Multi-area self-adaptive gas circulation AI physical fresh-keeping technology control system

Abstract

The invention relates to the technical field of fresh-keeping control, in particular to a multi-area self-adaptive gas circulation AI physical fresh-keeping technology control system, which comprises a first data acquisition module, a second data acquisition module and a control module, and is characterized in that the first data acquisition module is used for acquiring stored article parameters of the ith independent airtight area of the whole storage area, and the stored article parameters comprise article types and target fresh-keeping duration; the first strategy generation module is used for inputting the obtained stored article parameters into a pre-constructed first article fresh-keeping strategy model and outputting a first article fresh-keeping strategy; the second data acquisition module is used for acquiring cooking degree characteristic parameters of the stored articles; and the second strategy generation module outputs a second article fresh-keeping strategy. According to the control system, the stable state of the storage environment can be maintained, the gas component proportion can be accurately adjusted according to the real-time state of the articles, finally, the accurate matching of environmental parameters and the physiological needs of the articles is achieved, and the fresh-keeping self-adaptive capacity is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of preservation control technology, specifically to an AI physical preservation technology control system with multi-zone adaptive gas circulation. Background Technology

[0002] Existing food preservation technologies primarily employ constant atmosphere control (such as a set ratio of oxygen, carbon dioxide, and nitrogen) or a single refrigeration and insulation mode to regulate the environment, slowing down the metabolism of goods and inhibiting microbial growth, thereby achieving basic preservation. However, these traditional systems generally use static control methods, with most operating parameters remaining preset throughout the storage period, unable to be adjusted in real time according to the dynamic changes of the stored goods. More seriously, existing technologies generally lack the ability to perceive the evolution of the stored goods' state during preservation, failing to obtain crucial information such as whether the goods have begun to deteriorate and the rate of deterioration. Consequently, they cannot dynamically optimize control parameters such as atmosphere composition and temperature / humidity based on these feedback signals. Summary of the Invention

[0003] To address the aforementioned issues, this invention provides an AI physical preservation technology control system with multi-region adaptive gas circulation.

[0004] This invention employs the following technical solution: a multi-region adaptive gas circulation AI physical preservation technology control system, comprising: The first data acquisition module acquires the storage item parameters of the i-th independent airtight zone of the overall storage area, where i≥1. The storage item parameters include the item category and the target preservation time. The first strategy generation module inputs the acquired storage item parameters into the pre-built first item preservation strategy model and outputs the first item preservation strategy. The second data acquisition module acquires the maturity characteristic parameters of the stored items and obtains the maturity coefficient of the items based on the maturity characteristic parameters of the stored items. The second strategy generation module inputs the obtained maturity coefficient, the first item preservation strategy, and the stored item parameters into the pre-built second item preservation strategy model and outputs the second item preservation strategy. The preservation control module controls the parameter ratio, temperature, and humidity of the circulating gas in the i-th independent airtight zone based on the generated first and second item preservation strategies. The third data acquisition module acquires comprehensive environmental parameters of the i-th independent airtight zone in real time using a set of sensors based on a preset time interval. These comprehensive environmental parameters include ethylene concentration and TVOC gas content. Feature extraction is performed on the acquired comprehensive environmental parameters to obtain the ethylene concentration change rate. and the rate of change of TVOC gas content ; The strategy optimization module is based on the rate of change in ethylene concentration. and the rate of change of TVOC gas content Obtain the spoilage level of stored items, obtain the current second item preservation strategy, and trigger different atmosphere optimization strategies for different spoilage levels of stored items.

[0005] As a further description of the above technical solution: the first item preservation strategy includes storage temperature and storage humidity; The second item preservation strategy is a gas parameter ratio, which is the ratio of nitrogen, oxygen and carbon dioxide content.

[0006] As a further description of the above technical solution: the training method for the first item preservation strategy model includes: G sets of training data are collected in advance, where G is a positive integer greater than 0. The training data includes the item category and the target shelf life, as well as the first item preservation strategy corresponding to the item category and the target shelf life. A multi-output gradient boosting regression model was selected. The collected data is divided into training set, validation set and test set according to a preset ratio; The model is trained using the training set, the optimizer and learning rate are set, the model parameters are updated using the backpropagation algorithm, the loss function is minimized, and the model is validated using the validation set to achieve the best performance on the validation set. The trained model is evaluated using a test set, and the mean squared error index of the model is calculated to assess its performance. Once the model's performance evaluation meets the standards, the trained first-item preservation strategy model is deployed and applied.

