow parameters from the initial environmental parameters. Outputs the airflow rate and gas diffusion coefficient for each region, reflecting the ease of gas diffusion. The metabolic layer couples the stage-based metabolic model with the physical layer results, calculating the generation and oxygen consumption of characteristic metabolic gases in each region and correcting the gas concentration field, including local metabolic gas concentration, local oxygen concentration, and local temperature. The prediction layer uses the intermediate results from the physical and metabolic layers, along with matching association groups, as input to train a gradient boosting tree (GBDT) model. It fits the intermediate results from the physical and metabolic layers and the matching association groups generated based on historical screening conditions. Through interpolation, it outputs the gradient prediction values ​​for each region, including oxygen deviation, metabolic gas deviation, and temperature deviation.

[0093] Based on the gradient prediction values ​​output by the prediction layer, combined with the prediction error distribution of historical samples, Bayesian estimation is used to calculate the 95% confidence interval of the gradient for each region, so as to generate the gradient prediction interval, including the oxygen deviation interval, metabolic gas deviation interval, and temperature deviation interval for each region.

[0094] Vertically, based on the number of stacking layers, and horizontally based on a preset horizontal division area, such as the size of a single fruit, the stacking area is divided into grids. Each grid is assigned a unique coordinate, and the current stacking feature parameters, initial environmental parameters, and metabolic stage are input into the gradient prediction model to output the gradient prediction interval for each grid.

[0095] Based on historical runaway samples, safety deviation thresholds are set for the current stored materials, including oxygen deviation threshold, metabolic deviation threshold, and temperature deviation threshold. For each grid, if the upper limit of any deviation interval is greater than the redundancy ratio of the safety deviation threshold, the corresponding grid is identified as a high-risk area. The risk type is labeled based on the airflow resistance coefficient and metabolic activity index. At the same time, a risk score is calculated based on the ratio of the upper limit of the deviation interval to the safety deviation threshold. Monitoring priorities are divided into three levels based on the set thresholds. The risk types are labeled as follows: if the airflow resistance coefficient is greater than the sum of the mean and the standard deviation, it is labeled as airflow-restricted; if the metabolic activity index is greater than the sum of the mean and the standard deviation, it is labeled as high-metabolic; if both are satisfied, it is labeled as coupled.

[0096] Specifically, the steps for identifying abnormal deviation areas include:

[0097] Based on the coordinates of high-risk areas, monitoring priorities, and stacking characteristic parameters, a node layout scheme was formulated, and the node density was determined for each region. For primary monitoring areas, following the principle of full coverage of the risk grid, one monitoring node was placed at the center of each high-risk grid. If adjacent primary monitoring areas exist, additional monitoring nodes were added at the overlapping edges to ensure no monitoring blind spots. For secondary monitoring areas, according to... The risk grid shares a density arrangement of monitoring nodes, which are located at the grid intersection center. For the three-level monitoring area, monitoring nodes are arranged along the edge contour line and central axis of each layer of stack, such as a total of 3 nodes per layer, to ensure basic perception of the overall parameter distribution. The sensors integrate oxygen, characteristic metabolic gas concentration detection and temperature detection to generate a hierarchical node monitoring network covering the entire stack area, which includes the location coordinates of the monitoring nodes, sensor deployment map and communication topology.

[0098] The sampling frequency for each region is set based on monitoring priority. For example, monitoring nodes in the first-level monitoring region use a high-frequency sampling interval to capture rapid changes in a timely manner; monitoring nodes in the second-level monitoring region use a secondary sampling frequency to balance accuracy and energy consumption; and monitoring nodes in the third-level monitoring region use a basic sampling frequency to meet the slow-change monitoring needs of low-risk areas. At the same time, a sudden change sampling mechanism is set up so that temporary sampling is automatically triggered once a sudden change occurs in the storage environment where the monitoring node is located, such as the opening of the hatch or the start and stop of the fan. The regional sampling frequencies are ordered from highest to lowest as high-frequency sampling frequency, secondary sampling frequency, and basic sampling frequency.

