Community group purchase intelligent cabinet management system and method

Abstract

The invention discloses a community group buying intelligent cabinet management system and method, and the system combines the basic information of a commodity, such as the type and name of the commodity, with the real-time temperature change condition of a cabinet grid where the commodity is located, and builds a more refined commodity deterioration early warning scheme.

Description

Technical Field

[0001] The present application relates to the field of intelligent management, and more specifically, to a community group purchasing intelligent cabinet management system and method. Background Art

[0002] In the community group-buying scenario, smart cabinets serve as the core carrier for commodity storage and delivery. Their temperature control management directly affects the quality assurance of fresh and frozen commodities. Existing technologies still have significant deficiencies in dynamic monitoring and risk warning of commodity storage environments. For example, for the full-process temperature control management of temperature-sensitive commodities such as frozen and fresh products, traditional solutions mostly rely on fixed temperature thresholds to trigger alarms. In addition, some more advanced solutions may also combine the shelf life information of the commodity to issue reminders when the shelf life is approaching, in order to reduce losses caused by expiration and deterioration. However, these existing early warning methods often focus on monitoring the ambient temperature, or rely solely on the inherent properties of the commodity itself, and rarely consider the dynamic correlation between the characteristics of the commodity itself and the actual storage environment.

[0003] In community group buying scenarios, products vary greatly, each with its own unique storage environment requirements. Relying solely on standardized environmental parameter thresholds or simple expiration date determinations makes it difficult to accurately assess each item's actual risk of spoilage. To more precisely manage products in community group buying smart lockers and improve the accuracy of product spoilage warnings, an optimized management solution for community group buying smart lockers is needed. Summary of the Invention

[0004] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a community group purchase smart cabinet management system and method, which combines the basic information of the product itself, such as product type and name, with the real-time temperature changes of the cabinet where it is located to build a more refined product deterioration warning solution.

[0005] According to one aspect of the present application, a community group buying smart cabinet management method is provided, which includes: obtaining commodity entry information, the commodity entry information including order ID, user ID, cabinet ID, commodity type, commodity name, required temperature zone and cabinet entry timestamp; based on the cabinet ID, extracting the time series of target cabinet status data from the background database, the target cabinet status data including cabinet ID, timestamp and real-time temperature; performing structured fusion coding on the commodity type and commodity name to obtain a structured embedded coding vector of the commodity basic information; performing data processing on the time series of the target cabinet status data to obtain a time series associated implicit coding vector of the commodity real-time temperature; performing feature-level commodity information-commodity temperature semantic collaborative verification on the structured embedded coding vector of the commodity basic information and the commodity real-time temperature time series associated implicit coding vector to obtain a commodity information-commodity temperature semantic collaborative verification response coding vector; based on the commodity information-commodity temperature semantic collaborative verification response coding vector, determining whether to generate a commodity deterioration warning prompt.

[0006] Compared with the existing technology, the community group purchasing smart cabinet management system and method provided in this application combines the basic information of the product itself, such as product type and name, with the real-time temperature changes of the cabinet where it is located to build a more refined product deterioration warning plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended 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 drawings, the same reference numerals generally represent the same components or steps.

[0008] Figure 1 This is a flowchart of the community group purchasing smart cabinet management method according to an embodiment of the present application.

[0009] Figure 2 This is a data flow diagram of the community group purchasing smart cabinet management method according to an embodiment of the present application.

[0010] Figure 3 This is a block diagram of the community group buying smart cabinet management system according to an embodiment of the present application. DETAILED DESCRIPTION

[0011] 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 herein.

[0012] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0013] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0014] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0015] 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 herein.

[0016] In the technical solution of this application, a community group purchasing smart cabinet management method is proposed. Figure 1 This is a flowchart of the community group purchasing smart cabinet management method according to an embodiment of the present application. Figure 2 This is a data flow diagram of the community group purchase smart cabinet management method according to the embodiment of the present application. Figure 1 and Figure 2As shown, the community group buying smart cabinet management method according to the embodiment of the present application includes the following steps: S1, obtaining the commodity entry information, the commodity entry information including the order ID, user ID, cabinet ID, commodity type, commodity name, required temperature zone and cabinet entry timestamp; S2, based on the cabinet ID, extracting the time series of the target cabinet status data from the background database, the target cabinet status data including the cabinet ID, timestamp and real-time temperature; S3, performing structured fusion coding on the commodity type and commodity name to obtain a structured embedded coding vector of the basic information of the commodity; S4, performing data processing on the time series of the target cabinet status data to obtain a time series associated implicit coding vector of the real-time temperature of the commodity; S5, performing feature-level commodity information-commodity temperature semantic collaborative verification on the structured embedded coding vector of the basic information of the commodity and the time series associated implicit coding vector of the real-time temperature of the commodity to obtain a commodity information-commodity temperature semantic collaborative verification response coding vector; S6, based on the commodity information-commodity temperature semantic collaborative verification response coding vector, determining whether to generate a commodity deterioration warning prompt.

[0017] In particular, the S1 obtains the information on the entry of goods into the cabinet, which includes the order ID, user ID, cabinet ID, product type, product name, required temperature zone and cabinet entry timestamp. Considering that the existing technology lacks the ability to deeply correlate the individual characteristics of goods with the dynamic changes of the storage environment when dealing with community group buying smart cabinets, especially the management of temperature-sensitive goods (such as fresh and frozen products). Therefore, in order to achieve accurate prediction and early warning of the risk of deterioration of goods, in the technical solution of this application, first of all, the basic information of each goods entering the cabinet and the specific environment in which they are located are mastered by obtaining the information on the entry of goods. Specifically, the order ID (commodity transaction record), user ID (associated with the person who picks up the goods), cabinet ID (specify the specific physical compartment number where the goods are stored), product type (such as frozen, refrigerated, room temperature, indicating the major temperature control category of the goods), product name (specific product identity), required temperature zone (the temperature range standard for ideal storage of goods) and cabinet entry timestamp (recording the exact time when the goods begin to be stored in the cabinet) together constitute a complete description of the product storage event.

