AI model-driven optimization control method and system for Ehime orange preservation

Abstract

The present application relates to the field of intelligent fruit preservation technology, and specifically discloses an AI model-driven Ehime orange preservation optimization control method and system thereof, which utilizes an AI model to perform full-time dynamic analysis on the data set of the collected in-store preservation parameters to infer the recommended in-store ambient temperature value under the preset in-store environmental conditions. Furthermore, based on the comparison between the recommended value of the in-store ambient temperature and the real-time in-store ambient temperature value, the preservation temperature is dynamically adjusted to adapt to the preservation requirements under different environmental conditions, so as to effectively extend the shelf life of Ehime oranges and reduce the quality degradation and loss of Ehime oranges.

Description

Technical Field

[0001] The present application relates to the technical field of fruit preservation, and more specifically, to an AI model-driven Ehime orange preservation optimization control method and system. Background Art

[0002] Ehime orange is a plant of the genus Citrus in the Rutaceae family. Its fruit is bright and smooth in color and of excellent quality. It is a fine variety of citrus cultivated in countries around the world. It is loved by people because it is rich in a large number of essential elements and functional ingredients that are beneficial to the human body. The functional active substances contained in Ehime orange fruit, such as hesperidin, limonoids, D-limonene, and carotenoids, can inhibit bacteria, resist oxidation, reduce the carcinogenic activity of carcinogens, inhibit tumor cell proliferation, and reduce the incidence of cardiovascular diseases. Therefore, long-term consumption of Ehime orange has a certain effect on the prevention and treatment of various chronic diseases.

[0003] With the advent of the "Internet +" era, more and more agricultural products are sold to distant places through online sales channels, and Ehime oranges are a typical example. Long-distance and large-volume transportation puts forward higher requirements for freshness preservation. At the same time, due to changes in the external environment, it is easy to lose water, shrink, discoloration and other phenomena. These phenomena are often accompanied by changes in nutritional components, reduced taste and product loss, which seriously affect product quality and sales reputation. With the improvement of e-commerce channels and the improvement of consumer requirements for products, freshness preservation has become an increasingly important issue.

[0004] Chemical methods are currently the most widely used preservation methods. Chemical preservatives can be divided into three categories according to their preventive and therapeutic functions: protective, broad-spectrum systemic, and fumigation. Broad-spectrum systemic types are more commonly used in the preservation of citrus fruits, mainly carbendazim, thiophanate, myclobutanil, thiophanate-methyl, and imazalil. In addition, waxing is also a commonly used chemical preservation method in business. In recent years, researchers have found that calcium ions, potassium sorbate, chlorine dioxide, carbonates and bicarbonates have a good effect on improving the storage stability and disease resistance of sweet oranges. Combining them with traditional chemical fungicides has a better effect. The preservation effect of chemical preservatives is significant, but it also brings problems such as chemical residues and pollution. Therefore, exploring safe and effective physical preservation technology has become the goal of more fruit and vegetable preservation experts.

[0005] As an innovative and forward-looking means of physical preservation, AI physical preservation technology significantly extends the shelf life of fruits through intelligent and precise control and optimization of preservation conditions, such as by changing the storage environment conditions to slow down the evaporation and respiration of the fruit, bringing new breakthroughs to the preservation of fruits such as Ehime oranges.

[0006] However, the key to agricultural product storage and preservation technology lies in the precise control of the environmental indicators within the warehouse. Some existing preservation technologies mainly rely on constant low-temperature storage conditions. Although they can delay the aging process of fruits and vegetables to a certain extent, they cannot provide personalized preservation solutions for fruits of different batches and different degrees of maturity. It is often difficult to achieve the optimal preservation effect and may even lead to excessive cooling or drying, affecting the quality of the fruit.

[0007] Therefore, an optimized Ehime orange preservation control method and system thereof are expected. Summary of the invention

[0008] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an AI model-driven Ehime orange preservation optimization control method and system thereof, and an AI model-driven Ehime orange preservation optimization control method and system thereof, which utilizes an AI model to perform a full-time dynamic analysis of the collected in-warehouse preservation parameter data set to infer the recommended in-warehouse ambient temperature value under the preset in-warehouse environmental conditions. Furthermore, based on the comparison between the recommended value of the in-warehouse ambient temperature and the real-time in-warehouse ambient temperature value, the preservation temperature is dynamically adjusted to adapt to the preservation requirements under different environmental conditions, so as to effectively extend the shelf life of Ehime oranges and reduce the quality degradation and loss of Ehime oranges.

[0009] Accordingly, according to one aspect of the present application, a method for optimizing the preservation of Ehime oranges driven by an AI model is provided, which includes:

[0010] A time queue for receiving in-store environmental parameters collected by a sensor matrix, wherein the in-store environmental parameters include real-time humidity value, real-time carbon dioxide concentration value, real-time oxygen concentration value and real-time ethylene concentration value;

[0011] Performing parameter-level local time domain feature analysis on the time queue of the environmental parameters in the storage to obtain a sequence of local time series implicitly associated feature vectors of real-time humidity, a sequence of local time series implicitly associated feature vectors of real-time carbon dioxide concentration, a sequence of local time series implicitly associated feature vectors of real-time oxygen concentration, and a sequence of local time series implicitly associated feature vectors of real-time ethylene concentration;

[0012] Performing feature propagation aggregation based on time series gradient attenuation on the sequence of local time series implicitly associated feature vectors of the real-time humidity, the sequence of local time series implicitly associated feature vectors of the real-time carbon dioxide concentration, the sequence of local time series implicitly associated feature vectors of the real-time oxygen concentration, and the sequence of local time series implicitly associated feature vectors of the real-time ethylene concentration, respectively, to obtain a real-time humidity fine-grained time series associated aggregated feature vector, a real-time carbon dioxide concentration fine-grained time series associated aggregated feature vector, a real-time oxygen concentration fine-grained time series associated aggregated feature vector, and a real-time ethylene concentration fine-grained time series associated aggregated feature vector;

[0013] Performing deep aggregation analysis on the real-time humidity fine-grained time series associated aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series associated aggregation feature vector, the real-time oxygen concentration fine-grained time series associated aggregation feature vector, and the real-time ethylene concentration fine-grained time series associated aggregation feature vector to obtain a multi-source time series deep association representation vector of the in-reservoir environmental parameters;

[0014] Based on the multi-source time series deep association representation vector of the environmental parameters in the library, the Ehime orange preservation temperature control strategy is determined.