[0099] Each monitoring node synchronously collects data at a set frequency, including local oxygen concentration, characteristic metabolic gas concentration, and local temperature. The data includes a collection timestamp, layer number, and horizontal grid coordinates. The collected data is preprocessed, and a wavelet threshold denoising algorithm is used to eliminate sensor circuit noise and instantaneous interference caused by material swaying. The collected data from different monitoring nodes are aligned by timestamp to generate a parameter distribution matrix at the same time. Outliers are corrected by interpolation of neighboring monitoring nodes, thereby generating a local parameter set.

[0100] Based on the deployed base sensors, an overall update interval is set to collect overall environmental parameters within the storage chamber. Combined with the parameters of each monitoring node in the local parameter set, the local deviations are calculated by the difference between the monitoring node parameters and the overall environmental parameters. These deviations include oxygen deviation, characteristic metabolic gas deviation, and temperature deviation. Standardized deviation indices are then calculated. For gaseous parameters, such as oxygen or characteristic metabolic gases, the standard deviation of oxygen or characteristic metabolic gas is calculated by the ratio of the local deviation to the overall environmental parameters. For temperature, a safe temperature range is obtained, and the standard deviation of temperature is calculated by the ratio of the temperature deviation to the difference between the upper and lower limits. This information is then used to prioritize subsequent control measures.

[0101] Based on the gradient prediction interval, a reservation coefficient is set, and the product of the reservation coefficient and the upper limit of the interval is used as the reservation redundancy. The anomaly judgment threshold is set by the sum of the reservation redundancy and the upper limit of the region.

[0102] For each monitoring node, a single-node deviation exceeding the standard is determined. Once the local deviation exceeds the anomaly judgment threshold within two consecutive collection cycles, it is determined that the deviation exceeds the standard, including oxygen exceeding the standard, metabolic gas exceeding the standard, and temperature exceeding the standard.

[0103] For any adjacent There are at least [number] monitoring nodes, if there are at least [number] monitoring nodes. If multiple monitoring nodes simultaneously exceed the deviation limit, then adjacent nodes will be... The area covered by each monitoring node is identified as an abnormal deviation area, and the area coordinates and deviation type are recorded, including abnormal oxygen, abnormal metabolic gases, and abnormal temperature; in this embodiment, the area covered by each monitoring node is identified as an abnormal deviation area. , .

[0104] Specifically, the steps of predictive gradient intervention include:

[0105] For each abnormal deviation area, multi-dimensional features are extracted, including predicted spatial features, deviation features, stacking and metabolic features. Predicted spatial features include area coordinates, volume, and straight-line distance from the air supply port. Deviation features include local deviation values ​​of oxygen, metabolic gases, and temperature, standardized deviation index, and deviation duration (number of cycles exceeding the standard). Stacking and metabolic features include interlayer gap, filling density, airflow resistance coefficient, metabolic activity index, and current metabolic stage of the corresponding area, generating a multi-dimensional feature table for each abnormal deviation area.

[0106] Since the risk type classification is a predictive and qualitative classification, it does not consider the dynamic changes in actual regional storage. Furthermore, the contribution of factors cannot be quantified solely by risk type labels, which may lead to insufficient matching of control measures. Therefore, the analytic hierarchy process (AHP) is used to determine the contribution weight of each factor to the anomaly. The latest airflow resistance coefficient, metabolic activity index, and interaction characteristics are normalized to obtain hierarchical normalization coefficients, including the normalized airflow resistance coefficient, metabolic activity index, and interaction characteristics. Weight coefficient groups under different risk types are called, including three weight coefficient groups. By weighting the hierarchical normalization coefficients, the contribution weight of each risk type is obtained. The risk type with the largest contribution weight is selected as the dominant factor. For example, if the contribution weight calculated by the weight coefficient group of airflow-restricted type is the largest, then airflow-restricted type is selected as the dominant factor.