[0018] In particular, the S2 extracts the time series of the target cabinet status data from the background database based on the cabinet ID. The target cabinet status data includes the cabinet ID, timestamp and real-time temperature. It should be understood that in the actual operation of the community group buying smart cabinet, the cabinet ID, as a unique identifier, is directly associated with the physical location, temperature zone configuration and historical status data of the cabinet, and is the core link for achieving precise temperature control management. In addition, different cabinets may store different temperature-sensitive commodities (such as frozen products requiring -18°C and fresh products requiring 0-4°C), and their temperature fluctuation patterns are closely related to the risk of commodity deterioration. If the correlation of the cabinet ID is ignored, it may lead to a mismatch between the temperature control strategy and commodity demand, such as the mistaken use of refrigerator temperature data for the risk assessment of freezer products. By extracting the time series of the target cabinet status data (including timestamp and real-time temperature) from the background database based on the cabinet ID, the system can accurately trace the temperature change trajectory of a specific cabinet.

[0019] Specifically, precise matching of cabinet IDs ensures a strong correlation between temperature time-series data and the commodity storage environment, enabling the system to identify short-term anomalies (such as a brief temperature rise caused by opening a door) or long-term temperature trends (such as temperature drift caused by equipment aging) and, based on this, predict spoilage risks. For example, if the temperature curve corresponding to a freezer cabinet ID shows small periodic fluctuations, the system can combine the properties of the frozen goods in that cabinet to determine whether the cabinet is losing cold due to frequent storage and retrieval, thereby triggering an early warning rather than passively waiting for the temperature to exceed the standard. In this process, the cabinet ID is not only an index for data extraction, but also the technical foundation for realizing "one cabinet, one policy" dynamic management, ensuring that temperature control warnings are highly compatible with the actual storage needs of the goods.

[0020] Specifically, S3 performs structured fusion coding on the product type and product name to obtain a structured embedded coding vector for the basic product information. That is, in this embodiment of the present application, the product type is first structured embedded coding using the product type embedding matrix to obtain a structured embedding vector for the product type. It should be understood that product types inherently contain implicit temperature control knowledge (e.g., frozen goods require -18°C, dairy products require 4°C), but this knowledge cannot be directly expressed through simple coding. For example, while "quick-frozen dumplings" and "ice cream" belong to the same frozen category, they may have different tolerances for temperature fluctuations. Structured embedding coding can capture these fine-grained semantic differences. Using the product type embedding matrix, the system can map discrete product category labels (e.g., "frozen seafood" and "quick-frozen pastries") into a continuous, high-dimensional vector space. This allows semantically similar product types (e.g., "frozen dumplings" and "frozen buns") to exhibit clustering characteristics in the vector space, while products with significantly different temperature control requirements (e.g., "ice cream" and "fresh strawberries") are automatically separated into different vector clusters. This structured embedded coding transforms prior knowledge about product types (such as biological characteristics and industry storage specifications) into quantitative features that can be parsed by machine learning models, thus breaking through the over-reliance of traditional rule engines on expert experience. This coding approach effectively builds a dynamic mapping bridge between a product's physical attributes and digital features, upgrading temperature monitoring logic from "rule-based conditional judgment" to "feature semantics-based associative reasoning," significantly improving adaptability to long-tail products and complex temperature fluctuation scenarios.

[0021] Next, the product name embedding matrix is ​​used to perform structured embedding encoding on the product name to obtain a structured embedding encoding vector for the product name. It should be understood that because different specific products within the same product type have significant differences in sensitivity to temperature fluctuations, traditional methods that rely solely on coarse-grained classification of product type labels (such as "frozen seafood") are unable to capture these fine-grained differences. Using the product name embedding matrix, the system can map the semantic information of the product name (such as brand, origin, and processing technology) into a high-dimensional continuous vector, encoding the product characteristics implicit in the name (such as modifiers such as "organic," "quickly frozen to lock in freshness," and "ready-to-eat salad") into quantifiable feature dimensions. This structured embedding encoding overcomes the limitations of traditional classification systems in representing personalized product attributes. Through deep learning, it automatically mines the latent semantic associations related to temperature control requirements in the product name text, rather than relying solely on manually preset temperature zone labels. This encoding mechanism essentially establishes a digital bridge between the natural language description of the product and the physical deterioration mechanism, enabling the system to automatically extract implicit factors affecting qualitative change (such as processing methods and ingredient ratios) from massive product names, thereby achieving accurate personalized risk modeling in the complex and changeable community group purchasing product library, and significantly reducing the misjudgment rate caused by the loss of semantic information in the product name.

[0022] Furthermore, the product type structured embedding vector and the product name structured embedding coding vector are cascaded to obtain the product basic information structured embedding coding vector. It should be understood that although the product type embedding vector can reflect the temperature control rules of large categories (such as frozen products require -18°C), it cannot distinguish the subtle differences between similar products (such as "ice cream" is more sensitive to short-term warming than "frozen meat"); and the product name embedding vector, although it contains specific product features, lacks global constraints on temperature control categories. If only relying on a single attribute encoding, it is difficult to fully characterize the response mechanism of the product to temperature fluctuations. Therefore, in order to construct a feature representation that can more comprehensively reflect the individual temperature control needs of the product, in the technical solution of this application, the product type structured embedding vector and the product name structured embedding coding vector are cascaded to obtain the product basic information structured embedding coding vector.