[0007] As a further description of the above technical solution: the method for obtaining the maturity characteristic parameters of stored items includes: By installing industrial cameras on the inbound conveyor belt, image data of items being transported to the i-th independent airtight zone can be obtained; The acquired images are subjected to resolution unification, color correction, and exposure correction to obtain preprocessed images. Then, an image segmentation model is used to extract the object contours and generate feature images. Convert the feature image from the RGB color space to the HSV color space; Kernel density estimation is used to obtain the target color peaks in HSV images; Calculate and obtain the hue center value and standard deviation.

[0008] As a further description of the above technical solution: the method for obtaining the maturity coefficient of stored items based on maturity characteristic parameters includes: Define hue value A piecewise linear function to maturity score; Assign a weight to each hue value in turn; The maturity coefficient is calculated by weighting the maturity score and the set hue value.

[0009] As a further description of the above technical solution: the training method for the second item preservation strategy model includes: F sets of training data were collected in advance, where F is a positive integer greater than 0. The training data included maturity coefficient, storage temperature and storage humidity, and storage item parameters, as well as the second item preservation strategy corresponding to maturity coefficient, storage temperature and storage humidity, and storage item parameters. The F sets of training data are divided into a training set and a validation set, wherein the training set is used to train the second item preservation strategy model, and the validation set is used to evaluate the generalization performance of the second item preservation strategy model. During the training of the second item preservation strategy model, minimizing the cross-entropy loss function is used as the optimization objective. An early stopping strategy is used to monitor the performance of the validation set. The model performance is optimized by continuously adjusting the network parameters. When the prediction accuracy on the validation set reaches the expected accuracy, the second item preservation strategy model is considered to have converged and training is stopped. The second item preservation strategy model is trained using a deep neural network based on a multilayer perceptron. The maturity coefficient, storage temperature, storage humidity, and stored item parameters are converted into feature vectors. The input layer of the second item preservation strategy model receives the feature vectors and extracts nonlinear relationships in the data through multiple hidden layers. Finally, the output layer of the second item preservation strategy model calculates the probability distribution of gas parameter ratios through a softmax activation function and outputs the gas parameter ratio corresponding to the highest probability as the final prediction result.

[0010] As a further description of the above technical solution: the degradation levels include a first degradation level, a second degradation level, and a third degradation level, and the degree of degradation of the first degradation level, the second degradation level, and the third degradation level increases sequentially; Methods for obtaining the spoilage level of stored items include: Preset the threshold values ​​for the rate of change of ethylene concentration (Ymax) and the threshold values ​​for the rate of change of TVOC gas content (Tmax); when >Ymax, and When Tmax is greater than Tmax, the third mutation level is determined to be generated; when <Ymax, and When Tmax is less than Tmax, the first mutation level is determined to be generated; The remaining cases are judged to generate a second deterioration level.

[0011] As a further description of the above technical solution: The current second item preservation strategy is obtained, and different atmosphere optimization strategies are triggered for different spoilage levels of stored items, including: When the spoilage level of stored items is the first spoilage level, no atmosphere optimization strategy is generated; When the spoilage level of the stored goods is the second or third spoilage level, the second item preservation strategy and the spoilage level of the stored goods are input into the pre-built gas parameter optimization model, and the corrected gas parameter ratio is output as the atmosphere optimization strategy.

[0012] As a further description of the above technical solution: the training method of the gas parameter optimization model includes: Q sets of training data are collected in advance, where Q is a positive integer greater than 0. The training data includes the second item preservation strategy and the spoilage level of the stored item, as well as the gas parameter ratios corresponding to the second item preservation strategy and the spoilage level of the stored item. A deep learning model was selected as the gas parameter optimization model. Training data was used to train the gas parameter optimization model. The second item preservation strategy and the spoilage level of the stored items were used as inputs to the gas parameter optimization model, and the gas parameter ratio was used as the output. Stochastic gradient descent was employed, and the weights and biases of the gas parameter optimization model were adjusted using the backpropagation algorithm to minimize the error between the predicted and actual results. A loss function, the mean squared error, was set. When the loss function value converged, the training of the gas parameter optimization model was stopped, and the gas parameter optimization model corresponding to the convergence of the loss function value was taken as the trained gas parameter optimization model.