[0107] The control measures that can be implemented in the integrated storage system are classified according to their mechanism of action, including airflow control, gas supply, and temperature regulation. Among them, airflow control includes local fan speed regulation and directional air supply valve opening control; gas supply includes oxygen supply amount and characteristic metabolic gas adsorption / replacement rate; and temperature regulation includes regional heat sink power.

[0108] Based on the control cases recorded in the historical storage database, including the causes of anomalies and the effect data of corresponding control measures, an association mapping table containing the correspondence between dominant factors and control measures is established to match control measures with core issues. For example, when the dominant factor is airflow limitation, the mapping table recommends measures to enhance local airflow.

[0109] Using the parameters of the control measures as input and the local deviation value as output, a coupling relationship model is constructed to quantify the impact of the control measures on the local deviation. Partial least squares regression is used to train the model based on historical associated samples, enabling the model to predict the deviation correction effect under specific control parameters. At the same time, a metabolic stage trigger is embedded. When storage enters a new metabolic stage, the weight coefficients of the metabolic activity index in the model are automatically updated to ensure that the coupling relationship of the model can adapt to the dynamic changes of the material's metabolic characteristics.

[0110] For each abnormal deviation region, three levels of intervention targets are set based on the safety deviation threshold: emergency target, medium-term target, and long-term target. Using a dynamic programming algorithm, the optimal control parameters are solved in stages. The first stage is guided by the emergency target, constraining the control intensity to be no less than the preset minimum effective control threshold, with the optimization objective being the fastest deviation reduction rate. The medium- and long-term stages introduce energy consumption constraints, with the optimization objectives being the highest deviation stability and lowest energy consumption. Specifically, the emergency target requires reducing the deviation from the current excessive value to the first proportion (e.g., 80%) of the safety deviation threshold within a preset emergency period (e.g., 10 minutes) to quickly suppress deviation deterioration. The medium-term target requires reducing the deviation to the second proportion (e.g., 50%) of the safety deviation threshold within a preset medium-term period (e.g., 30 minutes) to stabilize the deviation trend. The long-term target requires maintaining the deviation within the third proportion (e.g., 30%) of the safety deviation threshold within a preset long-term period (e.g., 60 minutes) to achieve dynamic balance of local parameters. The target proportions, from largest to smallest, are the first proportion, the second proportion, and the third proportion.

[0111] During the optimization process, the deviation change curves of different parameter combinations are predicted by the coupling relationship model, that is, the deviation values ​​that change over time. The optimal control parameter sequence that meets the three-level intervention objectives is then selected as the optimal control parameters for each stage.

[0112] Based on a 3D point cloud model and real-time impedance monitoring, an improved A* algorithm is used to plan the optimal airflow path. By weighting the distance, average impedance, and number of turns, a path cost function is constructed. The smaller the cost function value, the better the path. The latest impedance data is collected at the impedance update interval, and the path is updated based on the new data. Temporary high-resistance areas are automatically avoided. Finally, a phased optimal control parameter sequence (including time nodes) and a dynamically updated optimal airflow path (including coordinate points and turning commands) are generated.

[0113] By using pressure sensors in the optimal airflow path, the inlet and outlet pressures of the airflow are collected. The real-time impedance is calculated by the ratio of the difference between the inlet and outlet pressures to the average wind speed along the path.

[0114] The initial impedance is obtained, and the impedance threshold coefficient is set. The impedance division threshold is obtained through multiplication. Once the real-time impedance is greater than the impedance division threshold, the corresponding abnormal deviation area is identified as a high impedance area, and the area coordinates and impedance increment are recorded simultaneously. The impedance increment is the difference between the real-time impedance and the impedance division threshold.

[0115] Set an incremental division threshold and classify high impedance areas into minor and significant areas based on impedance increments. Configure a graded adaptation strategy for high impedance areas. For minor areas, use pulsed air supply compensation to avoid continuous high power operation. For significant areas, activate the guide vane coordinated adjustment to increase the angle of the interlayer guide vanes on the path, reduce local resistance, and increase the fan speed to ensure that the wind speed reaching the abnormal deviation area is greater than the target value.