[0023] That is, by cascading vectors to fuse the macro-category characteristics and micro-individual characteristics of product attributes, a multi-dimensional semantic expression is constructed, enabling subsequent semantic collaborative verification to simultaneously capture the universal association of "type-temperature" and the specific association of "name-temperature." The generated structured embedded coding vector of basic product information not only reflects the broad category (type) of the product, nor is it just isolated name information, but is an organic combination of the two, capable of more precisely capturing the specific temperature control requirements and potential quality degradation patterns of specific products. This enables subsequent semantic collaborative verification to perform more accurate matching and reasoning, providing higher-quality, more discriminative input features for subsequent machine learning-based risk prediction models, thereby greatly improving the accuracy and timeliness of product deterioration warnings, effectively ensuring product quality, and reducing losses.

[0024] In particular, the S4 processes the time series of the target cabinet status data to obtain the implicit coding vector associated with the real-time temperature time series of the commodity. That is, in an embodiment of the present application, first, the time series of the target cabinet status data is sorted based on the timestamp to obtain a time queue of the real-time temperature value of the commodity. It should be understood that since the original temperature data stream collected by the smart cabinet sensor usually has problems such as disordered timestamps (such as network delays causing misordered data packets), inconsistent sampling frequencies (such as congestion in sensor data reporting during peak hours), and cross-cabinet data mixing (transmitting multiple cabinet states in the same batch), if the original data queue is used directly for temperature analysis, the system will not be able to accurately capture the true evolution law of temperature fluctuations. Therefore, in the technical solution of the present application, the time series of the target cabinet status data is sorted based on the timestamp to obtain a time queue of the real-time temperature value of the commodity.

[0025] During this process, first, all extracted data records are accurately sorted based on the timestamp information attached to each data point, ensuring that the temperature readings are arranged in chronological order starting from the timestamp of the product entering the cabinet (obtained in the first step); then, the core indicator of real-time temperature is filtered and only retained, removing other irrelevant information that may exist. The generated time queue of real-time temperature values ​​of the products not only ensures the temporal correctness of the data, that is, each temperature value accurately corresponds to the time point at which it occurs, and the entire sequence strictly follows the order of time passage, but also ensures the relevance of the data, that is, it only contains temperature records during the actual storage period of the product. In this way, the model can accurately learn the complex patterns of temperature changes over time, capture short-term abnormal fluctuations or long-term cumulative trends, thereby effectively overcoming the shortcomings of the existing technologies mentioned in the technical background, such as insufficient extraction of time-series correlation features and inability to accurately predict risks.

[0026] Furthermore, the time queue of the real-time temperature value of the commodity is encoded into a temperature sequence to obtain an implicit coding vector associated with the time series of the real-time temperature of the commodity. It should be understood that temperature data, as a numerical change at continuous time points, has obvious temporal and dynamic characteristics. Simply relying on static thresholds or simple statistical analysis is difficult to fully reflect its potential risks. In the technical solution of the present application, an LSTM model is used to deeply model the subtle fluctuations of temperature over time to extract the associated features hidden in the temperature fluctuations. Through the LSTM encoder, the system can convert complex and continuously changing temperature data into an implicit coding vector with concentrated semantic information. This vector not only contains the current temperature state, but also contains key features such as the temperature change trend, fluctuation amplitude, and abnormal fluctuation pattern over the past period of time. This deep feature expression helps subsequent risk prediction models to more accurately identify potential deterioration risks, rather than relying solely on instantaneous or local anomaly detection. By introducing the LSTM model for temperature sequence encoding, not only the system's sensitivity to dynamic environmental changes is enhanced, but also the accuracy and robustness of the overall monitoring and early warning system are improved. Compared to traditional methods that rely solely on static thresholds or simple statistical indicators, the implicit code vectors extracted using LSTM can more comprehensively and meticulously reflect the complex temporal relationships within the product storage environment, effectively reducing false positives and missed positives and improving early warning response speed. This ultimately helps ensure the quality of fresh and frozen products, enhances user trust, reduces economic losses caused by product spoilage, and ensures the efficient and secure operation of community group-buying smart cabinet management systems in practical applications.

[0027] In particular, S5 performs a feature-level semantic co-verification of the product information and product temperature on the structured embedded encoding vector of the basic product information and the implicit encoding vector associated with the product's real-time temperature time series, to obtain a response encoding vector for the semantic co-verification of the product information and product temperature. It should be understood that the impact of product attributes (such as protein content and shelf life) and temperature fluctuations (such as instantaneous jumps and gradual shifts) on product deterioration is not a simple linear superposition relationship, but rather involves complex cross-modal interactions. Existing technologies typically process product attributes and temperature data independently, associating them through simple logical rules or feature splicing. This fails to capture the nonlinear dynamic coupling relationship between product physical properties (such as ingredient stability and microbial activity) and ambient temperature fluctuations (such as periodic oscillations and abnormal shifts). Therefore, to break through the limitations of traditional single-point thresholds or static rules, a dynamic knowledge graph of commodity storage status is established to mine the multi-dimensional evolution of commodity storage status at the fine-grained feature level and achieve high-order semantic conversion from raw data to latent variables of spoilage risk. In the technical solution of this application, a feature-level commodity information-commodity temperature semantic collaborative verification is performed on the structured embedded coding vector of the commodity basic information and the implicit coding vector associated with the commodity real-time temperature time series to obtain the commodity information-commodity temperature semantic collaborative verification response coding vector. In other words, through hierarchical feature decoupling and serialized reasoning mechanisms, a deep collaborative mapping of the commodity attribute semantic space and the temperature time series feature space is constructed. Specifically, one-dimensional convolutional coding is first used to extract local implicit features from both the product attribute vector and the temperature time series vector, effectively stripping away noise from the original features and focusing on key pattern recognition (e.g., biochemical reaction rate features implicit in the product type vector, or gradual trend features representing cumulative effects in the temperature series). Next, a single semantic feature interaction engine learns microscopic correlation patterns between product attributes and temperature fluctuations (e.g., the sensitivity of specific component degradation rates to the frequency of temperature fluctuations) through parameterized interaction functions (e.g., multiplicative interactions or attention gating). Finally, attention weights are calculated based on the distributional properties of the interaction response vectors (e.g., entropy and variance), quantifying the contributions of different local interaction units to overall risk prediction. For example, temperature stability during the initial storage phase may be more predictive than transient fluctuations later in the storage phase, while the buffering effect of specific product attributes (e.g., antioxidant content) against temperature excursions may vary dynamically with storage duration. Finally, sequential inference using a forward LSTM integrates the temporal evolution of the product information-product temperature local semantic interaction units to form a global risk assessment encoding feature.The generated product information-product temperature semantic collaborative verification response encoding vector provides extremely rich and highly relevant input features for the subsequent machine learning risk prediction model. This enables the prediction model to more accurately quantify the probability of product deterioration, thereby triggering early warnings before the product quality substantially deteriorates, buying valuable time for timely intervention (such as adjusting cabinet temperature, notifying operations and maintenance, and reminding users to pick up goods as soon as possible), ultimately effectively ensuring product quality, reducing losses, and maintaining user trust.