[0013] As a further description of the above technical solution: the sensor group includes an optical ethylene sensor and a PID-type TVOC sensor. The optical ethylene sensor is used to collect the ethylene concentration in the i-th independent airtight zone, and the PID-type TVOC sensor is used to collect the TVOC gas content in the i-th independent airtight zone.

[0014] Beneficial effects: In the above technical solution, the control system achieves synergistic optimization of preservation efficiency through deep coupling of two strategies. The first strategy generation module generates basic temperature and humidity parameters based on the item parameters and the target preservation period to construct a static storage framework. The second strategy generation module, based on the constructed static storage framework and relying on real-time visual maturity analysis, dynamically analyzes the changes in the physiological state of the item, obtains the maturity coefficient, and inputs the maturity coefficient, the first item preservation strategy, and the stored item parameters into the neural network model. The two strategies form a closed-loop linkage—the basic temperature and humidity setting provides a stable environmental benchmark for maturity monitoring, while the maturity feedback drives the dynamic correction of the gas ratio. This enables the control system to maintain the steady state of the storage environment and accurately adjust the gas component ratio according to the real-time state of the item, ultimately achieving precise matching between environmental parameters and the physiological needs of the item, and significantly improving the self-adaptive capability of preservation. Furthermore, by introducing a third data acquisition module and a strategy optimization module, the ethylene concentration and TVOC content in each independent airtight zone can be collected in real time based on a set time interval, and their rate of change can be calculated. This accurately reflects the deterioration trend of stored goods. By constructing a deterioration level judgment, the dynamic change characteristics of ethylene and TVOC are mapped to three deterioration levels, realizing the scientific classification and identification of the preservation status of goods. Combined with the second goods preservation strategy and the gas parameter optimization model constructed by deep learning, a corrected gas ratio is generated under the second or third deterioration level, and the storage atmosphere is dynamically adjusted. This method not only improves the system's early perception ability and response speed of deterioration signs, avoiding control lag or excessive intervention, but also realizes the differentiation and precision of atmosphere adjustment, effectively extending the shelf life of goods, reducing energy consumption, and significantly enhancing the adaptability, intelligence and stability of the overall preservation system. Attached Figure Description

[0015] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a module connection diagram of the AI ​​physical preservation technology control system with multi-region adaptive gas circulation provided in Embodiment 1 of the present invention; Figure 2 This is a module connection diagram of the AI ​​physical preservation technology control system with multi-region adaptive gas circulation provided in Embodiment 2 of the present invention; Figure 3 This is a flowchart of the AI ​​physical preservation technology control method with multi-region adaptive gas circulation provided in Embodiment 3 of the present invention; Figure 4 This is a flowchart of a method for obtaining the maturity characteristic parameters of stored items according to Embodiment 1 of the present invention. Detailed Implementation

[0016] To make the technical means, creative features, objectives, and effects of this invention readily understandable, the invention is further described below with reference to specific illustrations. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0017] Example 1 Please see Figure 1 and Figure 4 This invention provides a technical solution: an AI physical preservation technology control system with multi-region adaptive gas circulation, comprising a first data acquisition module, a first strategy generation module, a second data acquisition module, a second strategy generation module, and a preservation control module, wherein the modules are connected via wired and / or wireless means. The first data acquisition module acquires the storage item parameters of the i-th independent airtight zone of the overall storage area, where i≥1. The storage item parameters include the item category and the target preservation time. It should be noted that the overall storage area has M independent airtight zones, where M is a positive integer greater than 1. The M independent airtight zones are isolated from each other by sealed doors or air curtains, and each independent airtight zone is equipped with an air inlet and an air outlet.