[0116] An adjustable supply port is set upstream of the optimal airflow path to release control gases, such as oxygen, in advance. These gases are carried by the airflow to the abnormal deviation area to achieve source compensation and reduce local concentration gradients. The calculation of the supply amount needs to take into account the gas loss in the transmission path, such as the amount adsorbed by the material. The transmission loss rate is fitted based on historical data and is positively correlated with the path length and the material adsorption characteristics. The longer the path and the stronger the material adsorption, the higher the loss rate. The target supply amount required for the abnormal deviation area is divided by (1 - transmission loss rate) to obtain the actual supply amount. At the same time, the injection angle of the supply port is adjusted by a stepper motor to ensure that the gas injection direction is consistent with the optimal airflow path and reduce diffusion loss.

[0117] Based on the standardized deviation index and the contribution weight of the dominant factors, a comprehensive priority index is calculated through weighted calculation. Based on the level of the comprehensive priority index, the control priority of abnormal deviation areas is ranked, and equipment resources are allocated according to priority, including first-level, second-level, and third-level priorities. First-level priority areas have exclusive dedicated control equipment, such as dedicated fans and feed pumps, to ensure rapid response. Second-level priority areas share general equipment, such as allocating running time for the same fan according to priority ratio. Third-level priority areas delay control until the parameters of higher priority areas stabilize before execution to avoid resource conflicts.

[0118] The optimized control parameters are converted into standardized execution instructions, including basic instructions, feedback instructions, and correction trigger conditions. The basic instructions include device ID, start time, duration, and core parameters. The feedback instructions include requests for feedback data from the device. The correction trigger conditions include ensuring that the control effect meets expectations if the deviation decrease rate of two consecutive feedbacks is less than 50% of the target value.

[0119] Specifically, the steps for deviation correction include:

[0120] Based on the time cycle of the three-level intervention target, an initial update interval is set to ensure that it matches the verification cycle of the control effect. Dynamic trigger conditions are set to initiate additional updates, thereby generating a dynamic update plan. The dynamic trigger conditions include: if the rate of deviation decrease fails to reach the target for two consecutive times in the feedback instruction, the cause of the control failure should be verified in a timely manner; new abnormal deviation areas should be identified and the matching degree between the newly added areas and the prediction should be verified simultaneously; and the metabolic stage trigger should be switched, such as from the high metabolic period to the stable period, the prediction may fail due to changes in the metabolic characteristics of materials, and the verification benchmark should be updated in a timely manner.

[0121] Based on the dynamic update schedule, at the update time point, the hierarchical node monitoring network collects the environmental parameters of the entire domain and performs preprocessing, including wavelet denoising, spatiotemporal alignment, and outlier correction, to generate an actual gradient dataset, which includes the parameter values ​​of each node, the gradient difference between layers (the difference in parameters at the same coordinate between adjacent layers), and the regional mean.

[0122] At the single-node level, the actual parameter value of each monitoring node is compared with the corresponding gradient prediction interval. By comparing with the upper and lower limits of the parameters, it is recorded whether the current monitoring node is within the interval, so as to distinguish between nodes that exceed the standard and normal nodes.

[0123] At the regional gradient level, for any two adjacent layers, the actual inter-layer gradient difference is calculated using the difference between the upper and lower layer parameters. Simultaneously, the gradient prediction intervals for these two adjacent layers are obtained. The upper and lower limits of the inter-layer gradient prediction are calculated using the difference between the upper and lower interval boundaries to obtain the inter-layer prediction range. By comparing the actual inter-layer gradient difference with the inter-layer prediction range, the gradient distribution is evaluated to determine if it meets expectations. Specifically, the upper prediction limit is the difference between the upper limit of the upper layer parameter gradient prediction interval and the lower limit of the lower layer parameter gradient prediction interval, and the lower prediction limit is the difference between the lower limit of the upper layer parameter gradient prediction interval and the upper limit of the lower layer parameter gradient prediction interval.