[0028] Specifically, the system first performs local latent feature extraction based on one-dimensional convolutional coding on the structured embedded code vector of the product's basic information and the implicit code vector associated with the product's real-time temperature, respectively. This yields a set of local latent feature code vectors for the product's basic information and a set of local latent feature code vectors for the product's real-time temperature. It should be understood that the impact of product attributes (such as the fat content of a "frozen steak") and temperature fluctuations (such as short-term temperature rise events) on spoilage risk is not globally linearly correlated, but rather nonlinearly coupled within local sensitive regions. Traditional methods that directly use fully connected layers or simple concatenation can overwhelm the semantic value of key local signals. Using one-dimensional convolutional coding, the system scans the structured embedded code vector of the product's basic information and the implicit code vector associated with the product's real-time temperature in a sliding window manner to capture local correlation patterns at different scales.

[0029] Specifically, local feature extraction of product attribute vectors focuses on the combined effects of the product's physical properties (e.g., the synergistic effect of the dough moisture content and the fat content of the filling in frozen dumplings), while local encoding of the temperature vector emphasizes the temporal correlation of fluctuation patterns (e.g., the instantaneous temperature jump caused by opening the door to pick up the product and the delayed effect of the refrigeration system recovery). Through adaptive learning of parameterized convolutional kernel weights, these local feature sets form local mappings between the semantic space of product attributes and the space of temperature fluctuation patterns, respectively, providing interpretable feature anchors for cross-modal interaction.

[0030] In a specific example of the present application, the following one-dimensional convolution formula is used to perform local implicit feature extraction based on one-dimensional convolution coding on the structured embedded coding vector of the basic information of the product and the implicit coding vector associated with the time series of the real-time temperature of the product to obtain a set of local implicit feature coding vectors of the basic information of the product and a set of local implicit feature coding vectors of the real-time temperature of the product; wherein the one-dimensional convolution formula is: ;in, The structural embedding coding vector of the basic information of the product, represents a one-dimensional convolution operation, are the first, second, and third in the set of local implicit feature coding vectors of the basic information of the product. and The local implicit feature encoding vector of the basic information of each product, is the implicit coding vector associated with the real-time temperature time series of the commodity, are the first, second, and third in the set of local implicit feature encoding vectors of the real-time temperature of the commodity. and The local implicit feature encoding vector of the real-time temperature of each product.

[0031] Next, each set of local latent feature encoding vectors for product basic information and local latent feature encoding vectors for product real-time temperature is fed into the product information-product temperature single semantic feature interaction engine to generate a set of product information-product temperature local semantic latent feature interaction response encoding vectors. It should be understood that the impact of product attributes (such as ingredient stability and processing technology) and temperature fluctuations (such as periodic oscillations and abnormal excursions) on spoilage risk is not a simple linear superposition, but rather involves complex cross-modal local interactions. Traditional methods using global feature concatenation or single threshold rules struggle to capture these nonlinear associations. Using the single semantic feature interaction engine, the system constructs a fine-grained association model between product attributes and temperature time series at the local latent feature level. Furthermore, by pairwise feeding the local latent feature encoding vectors for product basic information and product real-time temperature into the interaction engine, the system effectively mitigates interference from noise and redundant information in the original feature space. For example, filtering out interference signals from transient temperature jumps caused by cabinet door openings on non-sensitive products. This fine-grained, decoupled interactive representation greatly reduces the complexity of subsequent modeling and provides structured, information-rich input for subsequent chain reasoning, thus laying a solid foundation for ultimately achieving accurate and early warning of commodity deterioration risks.

[0032] In a specific example of the present application, the following feature interaction formula is used to input each corresponding set of product basic information local implicit feature coding vectors and product real-time temperature local implicit feature coding vectors in the set of product basic information local implicit feature coding vectors and the set of product real-time temperature local implicit feature coding vectors into the product information-product temperature monomer semantic feature interaction engine to obtain a set of product information-product temperature local semantic implicit feature interaction response coding vectors; wherein, the feature interaction formula is: ;in, Indicates cascade, Indicates point multiplication by position, It means adding by position. Indicates difference by position. and denote the weight matrix and bias vector respectively, express and The corresponding product information-product temperature local semantic implicit feature interaction response encoding vector.

