Intelligent cold chain logistics distribution method and system based on real-time monitoring of fresh milk quality

Abstract

The present application relates to fresh milk logistics distribution technical field, especially intelligent cold chain logistics distribution method and system based on real-time monitoring of fresh milk quality, the method is through analyzing historical cold chain data, identifying typical cold chain scene and constructing scene model, extracting environmental characteristic parameter, generating corresponding quality influence label; Combined with fresh milk real-time monitoring data, using space-time fusion analysis mechanism, output fresh milk in each cold chain link quality state trajectory and abnormal data; The above information is mapped to the digital twin platform, and the distribution path graph structure and the multi-dimensional quality correlation graph are constructed; Based on the graph analysis of the distribution time limit and the quality guarantee constraint boundary, the cold chain resource scheduling data and the distribution priority are fused, the dynamic distribution adjustment window set is generated, the intelligent distribution optimization strategy is made, and the intelligent decision-making ability of the fresh milk cold chain distribution is improved. Under the premise of guaranteeing the quality and safety of fresh milk, the distribution path and resource scheduling are dynamically optimized.

Description

Technical Field

[0001] This invention relates to the field of fresh milk logistics and distribution technology, and in particular to an intelligent cold chain logistics and distribution method and system based on real-time monitoring of fresh milk quality. Background Technology

[0002] With consumers increasingly demanding higher safety and quality standards for dairy products, ensuring the quality of fresh milk during cold chain transportation has become a core challenge for cold chain logistics management, given its high perishability and sensitivity. Fresh milk is highly susceptible to environmental factors such as temperature fluctuations, transport vibrations, humidity changes, and storage time, leading to nutrient loss, microbial growth, and even spoilage, severely impacting product quality and consumer safety. Traditional cold chain logistics distribution models rely heavily on static temperature control and fixed transportation routes, lacking real-time perception and dynamic response capabilities to the quality evolution of fresh milk during transportation, making it difficult to cope with complex and ever-changing transportation environments and unexpected delays.

[0003] Although multi-source sensor technology has been gradually applied to cold chain monitoring in recent years, enabling the collection of environmental parameters such as temperature, humidity, and vibration, existing systems generally remain at the level of "data recording + post-event traceability," lacking the ability to fuse and analyze multi-dimensional data and intelligently predict quality evolution trends. Furthermore, delivery decisions are often independent of the quality monitoring system, failing to link real-time quality status with route optimization and resource scheduling. This results in the inability to adjust delivery strategies in a timely manner when facing quality risks, leading to resource waste or quality control failures.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide an intelligent cold chain logistics distribution method and system based on real-time monitoring of fresh milk quality. This aims to solve the technical problems of existing cold chain logistics technologies lacking the ability to accurately perceive and dynamically respond to the real-time evolution of fresh milk quality, making it difficult to achieve coordinated optimization of quality monitoring and distribution decisions, resulting in delayed quality risk warnings and passive control measures.

[0006] To achieve the above objectives, the present invention provides an intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality, the method comprising: Acquire historical cold chain data from the cold chain transportation system, identify typical cold chain scenarios, construct cold chain scenario models, extract environmental characteristic parameters of typical cold chain scenarios, and generate corresponding quality impact labels. The quality impact label and the acquired real-time monitoring data of fresh milk are used as inputs, and the data are processed based on the spatiotemporal fusion analysis mechanism to output the quality status trajectory of fresh milk in different cold chain links and quality anomaly data. Set up a digital twin platform to map the quality status trajectory and quality anomaly data into the digital twin platform, and construct a delivery route map structure and a multi-dimensional quality correlation map. Based on the analysis of the constraint boundary between delivery timeliness and quality assurance using a multidimensional quality correlation graph, and combined with cold chain resource scheduling data and delivery priority parameters, a dynamic delivery adjustment window set is output to generate an intelligent delivery optimization strategy.

[0007] Optionally, the process of acquiring historical cold chain data of the cold chain transportation system, identifying typical cold chain scenarios, and constructing a cold chain scenario model includes: By acquiring historical cold chain data generated during the historical delivery process of the cold chain transportation system, the historical cold chain data is constructed into a cold chain data matrix according to spatial location and time sequence; The cold chain data matrix is ​​divided into blocks based on the spatiotemporal grid partitioning mechanism to obtain multiple historical cold chain units, and the local environmental fragments corresponding to each historical cold chain unit are extracted. Multidimensional environmental features are extracted for each local environmental segment. An environmental feature vector is formed by combining the multidimensional environmental features extracted from the same local environmental segment. An environmental feature vector set is constructed by combining the environmental feature vectors of all local environmental segments. Cluster analysis is performed on the set of environmental feature vectors to identify representative typical cold chain scenarios. Based on the multi-dimensional environmental features extracted from each typical cold chain scenario, a cold chain scenario model is constructed.