[0015] According to another aspect of the present application, an AI model-driven Ehime orange preservation optimization control system is provided, which includes:

[0016] The in-store environment monitoring module is used to receive the time queue of the in-store environment parameters collected by the sensor matrix, wherein the in-store environment parameters include real-time humidity value, real-time carbon dioxide concentration value, real-time oxygen concentration value and real-time ethylene concentration value;

[0017] An environmental parameter local time series analysis module is used to perform parameter-level local time domain feature analysis on the time queue of the environmental parameters in the library to obtain a sequence of real-time humidity local time series implicitly associated feature vectors, a sequence of real-time carbon dioxide concentration local time series implicitly associated feature vectors, a sequence of real-time oxygen concentration local time series implicitly associated feature vectors, and a sequence of real-time ethylene concentration local time series implicitly associated feature vectors;

[0018] The global time domain information aggregation module of environmental parameters is used to perform feature propagation aggregation based on time gradient attenuation on the sequence of local time series implicitly associated feature vectors of real-time humidity, the sequence of local time series implicitly associated feature vectors of real-time carbon dioxide concentration, the sequence of local time series implicitly associated feature vectors of real-time oxygen concentration and the sequence of local time series implicitly associated feature vectors of real-time ethylene concentration, respectively, to obtain a real-time humidity fine-grained time series associated aggregation feature vector, a real-time carbon dioxide concentration fine-grained time series associated aggregation feature vector, a real-time oxygen concentration fine-grained time series associated aggregation feature vector and a real-time ethylene concentration fine-grained time series associated aggregation feature vector;

[0019] A multi-source environmental parameter association analysis module is used to perform deep aggregation analysis on the real-time humidity fine-grained time series association aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series association aggregation feature vector, the real-time oxygen concentration fine-grained time series association aggregation feature vector, and the real-time ethylene concentration fine-grained time series association aggregation feature vector to obtain a multi-source time series deep association representation vector of the in-reservoir environmental parameter;

[0020] The temperature control module is used to determine the Ehime orange preservation temperature control strategy based on the multi-source time series deep association representation vector of the environmental parameters in the warehouse.

[0021] Compared with the prior art, the AI ​​model-driven Ehime orange preservation optimization control method and system provided in this application, the AI ​​model-driven Ehime orange preservation optimization control method and system, use the AI ​​model to perform full-time dynamic analysis on the collected data set of storage parameters to infer the recommended storage environment temperature value under the preset storage environment conditions. Furthermore, based on the comparison between the recommended value of the storage environment temperature and the real-time storage environment temperature value, the preservation temperature is dynamically adjusted to adapt to the preservation requirements under different environmental conditions, so as to effectively extend the shelf life of Ehime oranges and reduce the quality decline and loss of Ehime oranges. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0023] Figure 1 This is a flow chart of the Ehime orange preservation optimization control method driven by the AI ​​model according to an embodiment of the present application.

[0024] Figure 2 This is a data flow diagram of the Ehime orange preservation optimization control method driven by the AI ​​model according to an embodiment of the present application.

[0025] Figure 3 This is a flowchart of step S120 in the Ehime orange preservation optimization control method driven by the AI ​​model according to an embodiment of the present application.

[0026] Figure 4 This is a flowchart of step S130 in the Ehime orange preservation optimization control method driven by the AI ​​model according to an embodiment of the present application.

[0027] Figure 5 This is a block diagram of an Ehime orange preservation optimization control system driven by an AI model according to an embodiment of the present application.

[0028] Figure 6 This is a schematic diagram comparing the storage effects of Ehime oranges in AI fresh-keeping storage and ordinary cold storage.

[0029] Figure 7 Schematic diagram of the weight loss comparison of Ehime oranges in AI fresh-keeping storage and ordinary cold storage.

[0030] Figure 8 This is a schematic diagram comparing the mold growth of Ehime oranges in the AI ​​fresh-keeping storage and ordinary cold storage.

[0031] Fig. 9 This is a diagram comparing the hardness of Ehime oranges in AI fresh-keeping storage and ordinary cold storage.

[0032] Fig.10 This is a schematic diagram comparing the soluble solids content of Ehime oranges in AI fresh-keeping storage and ordinary cold storage.

[0033] Fig.11 This is a schematic diagram comparing the color of Ehime orange in AI fresh-keeping storage and ordinary cold storage.

[0034] The AI ​​fresh-keeping warehouse is an AI fresh-keeping warehouse constructed based on the Ehime orange preservation optimization control method driven by the AI ​​model of this application. DETAILED DESCRIPTION

[0035] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0036] Figure 1 This is a flow chart of the Ehime orange preservation optimization control method driven by the AI ​​model according to an embodiment of the present application. Figure 2 The data flow diagram of the AI ​​model driven Ehime orange preservation optimization control method according to the embodiment of the present application is shown in FIG. Figure 1 and Figure 2As shown, according to the AI ​​model-driven Ehime orange preservation optimization control method of the embodiment of the present application, the method includes the following steps: S110, receiving a time queue of in-store environmental parameters collected by a sensor matrix, wherein the in-store environmental parameters include real-time humidity value, real-time carbon dioxide concentration value, real-time oxygen concentration value and real-time ethylene concentration value; S120, performing parameter-level local time domain feature analysis on the time queue of the in-store environmental parameters to obtain a sequence of local time series implicitly associated feature vectors of real-time humidity, a sequence of local time series implicitly associated feature vectors of real-time carbon dioxide concentration, a sequence of local time series implicitly associated feature vectors of real-time oxygen concentration and a sequence of local time series implicitly associated feature vectors of real-time ethylene concentration; S130, performing parameter-level local time series implicitly associated feature vectors on the sequence of local time series implicitly associated feature vectors of real-time humidity, a sequence of local time series implicitly associated feature vectors of real-time carbon dioxide concentration, a sequence of local time series implicitly associated feature vectors of real-time oxygen concentration and a sequence of local time series implicitly associated feature vectors of real-time ethylene concentration. The sequence of associated feature vectors and the sequence of local time series implicit associated feature vectors of real-time ethylene concentration are respectively subjected to feature propagation aggregation based on time series gradient attenuation to obtain real-time humidity fine-grained time series associated aggregation feature vector, real-time carbon dioxide concentration fine-grained time series associated aggregation feature vector, real-time oxygen concentration fine-grained time series associated aggregation feature vector and real-time ethylene concentration fine-grained time series associated aggregation feature vector; S140, deep aggregation analysis is performed on the real-time humidity fine-grained time series associated aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series associated aggregation feature vector, the real-time oxygen concentration fine-grained time series associated aggregation feature vector and the real-time ethylene concentration fine-grained time series associated aggregation feature vector to obtain a multi-source time series deep association representation vector of the in-warehouse environmental parameters; S150, based on the multi-source time series deep association representation vector of the in-warehouse environmental parameters, the Ehime orange preservation temperature control strategy is determined.

[0037] In a specific example, the step S110 receives a time queue of the in-store environmental parameters collected by the sensor matrix, wherein the in-store environmental parameters include real-time humidity value, real-time carbon dioxide concentration value, real-time oxygen concentration value and real-time ethylene concentration value. It should be understood that temperature is the most basic and most critical data for controlling the preservation effect of fruits. Since the fruit is still alive after picking, it will continue to respire. The higher the temperature, the stronger the respiration of the fruit, the more nutrients consumed, and the faster the speed of maturation and decay. At a lower temperature, the metabolic activity of the fruit slows down, the respiration weakens, the carbon dioxide and ethylene produced are reduced, and the humidity is more easily controlled within an ideal range, which is conducive to maintaining the quality of the fruit and extending its shelf life. However, too low a temperature can also cause chilling damage to the fruit and affect its normal metabolism. Therefore, in the technical solution of the present application, the humidity, carbon dioxide concentration, oxygen concentration and ethylene concentration in the environment are monitored in real time to fully understand the metabolic activity and preservation state of Ehime orange, and dynamically adjust the preservation temperature in the storehouse accordingly. For example, when the ethylene concentration increases, it indicates that the fruit begins to mature or rot, and the temperature can be appropriately lowered at this time to slow down this process. Likewise, if CO2 levels rise, this may indicate that the fruit is experiencing increased respiration, and the temperature may need to be lowered to slow respiration.