[0033] Next, based on the characteristic distribution characteristics of each product information-product temperature local semantic implicit feature interaction response encoding vector in the set of product information-product temperature local semantic implicit feature interaction response encoding vectors, the chained reasoning semantic attention weights of each product information-product temperature local semantic implicit feature interaction response encoding vector are determined to obtain the initial set of product information-product temperature chained reasoning semantic attention weights. It should be understood that due to the dynamic differences in the interaction effects between different product attributes (such as ingredient stability and shelf life) and temperature fluctuation patterns (such as short-term jumps and gradual shifts), traditional methods using fixed weights or manual experience-based assignments cannot adapt to the risk evolution patterns in complex scenarios. For example, the temperature stability of refrigerated yogurt during the initial storage period has a higher impact on lactic acid bacteria activity, while the penalty coefficient for short-term temperature rise events near the expiration date needs to be dynamically increased. The sensitivity of frozen seafood to ice crystal formation is nonlinearly related to the frequency of temperature fluctuations. By calculating statistical indicators of the interaction response vectors (such as entropy to reflect information complexity and norm to represent interaction strength), the system can quantify the risk contribution of interaction units in different spatiotemporal dimensions, for example, assigning higher weights to sensitivity to high-frequency, small fluctuations during the storage period. Specifically, by analyzing the local interactive response distribution between product attributes and temperature time series, the system can identify key risk triggering nodes (such as a sudden change in the evaporation rate of high-moisture products under specific temperature fluctuation patterns) and adjust attention weights accordingly. This mechanism, by constructing a dynamic weight mapping network, significantly improves the timeliness of early warnings and the credibility of decisions for multi-category products in community group buying scenarios, transforming the delayed warnings of traditional solutions into predictive interventions based on the risk evolution path.

[0034] In a specific example of the present application, the following attention formula is used to determine the chain reasoning semantic attention weight of each product information-product temperature local semantic implicit feature interaction response encoding vector to obtain a set of initial product information-product temperature chain reasoning semantic attention weights; wherein, the attention formula is: ;in, express function, The first in the set of interactive response encoding vectors representing the local semantic implicit features of product information and product temperature The feature value of the jth position of the interaction response encoding vector of the product information-product temperature local semantic implicit feature, represents the square of the norm, Indicates the length of the interaction response encoding vector of the local semantic implicit feature of product information and product temperature, express The corresponding initial product information-product temperature chain reasoning semantic attention weight.

[0035] Then, the initial set of semantic attention weights for the chained reasoning of product information and product temperature is spatially modified based on the semantic interaction norm of product information and product temperature to obtain the set of semantic attention weights for the chained reasoning of product information and product temperature. It is understandable that due to the spatial heterogeneity of the interaction effects between product attributes (e.g., fat oxidation sensitivity of frozen steak) and temperature fluctuation patterns (e.g., gradual temperature rise or sudden jump), the contributions of different interaction norms (e.g., translation and fluctuation) to spoilage risk exhibit a nonlinear coupling relationship. For example, when refrigerated yogurt experiences a temperature jump caused by a brief door opening, the risk of lactic acid bacteria activity decay is not only related to the magnitude of the instantaneous temperature deviation but also to the spatial synergistic influence of the temperature recovery rate (fluctuation) and storage location (translation). Products located near the door, due to delayed temperature recovery, have a significantly higher cumulative risk effect than those located inside. Traditional methods that directly use initial attention weights (e.g., weights calculated based on variance) have difficulty distinguishing these differences in spatial interaction norms, leading to systematic biases in risk prediction. Therefore, in a preferred example of the present application, the set of initial product information-product temperature chain reasoning semantic attention weights is spatially corrected based on the product information-product temperature semantic interaction specification to obtain the set of product information-product temperature chain reasoning semantic attention weights.

[0036] Specifically, by decomposing the spatial characteristics of interaction specifications, fine-grained calibration of attention weights is achieved. Specifically, the system divides the interaction between product attributes and temperature time series into translational effects (where the gradient direction aligns with the spatial interaction direction, such as the temperature distribution gradient caused by differences in shelf position) and fluctuation effects (where the gradient direction is orthogonal, such as local oscillations caused by sudden temperature jumps), and calculates periodic compensation factors for each. For example, for high-moisture leafy vegetables, the system identifies that the risk of moisture evaporation is primarily driven by translational effects (sustained temperature rise at the shelf edges), requiring increased weight compensation in the translational direction. Conversely, the risk of ice crystal formation in frozen seafood is strongly correlated with fluctuation effects (the frequency of temperature jumps), requiring weight adjustment using a fluctuation compensation factor. This spatial correction mechanism dynamically decouples the influence of different interaction specifications through mathematical modeling (e.g., using a logarithmic size growth function to compensate for translational effects and a periodic phase to correct for fluctuation effects), enabling attention weights to accurately reflect differences in risk transmission paths across spatial dimensions. This correction mechanism reduces the false alarm rate of spoilage risk predictions while improving the efficiency of identifying high-risk events, significantly optimizing the proactive management capabilities and user experience of smart lockers.

[0037] In this example, the set of initial product information-product temperature chain reasoning semantic attention weights is spatially corrected based on the product information-product temperature semantic interaction specification to obtain the set of product information-product temperature chain reasoning semantic attention weights using the following correction formula; wherein the correction formula is: ;in, and represents the translation action vector statistics, represents the statistics of the wave action vector, represents periodic local regular compensation, represents the periodic local auxiliary phase transition, and represents the modulation weight, express Corresponding product information-product temperature chain reasoning semantic attention weight.

[0038] Specifically, when calculating the interaction response encoding vector of each product information-product temperature local semantic implicit feature, the interaction features between the corresponding product basic information local implicit feature encoding vector and the product real-time temperature local implicit feature encoding vector are introduced, for example Etc., that is, they essentially correspond to different interaction norms, and thus have different spatial auxiliary constraint associations in the interaction space.

[0039] Therefore, in order to enhance the norm variability assistance in chain reasoning attention weights, it is preferred to modify the set of semantic attention weights of the initial product information-product temperature chain reasoning based on the spatial action decomposition of the interaction norm. Specifically, is regarded as a translation effect, that is, the gradient direction is consistent with the spatial interaction direction, and It is regarded as a wave effect, that is, the gradient direction is orthogonal to the spatial interaction direction. In this way, for vector statistics of different effects, for example , it is believed that the wave action leads to a localized periodic effect in the translation direction, and thus the periodic local regular compensation is calculated as: That is, As a localized representation of translation, the wave action It increases with the logarithmic size of the translation domain.