[0008] Optionally, the process of extracting environmental feature parameters of typical cold chain scenarios and generating corresponding quality impact labels includes: For various typical cold chain scenarios included in the cold chain scenario model, environmental feature parameters are extracted as quality impact descriptors based on statistically representative multidimensional environmental features in the environmental feature vector set. Each quality impact description item is input into a preset quality rule set. The quality rule set is constructed based on historical quality monitoring data of fresh milk and quality grade samples through multi-factor variance analysis or Bayesian network inference, and outputs the range of quality impact intensity coefficients and its quality risk level. By identifying the quality impact intensity coefficient ranges for various typical cold chain scenarios, quality risk levels are linked to corresponding typical cold chain scenarios, and quality impact labels are generated.

[0009] Optionally, the process of taking the quality impact label and the acquired real-time monitoring data of fresh milk as input, processing them based on a spatiotemporal fusion analysis mechanism, and outputting the quality status trajectory of fresh milk in different cold chain links and quality anomaly data is as follows: The real-time monitoring data of fresh milk acquired by multiple sensors during cold chain transportation is continuously collected and spatiotemporally registered. The real-time monitoring data includes fresh milk temperature data, humidity data, vibration data, gas composition data, and location trajectory records. Based on the spatiotemporal fusion analysis mechanism, real-time monitoring data is processed for spatiotemporal correlation to form quality monitoring segments with spatiotemporal coupling characteristics; each quality monitoring segment is associated and bound with the corresponding quality impact label to construct a quality evolution sample set. Based on the quality evolution sample set and each typical cold chain link, multiple quality monitoring segments are fused and trend analyzed in spatiotemporal order. The dynamic change characteristics of quality parameters in each cold chain link are extracted, quality feature nodes are constructed, and the links are correlated in spatiotemporal order to form a quality change trajectory. The quality change trajectory is then analyzed based on deep learning combined with multidimensional environmental features.

[0010] Optionally, the process of analyzing the quality change trajectory based on deep learning and multi-dimensional environmental features includes: Identify the spatiotemporal intervals of quality anomalies, extract the change gradient, fluctuation amplitude, and critical threshold points of environmental characteristic parameters for the spatiotemporal intervals of quality anomalies, and extract the remaining quality maintenance time in the cold chain corresponding to the spatiotemporal intervals of quality anomalies, which is recorded as the quality safety period. The change gradient, fluctuation amplitude, critical threshold points, and quality safety period are recorded as quality anomaly data, and the quality change trajectory corresponding to the spatiotemporal intervals of quality anomalies is recorded as the quality status trajectory.

[0011] Optionally, the process of setting up a digital twin platform, mapping quality status trajectories and quality anomaly data to the digital twin platform, and constructing a delivery route map structure and a multi-dimensional quality correlation map includes: The quality status trajectory and quality anomaly data are correlated to each virtual cold chain node in the digital twin platform. Based on the spatiotemporal sequence of each cold chain link, a multi-node status mapping chain for the fresh milk delivery process is established, and the multi-node status mapping chain is used as the basic spatiotemporal axis. Based on location trajectory records, the corresponding quality feature nodes in each cold chain link are aggregated to construct delivery nodes that correspond one-to-one with the cold chain links, and quality anomaly data is bound to the corresponding delivery nodes. The quality anomaly data and the corresponding change parameters in each delivery node in the quality status trajectory are recorded as the quality attribute markers of the delivery node; the quality safety period in the quality anomaly data is used as the quality label in the current cold chain link and marked to the corresponding delivery node. Based on the quality and safety period, the cold chain process is analyzed to identify the upper and lower thresholds of the timeliness constraints of the delivery nodes, and to form the delivery timeliness boundary. Based on the basic spatiotemporal axis, the delivery nodes are arranged in spatiotemporal order, and path edges are constructed between the delivery nodes. The delivery nodes are connected through the path edges to construct a delivery path graph. Quality attribute markers, quality labels, and delivery timeliness boundaries are bound to the corresponding delivery nodes and edges, and mapped to the virtual cold chain model in the digital twin platform to construct a multi-dimensional quality association graph.