[0018] It should be noted that the item categories include leafy green vegetables, apples, strawberries, shiitake mushrooms, etc. The item category and target shelf life are manually entered by the user through the operation interface. The first strategy generation module inputs the acquired storage item parameters into the pre-built first item preservation strategy model and outputs the first item preservation strategy, which includes storage temperature and storage humidity. The training method for the first item preservation strategy model includes: G sets of training data are collected in advance, where G is a positive integer greater than 0. The training data includes the item category and the target shelf life, as well as the first item preservation strategy corresponding to the item category and the target shelf life. A multi-output gradient boosting regression model was selected. The collected data is divided into training set, validation set and test set according to a preset ratio; The model is trained using the training set, the optimizer and learning rate are set, the model parameters are updated using the backpropagation algorithm, the loss function is minimized, and the model is validated using the validation set to achieve the best performance on the validation set. The trained model is evaluated using a test set, and the mean squared error index of the model is calculated to assess its performance. Once the model's performance evaluation meets the standards, the trained first-item preservation strategy model is deployed and applied.

[0019] The second data acquisition module acquires the maturity characteristic parameters of the stored items and obtains the maturity coefficient of the items based on the maturity characteristic parameters of the stored items. The method for obtaining the maturity characteristic parameters of stored items includes: By installing industrial cameras on the inbound conveyor belt, image data of items being transported to the i-th independent airtight zone can be obtained; The acquired images are subjected to resolution unification, color correction, and exposure correction to obtain preprocessed images. Then, an image segmentation model (such as Mask R-CNN) is used to extract the contours of the objects and generate feature images. It should be noted that the pre-trained Mask R-CNN model is used. The input is a pre-processed image, and the model outputs a binary mask of the object. The mask identifies the object region (1 for the foreground and 0 for the background). The feature image is the pixel-wise product of the mask and the original image, retaining only the object region.

[0020] Convert the feature image from the RGB color space to the HSV color space; Kernel density estimation is used to obtain the target color peaks in HSV images; Methods for obtaining the target color peak include: Collect the H values ​​of all non-zero pixels in the HSV image. ,in This represents the total number of pixels in the item area. For the first The H value of a non-zero pixel.

[0021] The kernel density is used to estimate the target color peak. The kernel density formula is as follows: ; in, ; ; In the formula, For hue value Density estimation at [location] For bandwidth parameters, It is a circular distance function. The Gaussian kernel function; Peak detection, in Find all peak values ​​within the range that satisfy the following conditions : ; in, To search for strides, the options are: ; , This is a constant, and can be set to 1.5. for The mean, for Standard deviation; Calculate the hue center value and standard deviation; It should be noted that the hue center value can be used to determine the maturity stage, while the standard deviation is used to determine the uniformity of maturity.

[0022] Hue center value ; For the first The hue center value (in degrees) of each primary color represents the average hue angle of that primary color. The sequence number of the primary color peak (e.g.) =1 indicates the first primary color. =2 indicates the second primary color, etc. The first one found by kernel density estimation There are several peak positions (unit: °), ranging from [0°, 360°); The formula for calculating the standard deviation is: ; In the formula, For the first The standard deviation of each primary color (in degrees) is used to quantify the dispersion of the color distribution; A smaller value indicates that the color is pure and concentrated; A large value indicates that the color is scattered and mixed; For the neighborhood The number of effective pixels within. Therefore The ring-shaped neighborhood centered on the center, The hue center value of the current primary color. For the neighborhood Inner The hue value of each pixel; Methods for obtaining the maturity coefficient of stored items based on their maturity characteristic parameters include: Define hue value A piecewise linear function to maturity score; In the formula The maturity score is... The immature threshold, This is the maturity threshold; A hue value weight is assigned to each hue value sequentially, and the method for assigning the hue value weight includes: In the formula For the first The weight of the hue value of each primary color ; This is a sensitivity parameter, which is optional. =20°, For the first The standard deviation of the primary color; , Neighborhood Intra-pixel count, This represents the total number of pixels.