[0040] On the other hand, the wave action will also lead to a localized translational phase transition, thus obtaining a periodic local auxiliary phase transition: ; Therefore, based on the weighted sum of the above two items, the initial product information-product temperature chain reasoning semantic attention weight is modified : .

[0041] By focusing on the differences in the normative roles of different interaction norms corresponding to the interaction responses of the local semantic implicit features of each product information-product temperature in the interaction space, the correlation between different spatial auxiliary constraint norms in the interaction space can be improved, thereby improving the calculation accuracy of the semantic attention weight of the product information-product temperature chain reasoning.

[0042] Furthermore, based on the set of semantic attention weights derived from the chained reasoning of product information and product temperature, the set of interaction response encoding vectors of the local semantic latent features of product information and product temperature is weighted and modulated to obtain the set of modulated interaction response encoding vectors of the local semantic latent features of product information and product temperature. Because the interaction effects between different product attributes (e.g., fat oxidation rate, microbial activity) and temperature fluctuation patterns (e.g., sudden jumps, sustained excursions) vary significantly, traditional methods that assign equal weights to all local interaction units can drown out key risk signals. For example, when refrigerated yogurt experiences multiple short fluctuations of ±1°C, only the fluctuations in a specific time window (e.g., mid-storage) and the interaction with product attributes (e.g., probiotic content) are decisive for spoilage risk. Similar fluctuations during other time periods may be less affected by the product's inherent buffering properties (e.g., vacuum packaging). In this application's technical solution, by introducing an attention weight modulation mechanism, the system dynamically enhances high-value interaction signals (such as the sensitivity of initial storage temperature stability to short-shelf-life products) and suppresses low-value or noisy signals (such as the impact of instantaneous temperature jumps on frozen dumplings caused by opening the door to pick up goods), thus overcoming the limitations of traditional uniform weighting strategies. This dynamic modulation enables the model to adapt to the risk perception needs of different storage stages (such as initial stability monitoring and later cumulative effect assessment) and shelf locations (such as edge areas susceptible to disturbance).

[0043] In a specific example of the present application, the set of product information-product temperature local semantic implicit feature interaction response coding vectors is weighted modulated using the following modulation formula to obtain a set of modulated product information-product temperature local semantic implicit feature interaction response coding vectors; wherein the modulation formula is: ;in, is the set of interactive response encoding vectors of the product information-product temperature local semantic implicit feature, are the first, second, and third in the set of the interaction response encoding vectors of the modulated product information-product temperature local semantic implicit feature. and Modulated product information-product temperature local semantic implicit feature interaction response encoding vector.

[0044] Subsequently, the set of modulated product information-product temperature local semantic implicit feature interaction response encoding vectors is input into a chained inference engine based on a forward LSTM model to generate a product information-product temperature semantic collaborative verification response encoding vector. It should be understood that while local interaction features can capture fine-grained associations between product attributes and temperature fluctuations (such as the interaction between temperature jumps and product composition during a certain period), these features are discrete and lack temporal logic connection, making it difficult to model the dynamic evolution of risk factors throughout the entire product storage cycle. For example, the impact of multiple short-term temperature increases on products may have a cumulative effect, requiring serialized reasoning to capture their long-term effects. Therefore, in the technical solution of this application, the set of modulated product information-product temperature local semantic implicit feature interaction response encoding vectors (such as product-temperature interaction features in different time windows) is gradually integrated in chronological order using the recurrent structure and gating mechanism of the forward LSTM. The historical interaction state is retained through memory cells, and key information is filtered based on the gating mechanism to ultimately generate a global semantic collaborative verification code. This process is essentially a chain aggregation and logical deepening of local risk signals, enabling the model to understand "how local anomalies trigger global deterioration risks" from a temporal dimension.

[0045] Here, the chained reasoning of the forward LSTM overcomes the limitations of traditional static feature fusion, identifying the complex temporal correlations between temperature fluctuation patterns (such as periodicity and trends) and product attributes (such as ingredient stability and shelf life), thereby predicting hidden risks (such as slow oxidation and microbial proliferation) within the compliant temperature range. This global feature representation significantly improves the robustness of spoilage risk prediction, enabling the system to trigger warnings based on dynamic reasoning rather than isolated thresholds. Ultimately, this enables an upgrade in management from passive alerts to proactive risk intervention, optimizing product loss control and user experience.

[0046] In a specific example of the present application, the set of modulated product information-product temperature local semantic implicit feature interaction response encoding vectors is input into a chain inference engine based on a forward LSTM model using the following inference formula to obtain a product information-product temperature semantic collaborative verification response encoding vector; wherein, the inference formula is: ;in, Indicates forward , The encoding vector of the product information-product temperature semantic collaborative verification response.

[0047] In particular, the S6 determines whether to generate a product deterioration warning prompt based on the product information-product temperature semantic collaborative verification response coding vector. That is, in an embodiment of the present application, first, the product information-product temperature high-dimensional collaborative verification response coding vector is input into the trained machine learning risk prediction model to obtain the product deterioration risk probability value. Specifically, in the technical solution of the present application, the product information-product temperature high-dimensional collaborative verification response coding vector is input into the classifier-based product deterioration risk prediction model to obtain the product deterioration risk probability value. It should be understood that although the product information-product temperature high-dimensional collaborative verification response coding vector deeply integrates the inherent attributes of the product (such as the temperature-sensitive characteristics implied by the type and name) and the dynamic changes in the actual storage environment (temperature time series characteristics) and the complex interactive verification results between the two, its high-dimensional abstract characteristics cannot be directly mapped into quantifiable risk indicators. Traditional models lack the ability to decouple high-dimensional nonlinear relationships and are difficult to extract key risk signals from complex features. In the technical solution of this application, the high-dimensional collaborative verification response encoding vector of product information and product temperature is input into a trained classifier model (such as a deep neural network, ensemble learning, etc.). This is to leverage the powerful nonlinear pattern recognition capabilities of the machine learning algorithm to effectively interpret these abstract features and convert multi-source heterogeneous information into intuitive and usable spoilage risk probability values. This method, based on the classifier output probability value, enables managers to flexibly set thresholds based on actual business needs, implement more flexible quality assurance strategies, effectively reduce losses and enhance user trust, provide more targeted operational recommendations and service experiences for operation and maintenance personnel and consumers, and fundamentally promote the upgrade of community group purchasing supply chain management to data-driven and intelligent decision-making.