[0012] Optionally, the process of analyzing the constraint boundary between delivery timeliness and quality assurance based on multi-dimensional quality correlation graphs, combining cold chain resource scheduling data and delivery priority parameters, outputting a dynamic delivery adjustment window set, and generating an intelligent delivery optimization strategy includes: Obtain the set of delivery nodes, the set of quality tags, and the delivery time boundary in the multidimensional quality correlation graph; identify the delivery node group that has a risk of conflict between the delivery time boundary and the quality status trajectory; and determine the corresponding quality risk period. By combining delivery priority parameters, delivery nodes are sorted, and delivery nodes with lower priority that are in the quality risk period are marked as adjustable delivery groups. Optimizable time segments in their delivery time boundary are extracted to construct an adjustable time window set. Based on the cold chain resource scheduling data, identify the time segments that meet the quality assurance conditions in the adjustable time window set, output the dynamic delivery adjustment window set as candidates, and send the adjustable delivery group, the optimizable time segment, and the dynamic delivery adjustment window set to the delivery scheduling center. Combine the adjustable delivery group, the optimizable time segment, and the dynamic delivery adjustment window set to generate the corresponding intelligent delivery optimization plan.

[0013] Furthermore, to achieve the above objectives, the present invention also provides an intelligent cold chain logistics and distribution system based on real-time monitoring of fresh milk quality, the system comprising: The scenario modeling module is used to acquire historical cold chain data of the cold chain transportation system, identify typical cold chain scenarios, build cold chain scenario models, extract environmental feature parameters of typical cold chain scenarios, and generate corresponding quality impact labels. The quality tracking module takes quality impact labels and real-time monitoring data of fresh milk as input, processes them based on a spatiotemporal fusion analysis mechanism, and outputs the quality status trajectory of fresh milk in different cold chain links and quality anomaly data. The digital mapping module is used to set up the digital twin platform, map the quality status trajectory and quality anomaly data to the digital twin platform, and construct the delivery route map structure and multi-dimensional quality correlation map. The intelligent scheduling module is used to analyze the constraint boundary between delivery timeliness and quality assurance based on multi-dimensional quality correlation graph analysis. Combining cold chain resource scheduling data and delivery priority parameters, it outputs a set of dynamic delivery adjustment windows and generates intelligent delivery optimization strategies.

[0014] Furthermore, to achieve the above objectives, the present invention also provides an intelligent cold chain logistics distribution device based on real-time monitoring of fresh milk quality. The device includes: a memory, a processor, and an intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality stored in the memory and executable on the processor. The intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality is configured to implement the steps of the intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality as described above.

[0015] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing an intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality. When the intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality is executed by a processor, it implements the steps of the intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality as described above.

[0016] This invention provides an intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality. The method integrates historical cold chain scenario modeling with real-time multi-source sensor data of fresh milk to construct a spatiotemporal dynamic perception system oriented towards quality evolution, enabling continuous tracking and anomaly warning of the quality status of fresh milk during cold chain transportation. By leveraging a digital twin platform to map the quality trajectory during physical transportation to the delivery path, a multi-dimensional quality correlation map including quality attributes, timeliness boundaries, and environmental associations is constructed, enhancing the visualization and analyzability of the cold chain system status. Furthermore, by combining delivery priority and resource scheduling data, a dynamically adjustable window is identified and optimization strategies are generated, realizing a shift from passive response to proactive control. This effectively improves the intelligent decision-making capability of cold chain logistics, optimizing delivery efficiency and resource utilization while ensuring the safety and quality of fresh milk, demonstrating significant practical value and promising prospects for wider application. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an embodiment of the intelligent cold chain logistics and distribution method based on real-time monitoring of fresh milk quality according to the present invention. Figure 2 This is a structural block diagram of an embodiment of the intelligent cold chain logistics and distribution system based on real-time monitoring of fresh milk quality according to the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] Reference Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the intelligent cold chain logistics and distribution method based on real-time monitoring of fresh milk quality according to the present invention.

[0021] In one embodiment, the intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality includes: Step S100: Obtain historical cold chain data of the cold chain transportation system, identify typical cold chain scenarios, construct a cold chain scenario model, extract environmental feature parameters of typical cold chain scenarios, and generate corresponding quality impact labels.