[0038] In a specific example, if Figure 3 As shown, the step S120 includes: S121, queuing the time queue of the environmental parameters in the library according to the environmental parameter sample dimension to obtain the time queue of real-time humidity values, the time queue of real-time carbon dioxide concentration values, the time queue of real-time oxygen concentration values ​​and the time queue of real-time ethylene concentration values; S122, inputting the time queue of real-time humidity values, the time queue of real-time carbon dioxide concentration values, the time queue of real-time oxygen concentration values ​​and the time queue of real-time ethylene concentration values ​​into a sequence encoder based on the 1D-CNN model respectively to obtain a sequence of local temporal implicit association feature vectors of the real-time humidity, a sequence of local temporal implicit association feature vectors of the real-time carbon dioxide concentration, a sequence of local temporal implicit association feature vectors of the real-time oxygen concentration and a sequence of local temporal implicit association feature vectors of the real-time ethylene concentration.

[0039] Specifically, in step S121, since different environmental parameters have different impact mechanisms on fruit preservation, in order to more finely analyze the actual state of the preservation environment, the present application further groups the time queues of the environmental parameters in the library according to the environmental parameter sample dimensions to form independent time series data for different environmental parameters, maintain the independence and integrity of the data, and facilitate more accurate analysis of the time change trends of various environmental parameters, thereby providing more accurate environmental information for subsequent data processing and preservation decisions.

[0040] In the step S122, in the time series data analysis process, directly performing time series analysis on a long-span data set is not only computationally inefficient, but may also result in the loss and inaccuracy of local detail information. Therefore, in the technical solution of the present application, a local time domain analysis method is adopted, by dividing the long-span environmental parameter time series data into shorter local subsequences, and using a 1D-CNN model to perform feature extraction in the local time domain on the time queue of each environmental parameter, so as to capture the subtle changes of each environmental parameter, dig out the dynamic change pattern of the environmental parameter time series under each local time domain, and generate corresponding time series feature representations, thereby obtaining a sequence of implicitly associated feature vectors of the local time series of real-time humidity, a sequence of implicitly associated feature vectors of the local time series of real-time carbon dioxide concentration, a sequence of implicitly associated feature vectors of the local time series of real-time oxygen concentration, and a sequence of implicitly associated feature vectors of the local time series of real-time ethylene concentration.

[0041] It is worth mentioning that the core of the 1D-CNN model is the 1D convolution layer, which is responsible for extracting local features from the input time series data. These layers capture local patterns and trends in the time series by applying multiple one-dimensional convolution kernels of different sizes and strides. The convolution kernel slides through the input data, covering a small segment of the time series each time, and calculates the dot product of the segment of data and the convolution kernel to generate a new feature. In the process of setting model parameters, choosing an appropriate convolution kernel size helps capture temporal features of different scales. For example, a smaller convolution kernel can capture short-term temporal patterns, while a larger convolution kernel can capture long-term temporal trends. The stride determines the speed at which the convolution kernel moves on the time series. The choice of stride affects the size of the output feature map. In order to keep the size of the input and output feature maps consistent, zero padding can be added at both ends of the input data. Common activation functions include ReLU (Rectified Linear Unit), sigmoid, and tanh, which can increase the nonlinear ability of the model and help the model better fit complex data distributions.

[0042] In a specific example, the purpose of step S130 is to further obtain the global time-domain dynamic change information of various environmental parameters. It is necessary to represent the local time-domain features of various environmental parameters in the time dimension for feature aggregation, so as to achieve a global temporal understanding of various environmental parameters. It is worth mentioning that, considering that in the process of information propagation and aggregation, the early environmental parameter features will gradually lose their importance over time, while the recent environmental parameter features have a greater impact on the current preservation state. Based on this, the present application proposes a feature propagation aggregation method based on time gradient attenuation, which introduces the time factor so that the recent environmental parameter features occupy a greater weight in the global time-domain information propagation aggregation process, thereby more accurately reflecting the real-time state of the current preservation environment and improving the accuracy of subsequent preservation temperature adjustment.

[0043] The step S130 includes: processing the sequence of the local time series implicit correlation feature vectors of the real-time ethylene concentration using the following information time series dynamic propagation formula to obtain the real-time ethylene concentration fine-grained time series correlation aggregation feature vector, wherein the information time series dynamic propagation formula is:

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051] in, represents a sequence of implicitly associated feature vectors of the local time series of real-time ethylene concentration, is the number of vectors in the sequence of implicitly associated feature vectors of the local time series of real-time ethylene concentration, represents the implicit correlation feature vector of the local time series of the current real-time ethylene concentration, , and The first, second and third sequences of the local time series implicit correlation feature vectors of the historical real-time ethylene concentration are represented respectively. The implicit correlation feature vector of the local time series of historical real-time ethylene concentration, represents the principal component analysis function, Indicates The principal component representation vector of the local time series of historical ethylene concentration, represents the principal component representation vector of the local time series of the current ethylene concentration, represents the transpose of a vector, represents the length of the implicit correlation feature vector of the local time series of the historical real-time ethylene concentration, Indicates the The covariance matrix of the implicit correlation eigenvectors of the local time series of historical real-time ethylene concentrations is the covariance matrix of the local time series of historical ethylene concentrations, Indicates that through The matrix is ​​composed of the set of principal component eigenvectors of the local time series of historical ethylene concentrations obtained by eigenvalue decomposition of the covariance matrix of the local time series of historical ethylene concentrations. Indicates that by The diagonal matrix consisting of the set of principal component eigenvalues ​​of the local time series of historical ethylene concentrations obtained by eigenvalue decomposition of the local time series covariance matrix of historical ethylene concentrations is is the diag function, which is used to construct a diagonal matrix. , , and Respectively represent The first, second, and third sets of principal component eigenvalues ​​of the local time series of historical ethylene concentration and The principal component eigenvalues ​​of the local time series of historical ethylene concentration, is the number of eigenvalues ​​in the set of principal component eigenvalues ​​of the local time series of historical ethylene concentration, , , and Respectively represent The first, second, and third principal component eigenvectors of the local time series of historical ethylene concentration are and The principal component eigenvectors of the local time series of historical ethylene concentrations, Indicates the index corresponding to the maximum value. represents the index of the maximum eigenvalue in the set of principal component eigenvalues ​​of the local time series of historical ethylene concentration, Indicates The principal component representation vector of the local time series of historical ethylene concentration eigenvalues, The first principal component representing the local time series of the current ethylene concentration is represented by eigenvalues, represents the absolute value, represents the logarithmic function with base 2, represents the length of the principal component representation vector of the local time series of the current ethylene concentration, represents the exponential function with base e, Indicates The absolute attenuation factor of the characteristic time series propagation of historical ethylene concentration, and They represent the implicit correlation feature vector of the local time series of the current real-time ethylene concentration and the The timestamp of the implicit correlation feature vector of the local time series of historical real-time ethylene concentration, Indicates rounding down. Indicates The time span modulates the propagation attenuation factor, Indicates mask processing, represents the sigmoid function, Indicates the preset threshold value, Indicates The historical ethylene concentration characteristic time series gradient attenuation weight, The real-time ethylene concentration fine-grained time series associated aggregated feature vector is represented.