[0048] Furthermore, based on the comparison between the commodity deterioration risk probability value and a preset threshold, it is determined whether to generate a commodity deterioration warning prompt. Based on the comparison between the commodity deterioration risk probability value and a preset threshold, it is determined whether to generate a commodity deterioration warning prompt. Different commodities have significant differences in their tolerance to temperature fluctuations (e.g., frozen seafood is more sensitive to short-term temperature increases, while pre-packaged foods have stronger tolerance), and the deterioration risk has a nonlinear cumulative characteristic (e.g., the superposition effect of multiple small temperature increases). Traditional methods are unable to quantify such complex associations, while the risk value output by the probability model can comprehensively characterize the combined impact of commodity characteristics, temperature time series patterns, and storage time. Therefore, in the technical solution of the present application, a dynamic threshold mechanism is used to achieve refined adaptation of the warning strategy. In one example, in response to the commodity deterioration risk probability value being greater than or equal to the preset threshold, a commodity deterioration warning prompt is generated; otherwise, no commodity deterioration warning prompt is generated. This model effectively reduces the loss rate of commodities and realizes the management upgrade from "passive response alarm" to "active risk prevention."

[0049] In summary, the community group buying smart cabinet management method according to the embodiment of the present application is explained, which combines the basic information of the product itself, such as product type and name, with the real-time temperature changes of the cabinet in which it is located, to construct a more refined product deterioration warning plan.

[0050] Furthermore, a community group purchasing smart cabinet management system is also provided.

[0051] Figure 3 This is a block diagram of the community group purchase smart cabinet management system according to the embodiment of the present application. Figure 3 As shown, according to the community group buying smart cabinet management system 300 of the embodiment of the present application, it includes: a commodity cabinet information acquisition module 310, which is used to obtain commodity cabinet information, and the commodity cabinet information includes order ID, user ID, cabinet ID, commodity type, commodity name, required temperature zone and cabinet entry timestamp; a target cabinet state data extraction module 320, which is used to extract the time series of target cabinet state data from the background database based on the cabinet ID, and the target cabinet state data includes cabinet ID, timestamp and real-time temperature; a structured fusion coding module 330, which is used to perform structured fusion coding on the commodity type and commodity name to obtain the basic information of the commodity. The structured embedded coding vector of the basic information of the commodity is embedded in the data processing module 340; the data processing module 340 is used to process the time series of the target cabinet status data to obtain the implicit coding vector associated with the time series of the real-time temperature of the commodity; the semantic collaborative verification module 350 is used to perform feature-level commodity information-commodity temperature semantic collaborative verification on the structured embedded coding vector of the basic information of the commodity and the implicit coding vector associated with the time series of the real-time temperature of the commodity to obtain the commodity information-commodity temperature semantic collaborative verification response coding vector; the commodity deterioration warning prompt module 360 ​​is used to determine whether to generate a commodity deterioration warning prompt based on the commodity information-commodity temperature semantic collaborative verification response coding vector.

[0052] As described above, the community group purchasing smart cabinet management system 300 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a community group purchasing smart cabinet management algorithm. In one possible implementation, the community group purchasing smart cabinet management system 300 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the community group purchasing smart cabinet management system 300 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the community group purchasing smart cabinet management system 300 can also be one of the many hardware modules of the wireless terminal.

[0053] Alternatively, in another example, the community group purchasing smart cabinet management system 300 and the wireless terminal may also be separate devices, and the community group purchasing smart cabinet management system 300 may be connected to the wireless terminal through a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0054] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A community group purchase smart cabinet management method, characterized in that: include: Obtain product entry information, which includes order ID, user ID, compartment ID, product type, product name, required temperature zone, and entry timestamp; based on the compartment ID, extract the time series of target compartment status data from the background database, which includes the compartment ID, timestamp, and real-time temperature; perform structured fusion encoding on the product type and product name to obtain a structured embedded coding vector of the basic product information; perform data processing on the time series of the target compartment status data to obtain a time-series associated implicit coding vector of the real-time temperature of the product; perform feature-level product information-product temperature semantic collaborative verification on the structured embedded coding vector of the basic product information and the time-series associated implicit coding vector of the real-time temperature of the product to obtain a product information-product temperature semantic collaborative verification response coding vector; based on the product information-product temperature semantic collaborative verification response coding vector, determine whether to generate a product deterioration warning prompt.

2. The community group purchase smart cabinet management method according to claim 1 is characterized in that: The product type and product name are structured fusion-encoded to obtain a structured embedded coding vector of the basic product information, including: using the product type embedding matrix to perform structured embedded coding on the product type to obtain a structured embedded coding vector of the product type; using the product name embedding matrix to perform structured embedded coding on the product name to obtain a structured embedded coding vector of the product name; and cascading the structured embedded coding vector of the product type and the structured embedded coding vector of the product name to obtain a structured embedded coding vector of the basic product information.

3. The community group purchase smart cabinet management method according to claim 2 is characterized in that: Data processing is performed on the time series of the target cabinet status data to obtain an implicit coding vector associated with the real-time temperature time series of the product, including: data sorting is performed on the time series of the target cabinet status data based on the timestamp to obtain a time queue of the real-time temperature value of the product; temperature sequence encoding is performed on the time queue of the real-time temperature value of the product to obtain an implicit coding vector associated with the real-time temperature time series of the product.