[0124] At the level of special projects in abnormal areas, for abnormal deviation areas after regulation, the target achievement rate is calculated by the ratio of the actual parameter value to the third-level intervention target value (such as the emergency target requirement to be reduced to 80% of the safety threshold).

[0125] The system obtains the number of nodes exceeding the gradient prediction range at a single node level, calculates the node exceedance rate by the ratio of the number of exceedance nodes to the total number of nodes, calculates the theoretical inter-layer gradient difference by the difference between the upper and lower limits of the inter-layer prediction range, calculates the gradient deviation by subtracting the theoretical inter-layer gradient difference from the actual inter-layer gradient difference, and calculates the target non-achievement rate by the target achievement ratio. The system then performs a weighted summation of the node exceedance rate, gradient deviation rate, and target non-achievement rate to calculate the prediction deviation rate. Finally, it classifies the deviation into minor, moderate, and severe deviations using a preset two-level deviation threshold. The sum of the target achievement ratio and the target non-achievement rate is equal to 1.

[0126] A tiered correction strategy is implemented for different deviation levels, including: For minor deviations, only the optimal control parameters for each stage are fine-tuned. Based on subtle differences between the actual gradient and the prediction, such as a local oxygen concentration slightly lower than the predicted lower limit, the control intensity is moderately adjusted in the corresponding region, such as increasing the oxygen supply, without changing the overall control logic and the predicted range; For moderate deviations, the core predicted parameters are corrected. For regions with concentrated deviations (such as several layers of metabolic gas gradients with high deviations), the upper and lower limits of the gradient prediction range for the region are expanded by 10%-20% according to the actual fluctuation range. At the same time, the anomaly detection threshold is updated synchronously, and the correlation between the actual gradient data and the control parameters is sampled. This paper incorporates the coupling relationship model for retraining to improve the model's prediction accuracy for this region. For severe deviations, the prediction and control system is reset, the gradient prediction model is backtracked and corrected, and the multi-factor coupling algorithm is called again. The latest metabolic activity index, airflow resistance coefficient, and actual gradient data are used as inputs to regenerate the global gradient prediction interval. For the region with the highest deviation rate, such as a certain level 3 monitoring area where data is missing due to too few nodes, 1-2 monitoring nodes are temporarily added along the central axis to increase the data collection density. Based on the new gradient prediction interval and node data, the multi-objective collaborative control process is re-executed to generate new control instructions adapted to the current state.

[0127] After performing the graded correction, the actual gradient data is collected again at the initial update interval. The multi-dimensional verification and prediction deviation rate calculation are repeated. If the new prediction deviation rate is a slight deviation, the correction is deemed effective. If it is still a moderate or severe deviation, the graded correction process is repeated until the target is met or the emergency mechanism is triggered.

[0128] Based on the corrected environmental data, the stability index of parameters across the entire region is calculated by the ratio of the standard deviation of the parameters to the mean. Combined with the initial quality parameters, such as material hardness and moisture content, a storage effect assessment report is generated, including the parameter compliance rate of each region (the percentage of nodes within the safe range), the predicted trend of material quality changes based on the correlation between parameter stability and metabolic stage, and key recommendations for the next stage of regulation (such as the need to strengthen oxygen supply in a certain region).

[0129] Example 2

[0130] Another embodiment of the present invention provides a storage control system for multi-component gas parameters, comprising: a feature acquisition module, a gradient prediction module, a stratified monitoring module, a control module, and a dynamic correction module;

[0131] The feature acquisition module is used to collect initial environmental parameters in the storage compartment and images of the stacked form of the stored materials from multiple angles. It preprocesses the images to generate a stereo image set, extracts feature points and matches them to generate a three-dimensional point cloud model of the stacked area. Combined with the basic information of the materials, it calculates the number of stacked layers, vertical gap between layers, effective space volume of the pallet and filling density to identify stacking feature parameters and avoid prediction deviations caused by fuzzy descriptions of stacking features.