[0052] Figure 4 Flow chart of step S130 in the AI ​​model driven Ehime orange preservation optimization control method according to an embodiment of the present application. Figure 4 As shown, the step S130 includes: S131, dividing the sequence of the local time series implicitly associated feature vectors of the real-time ethylene concentration into time nodes and extracting the principal components to obtain a sequence of the local time series principal component representation vectors of the current ethylene concentration and a sequence of the local time series principal component representation vectors of the historical ethylene concentration; S132, based on the time span of each local time series principal component representation vector of the historical ethylene concentration relative to the local time series principal component representation vector of the current ethylene concentration, calculating the time series gradient attenuation weights of each local time series principal component representation vector of the historical ethylene concentration to obtain a sequence of characteristic time series gradient attenuation weights of the historical ethylene concentration; S133, based on the sequence of characteristic time series gradient attenuation weights of the historical ethylene concentration, performing feature time series transfer aggregation on the sequence of the local time series implicitly associated feature vectors of the real-time ethylene concentration to obtain the real-time ethylene concentration fine-grained time series associated aggregation feature vector.

[0053] Specifically, the data processing process of step S131 is as follows: first, extract the current real-time ethylene concentration local time series implicitly associated feature vector from the sequence of the real-time ethylene concentration local time series implicitly associated feature vectors, and perform eigenvalue-based principal component extraction on the current real-time ethylene concentration local time series implicitly associated feature vector to obtain the current ethylene concentration local time series principal component representation vector; then, define other real-time ethylene concentration local time series implicitly associated feature vectors in the sequence of the real-time ethylene concentration local time series implicitly associated feature vectors as historical real-time ethylene concentration local time series implicitly associated feature vectors, and perform eigenvalue-based principal component extraction on each historical real-time ethylene concentration local time series implicitly associated feature vector to obtain a sequence of historical ethylene concentration local time series principal component representation vectors.

[0054] Specifically, the eigenvalue-based principal component extraction process includes the following steps: the first step is to calculate the covariance matrix of the implicitly associated eigenvector of the current real-time ethylene concentration local time series to obtain the covariance matrix of the current ethylene concentration local time series; the second step is to perform eigenvalue decomposition on the current ethylene concentration local time series covariance matrix to obtain a set of principal component eigenvalues ​​of the current ethylene concentration local time series and a set of corresponding principal component eigenvectors of the current ethylene concentration local time series; the third step is to select the current ethylene concentration local time series principal component eigenvector corresponding to the maximum eigenvalue in the set of principal component eigenvalues ​​of the current ethylene concentration local time series as the principal component representation vector of the current ethylene concentration local time series.

[0055] In a specific example, the step S132 includes: calculating the time series propagation absolute attenuation factor of each historical ethylene concentration local time series principal component representation vector in the sequence of historical ethylene concentration local time series principal component representation vectors relative to the current ethylene concentration local time series principal component representation vector to obtain a sequence of historical ethylene concentration characteristic time series propagation absolute attenuation factors; performing time span modulation on the sequence of historical ethylene concentration characteristic time series propagation absolute attenuation factors based on the time span between each historical ethylene concentration local time series principal component representation vector in the sequence of historical ethylene concentration local time series principal component representation vectors and the current ethylene concentration local time series principal component representation vector to obtain a sequence of time span modulated propagation attenuation factors; inputting the sequence of time span modulated propagation attenuation factors into an information transfer screening module based on a gating function to obtain a sequence of historical ethylene concentration characteristic time series gradient attenuation weights.

[0056] Correspondingly, the process of calculating the absolute attenuation factor of time series propagation includes: calculating the dot division vector between the local time series principal component representation vector of the historical ethylene concentration and the local time series principal component representation vector of the current ethylene concentration, and calculating the base-two logarithmic value of the absolute value of each eigenvalue in the dot division vector to obtain the absolute standard representation vector of time series propagation attenuation; calculating the dot product vector between the absolute standard representation vector of time series propagation attenuation and the local time series principal component representation vector of the historical ethylene concentration, and calculating the exponential function value with base e and the sum of the eigenvalues ​​of the dot product vector as the exponent to obtain the characteristic time series propagation absolute attenuation factor of the historical ethylene concentration.

[0057] In a specific example of the present application, the process of time span modulating the sequence of absolute attenuation factors of the time series propagation of the historical ethylene concentration characteristics includes: calculating the timestamp difference between the historical ethylene concentration local time series principal component representation vector and the current ethylene concentration local time series principal component representation vector, rounding down the timestamp difference to obtain the time span factor; dividing the time series propagation absolute attenuation factor of the historical ethylene concentration local time series principal component representation vector by the natural exponential function value of the time span factor to obtain the time span modulated propagation attenuation factor.

[0058] In a specific example of the present application, the sequence of time span modulated propagation attenuation factors is input into an information transfer screening module based on a gating function to obtain a sequence of historical ethylene concentration characteristic time series gradient attenuation weights, including: based on a preset threshold, masking the sequence of time span modulated propagation attenuation factors to obtain a sequence of masked time span modulated propagation attenuation factors; inputting the sequence of masked time span modulated propagation attenuation factors into a Sigmoid function for normalization processing to obtain a sequence of historical ethylene concentration characteristic time series gradient attenuation weights.

[0059] Specifically, the step S133 includes: based on the sequence of historical ethylene concentration feature time series gradient attenuation weights, calculating the weighted sum of the sequence of historical ethylene concentration local time series principal component representation vectors to obtain the historical ethylene concentration local time series principal component significant transfer aggregation representation vector; calculating the positional sum of the historical ethylene concentration local time series principal component significant transfer aggregation representation vector and the current ethylene concentration local time series principal component representation vector to obtain the real-time ethylene concentration fine-grained time series associated aggregation feature vector.

[0060] Specifically, taking the sequence of the local time series implicit correlation feature vector of the real-time ethylene concentration as an example, first, the last time node is regarded as the current time node, and the other time nodes are regarded as historical time nodes. The PCA analysis is performed on the local time series implicit correlation feature vector of the real-time ethylene concentration at each time node to remove the noise interference in the data, highlight the key information, and extract the most representative principal component. Then, in order to capture the trend of the evolution of the ethylene concentration characteristics of each historical time node over time and the information loss, the characteristic principal component change rate of each historical time node relative to the current time node and its corresponding time series propagation absolute attenuation factor are further calculated to enhance the model's understanding of information transmission under long time intervals. Then, the above-mentioned time series propagation absolute attenuation factor is adjusted by introducing the time span factor to simulate the gradual forgetting process of data in the time dimension, so that the model can pay more attention to the recent ethylene concentration change trend and reduce excessive dependence on early data. Then, based on the gating mechanism, the information of the absolute attenuation factor of the time series propagation after time span modulation is screened, the most influential propagation path is identified and the corresponding weight is assigned accordingly, the feature discrimination is improved, and the main component features of the ethylene concentration at each time node are weighted and aggregated based on the generated weight, so as to effectively integrate the long-term change trend and short-term fluctuation of the ethylene concentration and generate the real-time ethylene concentration fine-grained time series associated aggregated feature vector. Through the above data processing method, the sequence of the real-time humidity local time series implicit associated feature vector, the sequence of the real-time carbon dioxide concentration local time series implicit associated feature vector, the sequence of the real-time oxygen concentration local time series implicit associated feature vector and the sequence of the real-time ethylene concentration local time series implicit associated feature vector are processed respectively, so as to obtain the real-time humidity fine-grained time series associated aggregated feature vector containing global time series associated change information, the real-time carbon dioxide concentration fine-grained time series associated aggregated feature vector, the real-time oxygen concentration fine-grained time series associated aggregated feature vector and the real-time ethylene concentration fine-grained time series associated aggregated feature vector.