[0023] The maturity coefficient is calculated by weighting the maturity score and the set hue value. The formula for calculating the maturity coefficient is as follows: ; This is the maturity coefficient. The number of primary color peaks detected The first The hue center value of the primary color, The maturity score of this main color. The first The weight of the hue value of the primary color; The second strategy generation module inputs the acquired maturity coefficient, the first item preservation strategy, and the stored item parameters into the pre-constructed second item preservation strategy model and outputs the second item preservation strategy. The second item preservation strategy is a gas parameter ratio, which is the ratio of nitrogen, oxygen, and carbon dioxide content. The training method for the second item preservation strategy model includes: F sets of training data were collected in advance, where F is a positive integer greater than 0. The training data included maturity coefficient, storage temperature and storage humidity, and storage item parameters, as well as the second item preservation strategy corresponding to maturity coefficient, storage temperature and storage humidity, and storage item parameters. The F sets of training data are divided into a training set and a validation set, wherein the training set is used to train the second item preservation strategy model, and the validation set is used to evaluate the generalization performance of the second item preservation strategy model. During the training of the second item preservation strategy model, minimizing the cross-entropy loss function is used as the optimization objective. An early stopping strategy is used to monitor the performance of the validation set. The model performance is optimized by continuously adjusting the network parameters. When the prediction accuracy on the validation set reaches the expected accuracy, the second item preservation strategy model is considered to have converged and training is stopped. The second item preservation strategy model is trained using a deep neural network based on a multilayer perceptron. The maturity coefficient, storage temperature, storage humidity, and stored item parameters are converted into feature vectors. The input layer of the second item preservation strategy model receives the feature vectors and extracts nonlinear relationships in the data through multiple hidden layers. Finally, the output layer of the second item preservation strategy model calculates the probability distribution of gas parameter ratios through a softmax activation function and outputs the gas parameter ratio corresponding to the highest probability as the final prediction result.

[0024] The preservation control module controls the parameter ratio, temperature, and humidity of the circulating gas in the i-th independent airtight zone based on the generated first and second item preservation strategies. In this embodiment, the control system achieves synergistic optimization of preservation efficiency through deep coupling of two strategies. The first strategy generation module generates basic temperature and humidity parameters based on item parameters and target preservation period to construct a static storage framework. The second strategy generation module, based on the constructed static storage framework and relying on real-time visual maturity analysis, dynamically analyzes changes in the physiological state of the items, obtains the maturity coefficient, and inputs the maturity coefficient, along with the first item preservation strategy and the stored item parameters, into a neural network model. The two strategies form a closed-loop linkage—the basic temperature and humidity setting provides a stable environmental benchmark for maturity monitoring, while the maturity feedback drives dynamic correction of the gas ratio. This enables the control system to maintain a steady state of the storage environment and accurately adjust the gas component ratio according to the real-time state of the items, ultimately achieving precise matching between environmental parameters and the physiological needs of the items, significantly improving the adaptive preservation capability.

[0025] Example 2 Please see Figure 2 This embodiment adds a third data acquisition module and a strategy optimization module based on the above embodiments; The third data acquisition module acquires comprehensive environmental parameters of the i-th independent airtight zone in real time based on a preset time interval using a set of sensors. The comprehensive environmental parameters include ethylene concentration and TVOC gas content. Feature extraction is performed on the acquired comprehensive environmental parameters to obtain the ethylene concentration change rate and TVOC gas content change rate. The sensor set includes an optical ethylene sensor and a PID-type TVOC sensor. The optical ethylene sensor is used to collect the ethylene concentration of the i-th independent airtight zone, and the PID-type TVOC sensor is used to collect the TVOC gas content of the i-th independent airtight zone.

[0026] The methods for obtaining the ethylene concentration change rate and the TVOC gas content change rate include: the calculation formula for the ethylene concentration change rate is: In the formula, The rate of change of ethylene concentration. For the detection time interval, This represents the difference between the current measured ethylene concentration and the previous measured ethylene concentration. The formula for calculating the change rate of TVOC gas content is: In the formula, The change rate of TVOC gas content For the detection time interval, This represents the difference between the current TVOC gas content and the previous TVOC gas content. The strategy optimization module is based on the rate of change in ethylene concentration. and the rate of change of TVOC gas content Get the spoilage level of stored items; get the current second item preservation strategy, and trigger different atmosphere optimization strategies for different spoilage levels of stored items; The degradation levels include a first degradation level, a second degradation level, and a third degradation level, and the degree of degradation increases sequentially from the first degradation level to the second degradation level and the third degradation level. Methods for obtaining the spoilage level of stored items include: Preset the threshold values ​​for the rate of change of ethylene concentration (Ymax) and the threshold values ​​for the rate of change of TVOC gas content (Tmax); when >Ymax, and When Tmax is greater than Tmax, the third mutation level is determined to be generated; when <Ymax, and When Tmax is less than Tmax, the first mutation level is determined to be generated; The remaining cases are judged to generate a second deterioration level.