[0132] The gradient prediction module is used to set priority matching rules, filter metabolic characteristic data that match the current material from the material database, construct a stage metabolic model with metabolic stage triggers according to storage time, generate an input feature set by combining initial environmental parameters, stack feature parameters and historical storage matching samples, construct a gradient prediction model using a hybrid architecture, output multi-component gas gradient prediction values ​​for each grid, generate gradient prediction intervals through Bayesian estimation, mark high-risk areas and divide monitoring priorities, capture the micro-environment gradient of multi-component gases inside the stack in advance, and solve the problem of neglecting spatial heterogeneity in existing technologies;

[0133] The hierarchical monitoring module is used to construct a hierarchical node monitoring network and set the acquisition frequency of multi-component gas and temperature parameters according to priority. It configures a temporary acquisition mechanism, synchronously acquires parameters of each node and preprocesses them to generate local parameter sets. Combined with the overall environmental parameters acquired by the basic sensors, it calculates local deviations and standardized deviation indices, accurately captures the differences between the local microenvironment and the whole, breaks through the monitoring limitations of relying on regional averages, and identifies abnormal deviation areas and records coordinates and deviation types by combining gradient prediction intervals. This ensures that anomaly identification takes into account both continuity and spatial correlation, and avoids control redundancy caused by misjudgment of a single node.

[0134] The control module is used to extract multi-dimensional features of abnormal areas to calculate the hierarchical normalization coefficient, call the weight coefficient group of the corresponding risk type to determine the dominant factors, classify the control measures according to their mechanism of action, establish an association mapping table based on historical data, construct a coupling relationship model, set three-level intervention targets and solve the optimal control parameter sequence through dynamic programming algorithm, combine the three-dimensional point cloud model and real-time impedance monitoring to plan and update the optimal airflow path, identify high impedance areas and configure hierarchical adaptation strategies, allocate equipment resources according to the comprehensive priority index, generate standardized execution instructions, adapt to the dynamic coupling characteristics of multi-component gases with metabolism and airflow, and achieve differentiated control.

[0135] The dynamic correction module is used to combine the three-level intervention target to set the initial update interval and dynamic triggering conditions, collect global parameters to generate an actual gradient dataset, compare the actual multi-component gas parameters with the gradient prediction interval, calculate the node exceedance rate, gradient deviation, and target non-achievement rate, and weight them to obtain the prediction deviation rate. Correction is performed according to the deviation level, and repeated verification is performed until the deviation reaches the target. The global parameter stability index is calculated and a storage effect evaluation report is generated to form a closed-loop control, continuously eliminate the risk of local runaway of multi-component gas, and ensure the stability of the overall storage effect.

[0136] Working principle and effects:

[0137] By collecting initial environmental parameters and material stacking images from the storage chamber, a 3D point cloud model is generated after preprocessing to identify stacking features. A gradient prediction model is constructed by combining material metabolic characteristics and historical data, outputting gradient intervals and marking high-risk areas to address the problem of insufficient prediction of spatial heterogeneity, providing targeted basis for subsequent control and avoiding blind regulation. A hierarchical node monitoring network is constructed according to risk level, and the collection frequency is set according to priority. Local parameters are collected in real time and preprocessed, and the deviation is calculated by comparing with the overall mean. Anomaly deviation areas are identified by combining the prediction interval, breaking through the limitation of relying on the overall mean and accurately capturing subtle local differences. Subsequently, the dominant factors of anomalies are quantified, and a coupled model of control measures is established. The optimal airflow path is dynamically planned, and multi-objective optimization generates targeted strategies that are adapted to the dynamic coupling characteristics of metabolism and airflow to achieve predictive intervention and avoid inefficient overall control. A dynamic update mechanism is used to verify the difference between actual and predicted results, and the model and strategy are corrected in stages to form a closed loop, continuously eliminating the risk of local loss of control and ensuring the stability of overall storage effect.