4. The community group purchase smart cabinet management method according to claim 3 is characterized in that: The method includes performing temperature sequence encoding on a time queue of real-time temperature values ​​of a commodity to obtain a temporally associated implicit encoding vector of the real-time temperature of the commodity, including: using a temperature sequence encoder based on an LSTM model to perform temperature sequence encoding on the time queue of real-time temperature values ​​of the commodity to obtain a temporally associated implicit encoding vector of the real-time temperature of the commodity.

5. The community group purchase smart cabinet management method according to claim 4 is characterized in that: A feature-level product information-product temperature semantic collaborative verification is performed on the structured embedded coding vector of the basic product information and the implicit coding vector associated with the real-time temperature of the product to obtain a product information-product temperature semantic collaborative verification response coding vector, including: calculating a set of product information-product temperature local semantic implicit feature interaction response coding vectors based on the local implicit coding features of the structured embedded coding vector of the basic product information and the implicit coding vector associated with the real-time temperature of the product; performing feature modulation on the set of product information-product temperature local semantic implicit feature interaction response coding vectors based on the chain reasoning attention mechanism based on the feature distribution characteristics of each product information-product temperature local semantic implicit feature interaction response coding vector in the set of product information-product temperature local semantic implicit feature interaction response coding vectors to obtain a set of modulated product information-product temperature local semantic implicit feature interaction response coding vectors; and inputting the set of modulated product information-product temperature local semantic implicit feature interaction response coding vectors into a chain reasoning engine based on a forward LSTM model to obtain a product information-product temperature semantic collaborative verification response coding vector.

6. The community group purchase smart cabinet management method according to claim 5 is characterized in that: Based on the local implicit coding features of the structured embedded coding vector of the basic information of the product and the implicit coding vector associated with the time series of the real-time temperature of the product, a set of interactive response coding vectors of the local semantic implicit features of the product information and the product temperature is calculated, including: performing local implicit feature extraction based on one-dimensional convolutional coding on the structured embedded coding vector of the basic information of the product and the implicit coding vector associated with the time series of the real-time temperature of the product to obtain a set of local implicit feature coding vectors of the basic information of the product and a set of local implicit feature coding vectors of the real-time temperature of the product; inputting each corresponding group of local implicit feature coding vectors of the basic information of the product and the local implicit feature coding vectors of the real-time temperature of the product into the product information-product temperature monomer semantic feature interaction engine to obtain a set of interactive response coding vectors of the local semantic implicit features of the product information and the product temperature.

7. The community group purchase smart cabinet management method according to claim 6 is characterized in that: Based on the characteristic distribution characteristics of each product information-product temperature local semantic implicit feature interaction response encoding vector in the set of product information-product temperature local semantic implicit feature interaction response encoding vectors, the set of product information-product temperature local semantic implicit feature interaction response encoding vectors is subjected to feature modulation based on the chain reasoning attention mechanism to obtain a set of modulated product information-product temperature local semantic implicit feature interaction response encoding vectors, including: based on the characteristic distribution characteristics of each product information-product temperature local semantic implicit feature interaction response encoding vector in the set of product information-product temperature local semantic implicit feature interaction response encoding vectors, determining each product information-product temperature The chain reasoning semantic attention weights of the local semantic implicit feature interaction response coding vectors are used to obtain a set of initial product information-product temperature chain reasoning semantic attention weights; the set of initial product information-product temperature chain reasoning semantic attention weights is spatially corrected based on the product information-product temperature semantic interaction specification to obtain a set of product information-product temperature chain reasoning semantic attention weights; based on the set of product information-product temperature chain reasoning semantic attention weights, the set of product information-product temperature local semantic implicit feature interaction response coding vectors is weighted modulated to obtain a set of modulated product information-product temperature local semantic implicit feature interaction response coding vectors.

8. The community group purchase smart cabinet management method according to claim 7 is characterized in that: Based on the product information-product temperature semantic collaborative verification response coding vector, determining whether to generate a product deterioration warning prompt includes: inputting the product information-product temperature high-dimensional collaborative verification response coding vector into a trained machine learning risk prediction model to obtain a product deterioration risk probability value; based on a comparison between the product deterioration risk probability value and a preset threshold, determining whether to generate a product deterioration warning prompt.

9. The community group purchase smart cabinet management method according to claim 8, characterized in that: Inputting the high-dimensional collaborative verification response encoding vector of commodity information and commodity temperature into the trained machine learning risk prediction model to obtain the commodity deterioration risk probability value, including: inputting the high-dimensional collaborative verification response encoding vector of commodity information and commodity temperature into the classifier-based commodity deterioration risk prediction model to obtain the commodity deterioration risk probability value.

10. A community group purchase smart cabinet management system, characterized in that: include: The product entry information acquisition module is used to obtain product entry information, which includes the order ID, user ID, cabinet ID, product type, product name, required temperature zone, and entry timestamp. The target cabinet status data extraction module is used to extract the time series of target cabinet status data from the backend database based on the cabinet ID. The target cabinet status data includes the cabinet ID, timestamp, and real-time temperature. The structured fusion coding module is used to perform structured fusion coding on the product type and product name to obtain a structured embedded coding vector of the basic product information. A data processing module is used to process the time series of the target cabinet status data to obtain the implicit coding vector associated with the real-time temperature of the product; A semantic collaborative verification module is used to perform feature-level semantic collaborative verification of product information and product temperature on the structured embedded coding vector of the product basic information and the implicit coding vector associated with the product's real-time temperature time series to obtain a product information and product temperature semantic collaborative verification response coding vector; The commodity deterioration warning prompt module is used to collaboratively verify the response coding vector based on the commodity information-commodity temperature semantics to determine whether to generate a commodity deterioration warning prompt.

Images