[0061] Furthermore, the step S140 includes: inputting the real-time humidity fine-grained time series association aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series association aggregation feature vector, the real-time oxygen concentration fine-grained time series association aggregation feature vector and the real-time ethylene concentration fine-grained time series association aggregation feature vector into the in-repository environmental parameter multi-source deep aggregation analysis engine based on the multi-layer perceptron model to obtain the in-repository environmental parameter multi-source time series deep association representation vector.

[0062] It should be understandable that, considering that the analysis of a single parameter cannot fully reflect the comprehensive state of the fruit preservation environment, in order to more comprehensively evaluate the metabolic activity and preservation environment of Ehime orange, the present application further adopts a multi-layer perceptron model to perform multi-source data deep aggregation analysis on the real-time humidity fine-grained time series association aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series association aggregation feature vector, the real-time oxygen concentration fine-grained time series association aggregation feature vector and the real-time ethylene concentration fine-grained time series association aggregation feature vector, so as to utilize the nonlinear mapping capability of the multi-layer perceptron model to map the multi-source environmental parameter features to a unified feature space, fully explore the mutual correlation and influence between different environmental parameters, realize the deep fusion of multimodal features, and obtain a comprehensive multi-source time series deep association representation vector of the in-library environmental parameters.

[0063] Furthermore, the data flow of step S150 includes: firstly, inputting the multi-source time-series deep association representation vector of the in-warehouse environmental parameters into the decoder-based in-warehouse environmental temperature dynamic planner to obtain the recommended value of the in-warehouse environmental temperature; then, based on the comparison between the recommended value of the in-warehouse environmental temperature and the real-time in-warehouse environmental temperature value, determining the Ehime orange preservation temperature control strategy.

[0064] That is, the powerful decoding prediction ability of the decoder model is used to perform feature analysis on the multi-source environmental parameter correlation change information contained in the multi-source temporal deep correlation representation vector of the environmental parameters in the warehouse. Specifically, through training, the decoder can effectively learn the dynamic change law of the metabolic activity of Ehime oranges reflected by the multi-source temporal deep correlation representation vector of the environmental parameters in the warehouse, so as to intelligently predict the most suitable warehouse environmental temperature for the current state, so as to maintain the best freshness of Ehime oranges.

[0065] Furthermore, a closed-loop feedback mechanism for temperature control is implemented by comparing the recommended value of the ambient temperature in the warehouse with the real-time ambient temperature value in the warehouse. For example, when there is a deviation between the recommended temperature and the actual temperature, the system automatically adjusts the operating state of the refrigeration or heating equipment to ensure that the temperature in the warehouse is stable near the recommended value. Through this dynamic adjustment mechanism, it is possible to effectively cope with ambient temperature fluctuations and ensure the freshness of Ehime oranges during storage.

[0066] In particular, in the technical solution of the present application, the real-time humidity fine-grained time series association aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series association aggregation feature vector, the real-time oxygen concentration fine-grained time series association aggregation feature vector and the real-time ethylene concentration fine-grained time series association aggregation feature vector respectively represent the time series significant propagation aggregation characteristics of the real-time humidity value, the real-time carbon dioxide concentration value, the real-time oxygen concentration value and the real-time ethylene concentration value. Taking into account the source time series distribution differences of the real-time humidity value, the real-time carbon dioxide concentration value, the real-time oxygen concentration value and the real-time ethylene concentration value, when the real-time humidity fine-grained time series association aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series association aggregation feature vector, the real-time oxygen concentration fine-grained time series association aggregation feature vector and the real-time ethylene concentration fine-grained time series association aggregation feature vector are input into the multi-source deep aggregation analysis engine of the in-store environmental parameters based on the multi-layer perceptron model, the source time series distribution differences will be amplified through feature encoding, thereby making the feature manifold of the multi-source time series deep association representation vector of the in-store environmental parameters obtained based on multi-layer fully connected encoding have expression sparsity and distribution imbalance in the high-dimensional feature space, affecting the accuracy of the recommended value of the in-store ambient temperature obtained by the decoder-based in-store ambient temperature dynamic planner.

[0067] Based on this, in a preferred embodiment, the multi-source time series deep association representation vector of the in-store environmental parameters is input into a decoder-based in-store environmental temperature dynamic planner to obtain a recommended value of the in-store environmental temperature, including the following steps:

[0068] Determine the number of zero eigenvalues ​​in the multi-source time series deep correlation representation vector of the environmental parameter in the library, and respectively calculate the reciprocal of the logarithm of the number of zero eigenvalues ​​with base two and the exponential value of the reciprocal of the number of zero eigenvalues ​​with base natural constants to obtain a first multi-source time series deep correlation representation value of the environmental parameter in the library and a second multi-source time series deep correlation representation value of the environmental parameter in the library;

[0069] Calculate the power function of each eigenvalue of the multi-source time series deep correlation representation vector of the in-repository environmental parameter with the inverse of the number of zero eigenvalues ​​as the exponent to obtain the in-repository environmental parameter multi-source time series deep correlation structure state vector;

[0070] Performing a dot multiplication of the in-repository environment parameter multi-source temporal depth association structure state vector and the first in-repository environment parameter multi-source temporal depth association representation value to obtain an in-repository environment parameter multi-source temporal depth association information representation vector;

[0071] The autocorrelation matrix of the state vector of the multi-source time series deep correlation structure of the environmental parameters in the library is respectively multiplied by the second multi-source time series deep correlation representation value and the weight hyperparameter of the environmental parameters in the library to obtain the multi-source time series deep correlation regression understanding matrix of the environmental parameters in the library;

[0072] Performing matrix multiplication of the in-repository environment parameter multi-source temporal deep association information representation vector and the in-repository environment parameter multi-source temporal deep association regression understanding matrix to obtain an optimized in-repository environment parameter multi-source temporal deep association representation vector; and

[0073] The optimized multi-source temporal depth association representation vector of the in-library environmental parameters is input into a surface protection film material recommender based on the first decoder to obtain a decoding result of the material of the surface protection film.

[0074] That is, the multi-source temporal depth correlation representation value of the first internal environment parameter and the multi-source temporal depth correlation representation value of the second internal environment parameter are expressed as:

[0075]

[0076]

[0077] in It is the number of zero eigenvalues ​​in the multi-source time series deep correlation representation vector of the environmental parameters in the library.

[0078] And the optimized multi-source time series deep association representation vector of the internal environment parameters of the library is expressed as:

[0079]

[0080]

[0081]

[0082] The multi-source time series deep correlation representation vector of the environmental parameters in the library is a row vector.