[0138] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for controlling the storage parameters of a multi-component gas, characterized in that, include: Initial environmental parameters are collected, images of the stacked morphology of stored materials are obtained, stacking feature parameters are identified, a gradient prediction model is constructed to obtain the gradient prediction interval, and high-risk areas are marked. The steps for obtaining the gradient prediction interval include: Based on the priority matching order, metabolic characteristic data matching the currently stored material are screened from the material database and divided into three storage stages according to storage time to construct a stage metabolic model; A metabolic stage trigger is embedded in the stage metabolic model so that when the storage time enters a new stage, the curve data of the corresponding stage is automatically called. Set historical filtering conditions to filter matching samples from the historical database and generate matching association groups; The initial environmental adoption count, stacked feature parameters, and stage metabolic model parameters are used as input variables, and feature engineering is performed to generate the input feature set. A gradient prediction model is constructed using a hybrid architecture that combines physical mechanisms and data-driven approaches. This model includes a physical layer, a metabolic layer, and a prediction layer to output gradient prediction values ​​for each region. Based on the gradient prediction value and combined with the prediction error distribution of historical samples, Bayesian estimation is used to calculate the 95% confidence interval of the gradient for each region in order to generate the gradient prediction interval. The stacked region is divided into grids, and the gradient prediction model is used to output the gradient prediction interval for each grid. Set a safety deviation threshold. If the upper limit of any deviation interval is greater than the redundancy ratio of the safety deviation threshold, the corresponding grid is identified as a high-risk area and the risk type is marked. Risk scores are calculated based on the ratio of the upper limit of the deviation range to the safety deviation threshold, and monitoring priorities are assigned based on the set thresholds. A hierarchical node monitoring network is constructed, and local parameters of each monitoring node are collected in real time. The abnormal deviation areas are identified by combining the total environmental mean. Construct coupling correlations and perform predictive gradient interventions to calculate the optimal airflow path and adjust the gas supply strategy. Utilize dynamic impedance matching strategies to identify high impedance regions and employ multi-objective optimization for differentiated regulation to generate targeted regulation strategies. Set the update interval and calculate the prediction deviation rate to correct the deviation.

2. The storage control method for multi-component gas parameters according to claim 1, characterized in that, The steps for identifying stacking feature parameters include: Images of the stacked region from different angles are acquired and preprocessed to generate a stereo image set; Feature points are extracted from each image in the stereo image set, feature point matching is performed, disparity map is calculated, and a three-dimensional point cloud model of the stacked area is generated by combining the calibration parameters of the camera device. Set an interlayer gap threshold, and perform layer segmentation on the three-dimensional point cloud model to obtain the number of stacked layers; For point clouds of two adjacent layers, calculate the height difference of each sampling point and obtain the vertical gap between layers; The effective volume of the pallet is calculated based on the material information, the theoretical volume occupied is calculated, and the filling density is calculated by combining the void ratio between fruits.

3. The storage control method for multi-component gas parameters according to claim 2, characterized in that, The steps for identifying abnormal deviation areas include: Based on the coordinates of high-risk areas, monitoring priorities, and stacking feature parameters, the node density is determined by region to generate a hierarchical node monitoring network. The regional data collection frequency is set based on monitoring priority, and a sudden change data collection mechanism is set to collect data synchronously based on each monitoring node. The collected data is preprocessed to generate a local parameter set; Set an overall update interval, collect overall environmental parameters inside the storage chamber, and combine them with local parameter sets to calculate local deviations and standardized deviation indices.

4. The storage control method for multi-component gas parameters according to claim 3, characterized in that, The steps for identifying abnormal deviation areas also include: Based on the gradient prediction interval, a reserved coefficient is introduced, and an anomaly detection threshold is set. For each monitoring node, a single-node deviation exceeding the standard is determined. If the local deviation exceeds the anomaly determination threshold within two consecutive collection cycles, it is determined that the deviation exceeds the standard. For any adjacent There are at least [number] monitoring nodes, if there are at least [number] monitoring nodes. If multiple monitoring nodes simultaneously exceed the deviation limit, then adjacent nodes will be... The area covered by each monitoring node is identified as an abnormal deviation area, and the area coordinates and deviation type are recorded.