[0027] In this embodiment, ethylene concentration and TVOC content are introduced as gaseous indicators of the deterioration of stored goods. This helps to promptly capture the key volatile organic compounds released by stored fruits and vegetables during the ripening-overripe-deterioration process. Compared with the traditional one-way decision-making limitation that relies solely on the maturity coefficient and the initial preservation strategy model, the use of preset thresholds to jointly judge the change rate of ethylene and TVOC scientifically classifies three deterioration levels, further refining the granularity of storage status identification.

[0028] The methods for obtaining the current second item preservation strategy and triggering different atmosphere optimization strategies for different spoilage levels of stored items include: When the spoilage level of stored items is the first spoilage level, no atmosphere optimization strategy is generated; When the spoilage level of the stored goods is the second or third spoilage level, the second item preservation strategy and the spoilage level of the stored goods are input into the pre-built gas parameter optimization model, and the corrected gas parameter ratio is output as the atmosphere optimization strategy.

[0029] The training method for the gas parameter optimization model includes: Q sets of training data are collected in advance, where Q is a positive integer greater than 0. The training data includes the second item preservation strategy and the spoilage level of the stored item, as well as the gas parameter ratios corresponding to the second item preservation strategy and the spoilage level of the stored item. A deep learning model was selected as the gas parameter optimization model. Training data was used to train the gas parameter optimization model. The second item preservation strategy and the spoilage level of the stored items were used as inputs to the gas parameter optimization model, and the gas parameter ratio was used as the output. Stochastic gradient descent was employed, and the weights and biases of the gas parameter optimization model were adjusted using the backpropagation algorithm to minimize the error between the predicted and actual results. A loss function, the mean squared error, was set. When the loss function value converged, the training of the gas parameter optimization model was stopped, and the gas parameter optimization model corresponding to the convergence of the loss function value was taken as the trained gas parameter optimization model.

[0030] In this embodiment, by introducing a third data acquisition module and a strategy optimization module, the ethylene concentration and TVOC content in each independent airtight zone can be collected in real time based on a set time interval, and their rate of change can be calculated. This accurately reflects the deterioration trend of stored goods. By constructing a deterioration level judgment, the dynamic change characteristics of ethylene and TVOC are mapped to three deterioration levels, realizing the scientific classification and identification of the preservation status of goods. Combined with the second goods preservation strategy and the gas parameter optimization model constructed by deep learning, a corrected gas ratio is generated under the second or third deterioration level, and the storage atmosphere is dynamically adjusted. This method not only improves the system's early perception ability and response speed of deterioration signs, avoiding control lag or excessive intervention, but also realizes the differentiation and precision of atmosphere adjustment, effectively extending the shelf life of goods, reducing energy consumption, and significantly enhancing the adaptability, intelligence and stability of the overall preservation system.

[0031] Example 3 Please see Figure 3 This invention provides a technical solution: an AI physical preservation technology control method with multi-region adaptive gas circulation, comprising: Obtain the storage item parameters for the i-th independent airtight zone of the overall storage area. The storage item parameters include the item category and the target preservation time. The item category includes leafy green vegetables, apples, strawberries, shiitake mushrooms, etc. The item category and the target preservation time are manually input by the user through the operation interface. The acquired storage item parameters are input into the pre-built first item preservation strategy model, and the first item preservation strategy is output; the first item preservation strategy includes storage temperature and storage humidity; Obtain the maturity characteristic parameters of the stored items, and obtain the maturity coefficient of the items based on the maturity characteristic parameters of the stored items; The obtained maturity coefficient, the first item preservation strategy, and the stored item parameters are input into the pre-constructed second item preservation strategy model, and the second item preservation strategy is output; the second item preservation strategy is the gas parameter ratio, which is the ratio of nitrogen, oxygen and carbon dioxide content. Based on the generated first and second item preservation strategies, the parameters, ratio, temperature, and humidity of the circulating gas in the i-th independent airtight zone are controlled. Based on a preset time interval, the comprehensive environmental parameters of the i-th independent airtight zone are acquired in real time through a set of sensors. The comprehensive environmental parameters include ethylene concentration and TVOC gas content. Feature extraction is performed on the acquired comprehensive environmental parameters to obtain the ethylene concentration change rate and TVOC gas content change rate. The spoilage level of stored items is obtained based on the rate of change of ethylene concentration and the rate of change of TVOC gas content. The current preservation strategy for the second item is then determined, and different atmosphere optimization strategies are triggered for different spoilage levels of stored items.