[0083] Based on this, in the above-mentioned preferred example, by modeling the vector field zero dimension of the multi-source temporal deep association representation vector of the in-library environmental parameters as a global field redundant dependency, effective long-range modeling of the multi-source temporal deep association representation vector of the in-library environmental parameters is achieved under the field linear complexity. In this way, in the feature regression process of the multi-source temporal deep association representation vector of the in-library environmental parameters, structured correlation self-distillation between information representation and decoding regression understanding is performed in a selective state space based on feature distribution, thereby maintaining the overall information complexity of the feature distribution of the multi-source temporal deep association representation vector of the in-library environmental parameters while promoting the long-range dynamic regression understanding balance of its feature convergence, so as to improve the regression understanding consistency between the feature regression and feature extraction processes, and improve the accuracy of the recommended value of the in-library ambient temperature obtained by the decoder-based in-library ambient temperature dynamic planner.

[0084] In summary, the AI ​​model-driven Ehime orange preservation optimization control method according to the embodiment of the present application is explained, which monitors the various environmental parameters in the warehouse in real time through the sensor matrix, and uses data processing technology based on deep learning to perform fine-grained local time domain analysis and global time domain information propagation aggregation on various environmental parameters, so as to dig out the global time series dynamic change pattern of various environmental parameters, and then, through multi-source data deep aggregation analysis of various environmental parameters, to achieve global deep perception of Ehime orange metabolic activities and preservation status, and based on this, intelligently plan the preservation temperature in the warehouse to achieve the best preservation effect. In this way, the preservation temperature can be dynamically adjusted according to the actual needs of Ehime oranges and the real-time changes in the environment in the warehouse to adapt to the preservation needs under different environmental conditions, thereby effectively extending the shelf life of Ehime oranges and reducing the quality decline and loss of Ehime oranges.

[0085] Implementation Data

[0086] This application uses the above-mentioned AI model-driven Ehime orange preservation optimization control method to build an AI preservation cold storage, selects the same batch of mature Ehime orange fruits that are fresh, highly mature, similar in size, free of pests and diseases, and intact in appearance as research materials, and takes two groups of the same number of fruits and stores them in the AI ​​preservation cold storage and an ordinary cold storage controlled by fixed temperature parameters for comparative experiments. The Ehime oranges are observed at a fixed time and place, and the fruit corruption is recorded in detail, including whether there are mold spots, softening degree, whether there is odor, etc. Among them, the fruits stored in the AI ​​preservation storage are the experimental group, and the fruits stored in the ordinary cold storage are the control group.

[0087] After continuous and close observation, the results showed that the Ehime oranges in the AI ​​fresh cold storage can be kept fresh for up to 60 days. Figure 6 As shown in the figure, the Ehime oranges were put into storage on September 23, 2023, and were taken out on October 27, 2023, with a storage period of 35 days. At the end of the shelf life, they still maintained a high degree of freshness, while the Ehime oranges in ordinary cold storage only lasted for 31 days, and showed obvious signs of rot and mold. This shows that AI fresh cold storage can significantly extend the shelf life of Ehime oranges, keeping them fresh longer.

[0088] In terms of weight loss rate, the average weight of Ehime oranges in AI fresh cold storage was 287 grams before storage, and the average weight of each fruit was 263 grams when it was shipped out on the 29th day, with a weight loss rate of only 8.4%. In contrast, the average weight of Ehime oranges in ordinary cold storage was 284 grams before storage, and the average weight of each fruit was 209 grams when it was shipped out on the 29th day, with a weight loss rate of 26.4%. Figure 7 As shown, AI fresh-keeping cold storage can better maintain the moisture of Ehime oranges, reduce the weight loss of fruits, and thus maintain their freshness.

[0089] In terms of mold infection, the total number of fruits released from the storage on the 35th day of the storage experiment was 180, and 112 moldy fruits were found in the control group, with a mold rate of 62.20%. At this time, the mold infection rate of Ehime oranges in the AI ​​fresh-keeping cold storage was 0.00%. On the 52nd day, there were 172 moldy fruits in the total number of 180 fruits in the control group, and the mold infection rate of Ehime oranges in the ordinary cold storage was 95.50%. There were 35 moldy fruits in the experimental group, and the mold infection rate of Ehime oranges in the AI ​​fresh-keeping cold storage was only 19.40%. Figure 8 As shown, AI fresh-keeping cold storage can more effectively prevent the growth of mold and ensure the product quality of Ehime oranges.

[0090] like Fig. 9 As shown in the figure, in the hardness test, the hardness of the Ehime oranges of the same batch before storage was 588.43 ± 23.27g. On the 29th day of the storage experiment, the hardness of the Ehime oranges in the AI ​​fresh cold storage remained at 563.51 ± 21.57g, while the hardness of the Ehime oranges in the ordinary cold storage dropped to 476.88 ± 23.53g. On the 45th day of the storage experiment, the hardness of the Ehime oranges in the AI ​​fresh cold storage dropped to 509.51 ± 23.56g, while the Ehime oranges in the ordinary cold storage had already rotted. This shows that the AI ​​fresh cold storage can better maintain the fruit hardness of the Ehime oranges as well as the taste and nutritional value of the Ehime oranges.

[0091] like Fig.10 As shown in the figure, on the 29th day of storage in ordinary cold storage, the soluble solid content of Ehime oranges dropped to 12°Brix, while the soluble solid content of Ehime oranges in AI fresh cold storage was 9°Brix. On the 45th day of storage, the soluble solid content of Ehime oranges in AI fresh cold storage was 13°Brix, while the soluble solid content of Ehime oranges in ordinary cold storage dropped to 10°Brix. This shows that AI fresh cold storage can better maintain the sugar content of Ehime oranges, so that they still have a sweet taste after the shelf life ends.

[0092] In terms of color, on the 21st day of the storage experiment, the Ehime orange in the AI ​​fresh cold storage still maintained a bright orange color, while the Ehime orange in the ordinary cold storage became darker. Fig.11 As shown, it shows that AI fresh-keeping cold storage can better maintain the appearance of Ehime oranges, so that they still have a high commodity value after the shelf life.

[0093] In summary, AI fresh cold storage has shown significant superiority in storing Ehime oranges. Its long-lasting fresh-keeping time, lower weight loss rate, mold rate, and excellent hardness and soluble solids content all prove the excellent performance of AI fresh cold storage in maintaining the freshness and taste of fruits. Therefore, compared with ordinary cold storage, AI fresh cold storage is more suitable for long-term storage of Ehime oranges.