5. The storage control method for multi-component gas parameters according to claim 4, characterized in that, The steps of predictive gradient intervention include: For each region of abnormal deviation, multi-dimensional features are extracted, and hierarchical normalization coefficients are calculated. Call the weight coefficient group under different risk types, calculate the contribution weight of each risk type, and select the risk type with the largest contribution weight as the dominant factor; The regulatory measures are classified according to their mechanisms of action, and a correlation mapping table is established based on the historical database. Using the parameters of the regulatory means as input and the local deviation value as output, a coupling relationship model is constructed. Partial least squares regression is used to train the model, and metabolic stage triggers are embedded to automatically update the model. Based on the safety deviation threshold, three levels of intervention targets are set, and the optimal control parameters are solved in stages using a dynamic programming algorithm. During the optimization process, the deviation change curves of different parameter combinations are predicted by the coupling relationship model, and the optimal control parameter sequence that meets the three-level intervention objectives is selected.

6. The storage control method for multi-component gas parameters according to claim 5, characterized in that, The steps of predictive gradient intervention also include: Based on the aforementioned three-dimensional point cloud model and real-time impedance monitoring, the optimal airflow path is planned. The latest impedance data is collected at the impedance update interval to update the optimal airflow path; Collect the inlet and outlet pressures of the airflow in the optimal airflow path and calculate the real-time impedance; Obtain the initial impedance, combine it with the impedance threshold coefficient, set the impedance division threshold, and once the real-time impedance is greater than the impedance division threshold, it is determined to be a high impedance region, and the region coordinates and impedance increment are recorded. Set an incremental division threshold to classify the high impedance region into different levels and configure a hierarchical adaptation strategy. Based on the deviation index and the contribution weight of the dominant factors, a comprehensive priority index is calculated, and the abnormal deviation areas are adjusted and prioritized to allocate equipment resources. The optimized control parameters are converted into standardized execution instructions.

7. The storage control method for multi-component gas parameters according to claim 6, characterized in that, The steps for deviation correction include: Based on the three-level intervention objectives, an initial update interval is set, and dynamic triggering conditions are set to generate a dynamic update plan table to collect environmental parameters across the entire domain, perform preprocessing, and generate an actual gradient dataset. The actual parameter values ​​are compared with the corresponding gradient prediction intervals to distinguish between out-of-target nodes and normal nodes; Calculate the actual interlayer gradient difference, obtain the interlayer prediction range, and evaluate whether the gradient distribution meets expectations; At the special level for abnormal areas, the target achievement rate is calculated for abnormal deviation areas after regulation. Calculate the node exceedance rate, gradient deviation, and target non-achievement rate, and calculate the prediction deviation rate by weighted summation, and perform deviation classification and classification correction; Based on the initial update interval, repeat the multi-dimensional verification and prediction deviation rate calculation. If the new prediction deviation rate is a slight deviation, the correction is deemed effective. If it is still a moderate or severe deviation, repeat the graded correction until the target is met or the emergency mechanism is triggered. Calculate the stability index of global parameters, and combine it with the initial quality parameters to generate a storage effect evaluation report.

8. A storage control system for multi-component gas parameters, used to implement the storage control method for multi-component gas parameters as described in any one of claims 1-7, characterized in that, include: Feature acquisition module, gradient prediction module, hierarchical monitoring module, control module, and dynamic correction module; The feature acquisition module is used to acquire initial environmental parameters, obtain images of the stacked morphology of stored materials, and identify stacked feature parameters. The gradient prediction module is used to construct a gradient prediction model to obtain the gradient prediction interval and mark high-risk areas; The hierarchical monitoring module is used to construct a hierarchical node monitoring network and collect local parameters of each monitoring node in real time to identify abnormal deviation areas. The regulation module is used to construct coupling correlations and perform predictive gradient interventions to generate targeted regulation strategies. The dynamic correction module is used to set the update time interval and calculate the prediction deviation rate in order to correct the deviation.

Images