[0022] Historical cold chain data can be a collection of environmental parameters, transportation routes, timestamps, and corresponding product quality inspection results recorded during past cold chain transportation processes. This data can provide the initial data foundation for building a cold chain scenario model. In this embodiment, historical cold chain data can be extracted from enterprise historical transportation logs, temperature control recorders, quality inspection reports, and other systems. Typical cold chain scenarios can be transportation situation categories that repeatedly appear in historical cold chain data and possess stable environmental characteristics and quality impact patterns. These can be used as basic units for building a cold chain scenario model. For example, typical cold chain scenarios may include, but are not limited to, one or more of the following: high-temperature and high-humidity urban delivery scenarios, long-distance constant-temperature trunk line transportation scenarios, and community delivery scenarios with frequent starts and stops.

[0023] A cold chain scenario model can be a structured knowledge model built based on historical cold chain data, representing the relationship between typical transportation environments and changes in fresh milk quality. It can provide prior knowledge support for real-time quality assessment and assist in determining the potential impact of the current transportation environment on fresh milk quality. In this embodiment, the cold chain scenario model can extract representative transportation scenarios and their corresponding environmental characteristics and quality results by clustering or pattern recognition of historical cold chain data, forming a reusable scenario template. Furthermore, the cold chain scenario model can collaborate with quality impact labels to map environmental characteristic parameters to quality change results and provide input for a spatiotemporal fusion analysis mechanism. Environmental characteristic parameters can be key variables describing the impact of the cold chain transportation environment on fresh milk quality, such as temperature fluctuation range, vibration frequency, and humidity level. They can be used to quantify the environmental characteristics of typical cold chain scenarios and establish a correlation with quality impact. In an exemplary embodiment, environmental characteristic parameters can be obtained through statistical analysis of historical sensor data or extraction from cluster centers.

[0024] Quality impact labels can be classification or rating labels that identify the actual impact of specific combinations of environmental characteristic parameters on the quality of fresh milk. They can be used to establish a mapping relationship between environment and quality for real-time analysis. In one specific embodiment, quality impact labels can be generated based on historical quality inspection data or expert rules. Furthermore, quality impact labels can be generated by inferring the quality deterioration level based on a microbial growth model and then labeling it, or by quality inspectors manually evaluating and labeling based on sensory and physicochemical indicators, thereby establishing an explainable causal relationship between environment and quality.

[0025] Obtaining historical cold chain data for the cold chain transportation system can be achieved by extracting structured datasets containing environmental parameters, route information, and quality inspection results from enterprise historical databases or third-party logistics platforms. Furthermore, this operation can be implemented by connecting to the enterprise's TMS / WMS system or calling historical API interfaces, thus providing a raw data foundation for building cold chain scenario models. Identifying typical cold chain scenarios can be done by clustering or pattern mining historical cold chain data to extract repetitive and representative transportation scenario categories. Further, this operation can be achieved by using the K-means clustering algorithm to perform unsupervised grouping of environmental feature parameters, or by identifying recurring route-environment combination patterns based on time series similarity measures (such as DTW), thereby enabling structured summarization of complex transportation environments.

[0026] Constructing a cold chain scenario model can involve structurally associating typical cold chain scenarios with their corresponding environmental characteristic parameters and quality impact labels, forming a queryable knowledge model. Furthermore, this operation can be achieved by building a scenario-label mapping table based on a rule engine, or by training a lightweight classification model as a predictor of scenario-to-quality impact, thereby establishing an environment-quality mapping relationship library that supports real-time inference. Extracting environmental characteristic parameters from typical cold chain scenarios can be achieved by statistically analyzing historical sensor data for each typical cold chain scenario, extracting representative features such as mean, variance, and extreme values. Further, this operation can be achieved using sliding window statistics or principal component analysis methods, thereby quantifying the environmental characteristics of the scenario for subsequent matching and comparison. Generating corresponding quality impact labels can be achieved by labeling each typical cold chain scenario with its actual impact category or level on fresh milk quality based on historical quality inspection results or expert experience. Further, this operation can be achieved by inferring and labeling the quality deterioration level based on a microbial growth model, or by having quality inspectors manually assess and label based on sensory and physicochemical indicators, thereby establishing an interpretable environment-quality causal relationship.

[0027] Step S200: The quality impact label and the acquired real-time monitoring data of fresh milk are used as inputs, and processed based on the spatiotemporal fusion analysis mechanism to output the quality status trajectory of fresh milk in different cold chain links and quality anomaly data.