[0094] According to another aspect of the present application, an AI model-driven Ehime orange preservation optimization control system is also provided. Figure 5 FIG. 1 is a block diagram of an optimization control system for preserving Ehime oranges driven by an AI model according to an embodiment of the present application. Figure 5 As shown, according to the AI ​​model driven Ehime orange preservation optimization control system 100 of the embodiment of the present application, it includes: an in-store environment monitoring module 110, which is used to receive the time queue of the in-store environment parameters collected by the sensor matrix, wherein the in-store environment parameters include real-time humidity value, real-time carbon dioxide concentration value, real-time oxygen concentration value and real-time ethylene concentration value; an environmental parameter local time series analysis module 120, which is used to perform parameter-level local time domain feature analysis on the time queue of the in-store environment parameters to obtain a sequence of local time series implicitly associated feature vectors of real-time humidity, a sequence of local time series implicitly associated feature vectors of real-time carbon dioxide concentration, a sequence of local time series implicitly associated feature vectors of real-time oxygen concentration and a sequence of local time series implicitly associated feature vectors of real-time ethylene concentration; an environmental parameter global time domain information aggregation module 130, which is used to perform parameter-level local time series implicitly associated feature vectors on the time queue of the in-store environment parameters to obtain a sequence of local time series implicitly associated feature vectors of real-time humidity, a sequence of local time series implicitly associated feature vectors of real-time carbon dioxide concentration, a sequence of local time series implicitly associated feature vectors of real-time oxygen concentration and a sequence of local time series implicitly associated feature vectors of real-time ethylene concentration The sequence of local time series implicit correlation feature vectors of gas concentration and the sequence of local time series implicit correlation feature vectors of real-time ethylene concentration are respectively subjected to feature propagation aggregation based on time series gradient attenuation to obtain real-time humidity fine-grained time series correlation aggregation feature vector, real-time carbon dioxide concentration fine-grained time series correlation aggregation feature vector, real-time oxygen concentration fine-grained time series correlation aggregation feature vector and real-time ethylene concentration fine-grained time series correlation aggregation feature vector; a multi-source environmental parameter association analysis module 140 is used to perform deep aggregation analysis on the real-time humidity fine-grained time series correlation aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series correlation aggregation feature vector, the real-time oxygen concentration fine-grained time series correlation aggregation feature vector and the real-time ethylene concentration fine-grained time series correlation aggregation feature vector to obtain a multi-source time series deep correlation representation vector of the in-warehouse environmental parameters; a temperature control module 150 is used to determine the Ehime orange preservation temperature control strategy based on the multi-source time series deep correlation representation vector of the in-warehouse environmental parameters.

[0095] Here, those skilled in the art can understand that the specific operations of each module in the Ehime orange preservation optimization control system driven by the above AI model have been referenced above. Figures 1 to 4 The description of the Ehime orange preservation optimization control method driven by the AI ​​model has been introduced in detail, and therefore, its repeated description will be omitted.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. An AI model-driven optimization control method for Ehime orange preservation, characterized in that: include: A time queue for receiving in-store environmental parameters collected by a sensor matrix, wherein the in-store environmental parameters include real-time humidity value, real-time carbon dioxide concentration value, real-time oxygen concentration value and real-time ethylene concentration value; Performing parameter-level local time domain feature analysis on the time queue of the environmental parameters in the storage to obtain a sequence of local time series implicitly associated feature vectors of real-time humidity, a sequence of local time series implicitly associated feature vectors of real-time carbon dioxide concentration, a sequence of local time series implicitly associated feature vectors of real-time oxygen concentration, and a sequence of local time series implicitly associated feature vectors of real-time ethylene concentration; The sequence of the real-time humidity local time series implicitly associated feature vectors, the sequence of the real-time carbon dioxide concentration local time series implicitly associated feature vectors, the sequence of the real-time oxygen concentration local time series implicitly associated feature vectors, and the sequence of the real-time ethylene concentration local time series implicitly associated feature vectors are respectively subjected to feature propagation aggregation based on time series gradient attenuation to obtain a real-time humidity fine-grained time series associated aggregated feature vector, a real-time carbon dioxide concentration fine-grained time series associated aggregated feature vector, a real-time oxygen concentration fine-grained time series associated aggregated feature vector, and a real-time ethylene concentration fine-grained time series associated aggregated feature vector; including: Performing time node division and principal component extraction on the sequence of implicitly associated feature vectors of the local time series of real-time ethylene concentration to obtain a sequence of principal component representation vectors of the local time series of current ethylene concentration and a sequence of principal component representation vectors of the local time series of historical ethylene concentration; Based on the time span of each historical ethylene concentration local time series principal component representation vector in the sequence of the historical ethylene concentration local time series principal component representation vector relative to the current ethylene concentration local time series principal component representation vector, the time series gradient attenuation weight of each historical ethylene concentration local time series principal component representation vector is calculated to obtain a sequence of historical ethylene concentration characteristic time series gradient attenuation weights; Based on the sequence of the historical ethylene concentration feature time series gradient attenuation weights, the sequence of the real-time ethylene concentration local time series implicit correlation feature vectors is subjected to feature time series transfer aggregation to obtain the real-time ethylene concentration fine-grained time series correlation aggregation feature vector; Performing deep aggregation analysis on the real-time humidity fine-grained time series associated aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series associated aggregation feature vector, the real-time oxygen concentration fine-grained time series associated aggregation feature vector, and the real-time ethylene concentration fine-grained time series associated aggregation feature vector to obtain a multi-source time series deep association representation vector of the in-reservoir environmental parameters; Based on the multi-source time series deep association representation vector of the environmental parameters in the library, the Ehime orange preservation temperature control strategy is determined.

2. The AI ​​model-driven Ehime orange preservation optimization control method according to claim 1 is characterized in that: Performing parameter-level local time domain feature analysis on the time queue of the environmental parameters in the storage to obtain a sequence of real-time humidity local time series implicitly associated feature vectors, a sequence of real-time carbon dioxide concentration local time series implicitly associated feature vectors, a sequence of real-time oxygen concentration local time series implicitly associated feature vectors, and a sequence of real-time ethylene concentration local time series implicitly associated feature vectors, including: The time queues of the environmental parameters in the library are sorted according to the environmental parameter sample dimensions to obtain a time queue of real-time humidity values, a time queue of real-time carbon dioxide concentration values, a time queue of real-time oxygen concentration values, and a time queue of real-time ethylene concentration values; The time queue of the real-time humidity values, the time queue of the real-time carbon dioxide concentration values, the time queue of the real-time oxygen concentration values ​​and the time queue of the real-time ethylene concentration values ​​are respectively input into a sequence encoder based on the 1D-CNN model to obtain a sequence of local temporal implicit association feature vectors of the real-time humidity, a sequence of local temporal implicit association feature vectors of the real-time carbon dioxide concentration, a sequence of local temporal implicit association feature vectors of the real-time oxygen concentration and a sequence of local temporal implicit association feature vectors of the real-time ethylene concentration.

3. The AI ​​model-driven Ehime orange preservation optimization control method according to claim 1 is characterized in that: The sequence of implicitly associated feature vectors of the local time series of real-time ethylene concentration is divided into time nodes and principal components are extracted to obtain a sequence of principal component representation vectors of the local time series of current ethylene concentration and a sequence of principal component representation vectors of the local time series of historical ethylene concentration, including: Extracting the current real-time ethylene concentration local time series implicit correlation feature vector from the sequence of the real-time ethylene concentration local time series implicit correlation feature vectors, and performing principal component extraction based on the eigenvalue on the current real-time ethylene concentration local time series implicit correlation feature vector to obtain the current ethylene concentration local time series principal component representation vector; The other real-time ethylene concentration local time series implicitly associated feature vectors in the sequence of the real-time ethylene concentration local time series implicitly associated feature vectors are defined as historical real-time ethylene concentration local time series implicitly associated feature vectors, and principal component extraction based on the eigenvalue is performed on each historical real-time ethylene concentration local time series implicitly associated feature vector to obtain a sequence of principal component representation vectors of the historical ethylene concentration local time series.