[0032] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. AI physical preservation technology control system of multi-zone adaptive gas circulation, characterized in that, include: The first data acquisition module acquires the storage item parameters of the i-th independent airtight zone of the overall storage area, where i≥1. The storage item parameters include the item category and the target preservation time. The first strategy generation module inputs the acquired storage item parameters into the pre-built first item preservation strategy model and outputs the first item preservation strategy. The second data acquisition module acquires the maturity characteristic parameters of the stored items and obtains the maturity coefficient of the items based on the maturity characteristic parameters of the stored items. The second strategy generation module inputs the obtained maturity coefficient, the first item preservation strategy, and the stored item parameters into the pre-built second item preservation strategy model and outputs the second item preservation strategy. The preservation control module controls the parameter ratio, temperature, and humidity of the circulating gas in the i-th independent airtight zone based on the generated first and second item preservation strategies. The third data acquisition module acquires the comprehensive environmental parameters of the i-th independent airtight area in real time based on a preset time interval through the set sensor group, the comprehensive environmental parameters including ethylene concentration and TVOC gas content, and the comprehensive environmental parameters acquired are subjected to feature extraction to acquire the ethylene concentration change rate and the TVOC gas content change rate . A policy optimization module based on the rate of change of ethylene concentration and the rate of change of TVOC gas content Obtain the deterioration level of the stored goods, obtain the current second goods preservation policy, trigger different atmosphere optimization strategies for different deterioration levels of the stored goods.

2. The AI ​​physical preservation technology control system with multi-zone adaptive gas circulation according to claim 1, characterized in that, The first item preservation strategy includes storage temperature and storage humidity; The second item preservation strategy is a gas parameter ratio, which is the ratio of nitrogen, oxygen and carbon dioxide content.

3. The AI ​​physical preservation technology control system with multi-zone adaptive gas circulation according to claim 1, characterized in that, The training method for the first item preservation strategy model includes: G sets of training data are collected in advance, where G is a positive integer greater than 0. The training data includes the item category and the target shelf life, as well as the first item preservation strategy corresponding to the item category and the target shelf life. A multi-output gradient boosting regression model was selected. The collected data is divided into training set, validation set and test set according to a preset ratio; The model is trained using the training set, the optimizer and learning rate are set, the model parameters are updated using the backpropagation algorithm, the loss function is minimized, and the model is validated using the validation set to achieve the best performance on the validation set. The trained model is evaluated using a test set, and the mean squared error index of the model is calculated to assess its performance. Once the model's performance evaluation meets the standards, the trained first-item preservation strategy model is deployed and applied.

4. The AI ​​physical preservation technology control system with multi-region adaptive gas circulation according to claim 1, characterized in that, The method for obtaining the maturity characteristic parameters of stored items includes: By installing industrial cameras on the inbound conveyor belt, image data of items being transported to the i-th independent airtight zone can be obtained; The acquired images are subjected to resolution unification, color correction, and exposure correction to obtain preprocessed images. Then, an image segmentation model is used to extract the object contours and generate feature images. Convert the feature image from the RGB color space to the HSV color space; Kernel density estimation is used to obtain the target color peaks in HSV images; Calculate and obtain the hue center value and standard deviation.

5. The AI ​​physical preservation technology control system with multi-zone adaptive gas circulation according to claim 4, characterized in that, The method for obtaining the maturity coefficient of stored items based on maturity characteristic parameters includes: Define hue value A piecewise linear function to maturity score; Assign a weight to each hue value in turn; The maturity coefficient is calculated by weighting the maturity score and the set hue value.

6. The AI ​​physical preservation technology control system with multi-zone adaptive gas circulation according to claim 2, characterized in that, The training method for the second item preservation strategy model includes: F sets of training data were collected in advance, where F is a positive integer greater than 0. The training data included maturity coefficient, storage temperature and storage humidity, and storage item parameters, as well as the second item preservation strategy corresponding to maturity coefficient, storage temperature and storage humidity, and storage item parameters. The F sets of training data are divided into a training set and a validation set, wherein the training set is used to train the second item preservation strategy model, and the validation set is used to evaluate the generalization performance of the second item preservation strategy model. During the training of the second item preservation strategy model, minimizing the cross-entropy loss function is used as the optimization objective. An early stopping strategy is used to monitor the performance of the validation set. The model performance is optimized by continuously adjusting the network parameters. When the prediction accuracy on the validation set reaches the expected accuracy, the second item preservation strategy model is considered to have converged and training is stopped. The second item preservation strategy model is trained using a deep neural network based on a multilayer perceptron. The maturity coefficient, storage temperature, storage humidity, and stored item parameters are converted into feature vectors. The input layer of the second item preservation strategy model receives the feature vectors and extracts nonlinear relationships in the data through multiple hidden layers. Finally, the output layer of the second item preservation strategy model calculates the probability distribution of gas parameter ratios through a softmax activation function and outputs the gas parameter ratio corresponding to the highest probability as the final prediction result.

7. The AI ​​physical preservation technology control system with multi-zone adaptive gas circulation according to claim 1, characterized in that, The degradation levels include a first degradation level, a second degradation level, and a third degradation level, and the degree of degradation increases sequentially from the first degradation level to the second degradation level and the third degradation level. Methods for obtaining the spoilage level of stored items include: Preset the threshold values ​​for the rate of change of ethylene concentration (Ymax) and the threshold values ​​for the rate of change of TVOC gas content (Tmax); when >Ymax, and When Tmax is greater than Tmax, the third mutation level is determined to be generated; when <Ymax, and When Tmax is less than Tmax, the first mutation level is determined to be generated; The remaining cases are judged to generate a second deterioration level.

8. The AI ​​physical preservation technology control system with multi-zone adaptive gas circulation according to claim 1, characterized in that, Obtain the current preservation strategy for the second item. For different spoilage levels of stored items, trigger different atmosphere optimization strategies, including: When the spoilage level of stored items is the first spoilage level, no atmosphere optimization strategy is generated; When the spoilage level of the stored goods is the second or third spoilage level, the second item preservation strategy and the spoilage level of the stored goods are input into the pre-built gas parameter optimization model, and the corrected gas parameter ratio is output as the atmosphere optimization strategy.

9. The AI ​​physical preservation technology control system with multi-zone adaptive gas circulation according to claim 8, characterized in that, The training method for the gas parameter optimization model includes: Q sets of training data are collected in advance, where Q is a positive integer greater than 0. The training data includes the second item preservation strategy and the spoilage level of the stored item, as well as the gas parameter ratios corresponding to the second item preservation strategy and the spoilage level of the stored item. A deep learning model was selected as the gas parameter optimization model. Training data was used to train the gas parameter optimization model. The second item preservation strategy and the spoilage level of the stored items were used as inputs to the gas parameter optimization model, and the gas parameter ratio was used as the output. Stochastic gradient descent was employed, and the weights and biases of the gas parameter optimization model were adjusted using the backpropagation algorithm to minimize the error between the predicted and actual results. A loss function, the mean squared error, was set. When the loss function value converged, the training of the gas parameter optimization model was stopped, and the gas parameter optimization model corresponding to the convergence of the loss function value was taken as the trained gas parameter optimization model.

10. The AI ​​physical preservation technology control system with multi-region adaptive gas circulation according to claim 1, characterized in that, The sensor group includes an optical ethylene sensor and a PID-type TVOC sensor. The optical ethylene sensor is used to collect the ethylene concentration in the i-th independent airtight zone, and the PID-type TVOC sensor is used to collect the TVOC gas content in the i-th independent airtight zone.

Images