4. The AI ​​model-driven Ehime orange preservation optimization control method according to claim 3 is characterized in that: Based on the time span of each historical ethylene concentration local time series principal component representation vector in the sequence of the historical ethylene concentration local time series principal component representation vector relative to the current ethylene concentration local time series principal component representation vector, the time series gradient attenuation weight of each historical ethylene concentration local time series principal component representation vector is calculated to obtain a sequence of historical ethylene concentration characteristic time series gradient attenuation weights, including: Calculating the time series propagation absolute attenuation factor of each historical ethylene concentration local time series principal component representation vector in the sequence of the historical ethylene concentration local time series principal component representation vector relative to the current ethylene concentration local time series principal component representation vector to obtain a sequence of historical ethylene concentration characteristic time series propagation absolute attenuation factors; Based on the time span between each historical ethylene concentration local time series principal component representation vector in the sequence of historical ethylene concentration local time series principal component representation vectors and the current ethylene concentration local time series principal component representation vector, the sequence of historical ethylene concentration characteristic time series propagation absolute attenuation factors is time span modulated to obtain a sequence of time span modulated propagation attenuation factors; The sequence of the time span modulated propagation attenuation factors is input into the information transfer screening module based on the gating function to obtain the sequence of the historical ethylene concentration characteristic time series gradient attenuation weights.

5. The AI ​​model-driven Ehime orange preservation optimization control method according to claim 4 is characterized in that: The method comprises: performing time span modulation on the sequence of absolute attenuation factors of the characteristic time series propagation of historical ethylene concentrations based on the time span between each historical ethylene concentration local time series principal component representation vector in the sequence of the historical ethylene concentration local time series principal component representation vector and the current ethylene concentration local time series principal component representation vector to obtain a sequence of time span modulated propagation attenuation factors, including: Calculating the timestamp difference between the historical ethylene concentration local time series principal component representation vector and the current ethylene concentration local time series principal component representation vector, and rounding down the timestamp difference to obtain a time span factor; The time-span modulation propagation attenuation factor is obtained by dividing the time-span modulation propagation attenuation factor of ... absolute attenuation factor of the local time-span principal component representation vector of the historical ethylene concentration by the natural exponential function value of the time-span factor.

6. The AI ​​model-driven Ehime orange preservation optimization control method according to claim 5, characterized in that: Based on the sequence of the historical ethylene concentration feature time series gradient attenuation weights, the sequence of the real-time ethylene concentration local time series implicit correlation feature vectors is subjected to feature time series transfer aggregation to obtain the real-time ethylene concentration fine-grained time series correlation aggregation feature vector, including: Based on the sequence of the gradient attenuation weights of the historical ethylene concentration characteristic time series, calculating the weighted sum of the sequence of the principal component representation vectors of the local time series of the historical ethylene concentration to obtain the principal component significant transfer aggregation representation vector of the local time series of the historical ethylene concentration; The position-wise sum of the historical ethylene concentration local time series principal component significant transfer aggregation representation vector and the current ethylene concentration local time series principal component representation vector is calculated to obtain the real-time ethylene concentration fine-grained time series associated aggregation feature vector.

7. The AI ​​model-driven Ehime orange preservation optimization control method according to claim 6 is characterized in that: The real-time humidity fine-grained time series associated aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series associated aggregation feature vector, the real-time oxygen concentration fine-grained time series associated aggregation feature vector, and the real-time ethylene concentration fine-grained time series associated aggregation feature vector are subjected to deep aggregation analysis to obtain a multi-source time series deep association representation vector of the in-reservoir environmental parameters, including: The real-time humidity fine-grained time series association aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series association aggregation feature vector, the real-time oxygen concentration fine-grained time series association aggregation feature vector and the real-time ethylene concentration fine-grained time series association aggregation feature vector are input into the in-reservoir environmental parameter multi-source deep aggregation analysis engine based on the multi-layer perceptron model to obtain the in-reservoir environmental parameter multi-source time series deep association representation vector.

8. The AI ​​model-driven Ehime orange preservation optimization control method according to claim 7, characterized in that: Based on the multi-source time series deep association representation vector of the environmental parameters in the library, the Ehime orange preservation temperature control strategy is determined, including: Inputting the multi-source time series deep association representation vector of the in-store environmental parameters into the in-store environmental temperature dynamic planner based on the decoder to obtain a recommended value of the in-store environmental temperature; Based on the comparison between the recommended value of the in-store ambient temperature and the real-time in-store ambient temperature value, the Ehime orange preservation temperature control strategy is determined.

9. An AI model driven optimization control system for Ehime orange preservation, characterized in that: include: The in-store environment monitoring module is used to receive the time queue of the in-store environment parameters collected by the sensor matrix, wherein the in-store environment parameters include real-time humidity value, real-time carbon dioxide concentration value, real-time oxygen concentration value and real-time ethylene concentration value; An environmental parameter local time series analysis module is used to perform parameter-level local time domain feature analysis on the time queue of the environmental parameters in the library to obtain a sequence of real-time humidity local time series implicitly associated feature vectors, a sequence of real-time carbon dioxide concentration local time series implicitly associated feature vectors, a sequence of real-time oxygen concentration local time series implicitly associated feature vectors, and a sequence of real-time ethylene concentration local time series implicitly associated feature vectors; The environmental parameter global time domain information aggregation module is used to perform feature propagation aggregation based on time gradient attenuation on the sequence of local time series implicitly associated feature vectors of the real-time humidity, the sequence of local time series implicitly associated feature vectors of the real-time carbon dioxide concentration, the sequence of local time series implicitly associated feature vectors of the real-time oxygen concentration, and the sequence of local time series implicitly associated feature vectors of the real-time ethylene concentration, respectively, to obtain a real-time humidity fine-grained time series associated aggregation feature vector, a real-time carbon dioxide concentration fine-grained time series associated aggregation feature vector, a real-time oxygen concentration fine-grained time series associated aggregation feature vector, and a real-time ethylene concentration fine-grained time series associated aggregation feature vector; comprising: Performing time node division and principal component extraction on the sequence of implicitly associated feature vectors of the local time series of real-time ethylene concentration to obtain a sequence of principal component representation vectors of the local time series of current ethylene concentration and a sequence of principal component representation vectors of the local time series of historical ethylene concentration; Based on the time span of each historical ethylene concentration local time series principal component representation vector in the sequence of the historical ethylene concentration local time series principal component representation vector relative to the current ethylene concentration local time series principal component representation vector, the time series gradient attenuation weight of each historical ethylene concentration local time series principal component representation vector is calculated to obtain a sequence of historical ethylene concentration characteristic time series gradient attenuation weights; Based on the sequence of the historical ethylene concentration feature time series gradient attenuation weights, the sequence of the real-time ethylene concentration local time series implicit correlation feature vectors is subjected to feature time series transfer aggregation to obtain the real-time ethylene concentration fine-grained time series correlation aggregation feature vector; A multi-source environmental parameter association analysis module is used to perform deep aggregation analysis on the real-time humidity fine-grained time series association aggregation feature vector, the real-time carbon dioxide concentration fine-grained time series association aggregation feature vector, the real-time oxygen concentration fine-grained time series association aggregation feature vector, and the real-time ethylene concentration fine-grained time series association aggregation feature vector to obtain a multi-source time series deep association representation vector of the in-reservoir environmental parameter; The temperature control module is used to determine the Ehime orange preservation temperature control strategy based on the multi-source time series deep association representation vector of the environmental parameters in the warehouse.

Images