[0028] The real-time monitoring data for fresh milk can be dynamic environmental and status data such as temperature, humidity, vibration, and location collected in real time by multi-source sensors during transportation. This data can reflect the actual environmental conditions during the current transportation process and be used for real-time quality assessment. In one specific embodiment, the real-time monitoring data is continuously collected by an onboard IoT device and uploaded via a wireless network. The spatiotemporal fusion analysis mechanism can be an analytical framework that integrates time and spatial dimensions to jointly model and infer multi-source heterogeneous sensor data. This framework can be used to achieve continuous tracking and anomaly warning of fresh milk quality status, overcoming the limitations of static recording. In this embodiment, the spatiotemporal fusion analysis mechanism can integrate historical labels and real-time monitoring data through time series modeling and spatial location association algorithms (such as neural networks and spatiotemporal convolution) to deduce the quality evolution path. Furthermore, the spatiotemporal fusion analysis mechanism can receive quality impact labels and real-time monitoring data for fresh milk as input and output quality status trajectory and quality anomaly data.

[0029] The quality status trajectory can be a continuous time series representation of the quality evolution of fresh milk throughout the entire transportation process. It can be used to track quality change trends and support anomaly warning and decision-making intervention. In an exemplary embodiment, the quality status trajectory is generated by a spatiotemporal fusion analysis mechanism based on historical labels and real-time data. Quality anomaly data can be monitoring points or time periods that deviate from the normal quality evolution path, indicating potential spoilage risks. It can be used to trigger an early warning mechanism, prompting the need for intervention measures. In a specific embodiment, quality anomaly data is identified from the quality status trajectory by setting a threshold or using an anomaly detection algorithm. The quality impact labels and the acquired real-time monitoring data of fresh milk are used as input. This can involve sending the real-time sensor data of the current transportation task along with the quality impact labels of the most similar historical scenarios into the analysis module. Furthermore, this operation can be achieved through nearest neighbor matching or embedded spatial similarity calculation, thereby enabling the fusion of historical priors and the current state as input.

[0030] Processing based on a spatiotemporal fusion analysis mechanism can involve jointly analyzing input data using spatiotemporal modeling methods to infer the current quality status and identify anomalies. Furthermore, this operation can be achieved by using a spatiotemporal graph convolutional network (ST-GCN) to model multi-dimensional sensor data at path nodes, or by using an LSTM network combined with GPS location embedding to perform time-series prediction of quality evolution, thereby enabling dynamic tracking of quality status and risk warning. Outputting the quality status trajectory and quality anomaly data of fresh milk in different cold chain stages can be achieved by outputting the analysis results in time series format and marking anomalies that exceed the normal range. Furthermore, this operation can be output to downstream systems through standardized data interfaces (such as JSON or Protobuf), thus providing a traceable and predictable dynamic view of quality.

[0031] Step S300: Set up a digital twin platform, map the quality status trajectory and quality anomaly data to the digital twin platform, and construct the delivery route map structure and multi-dimensional quality correlation map.

[0032] The digital twin platform can be a virtual simulation system that digitally mirrors the physical cold chain transportation process, supporting data mapping, state synchronization, and interactive analysis. It can be used to achieve a virtual-real mapping between the physical transportation process and the virtual model, supporting visual monitoring and strategy deduction. In this embodiment, the digital twin platform can access monitoring data and path information from the physical world in real time through API interfaces or message queues, constructing a dynamically updated transportation entity model in virtual space. Furthermore, the digital twin platform can receive quality status trajectories and quality anomaly data to construct a delivery path graph structure and a multi-dimensional quality correlation graph. The delivery path graph structure can be a delivery network represented in graph theory, with nodes representing logistics nodes (such as warehouses and stores) and edges representing feasible transportation paths. It can be used to structurally represent the physical delivery network in the digital twin platform, supporting path optimization calculations. In one specific embodiment, the delivery path graph structure is constructed based on GIS data and actual delivery routes. The multi-dimensional quality correlation graph can be a knowledge graph representing the relationship between fresh milk quality attributes, timeliness boundaries, and environmental factors in graph structure form. It can be used to enhance the interpretability and analyzability of the cold chain system state, supporting constraint boundary identification and strategy generation. In this embodiment, the multidimensional quality correlation graph can abstract elements such as quality status trajectory, delivery path, and environmental parameters into nodes and edges in a digital twin platform, establishing semantic relationships. Furthermore, the multidimensional quality correlation graph can serve as a basis for analyzing the boundaries of delivery timeliness and quality assurance constraints, and can participate in the generation of optimization strategies together with cold chain resource scheduling data and delivery priority parameters.

[0033] Setting up a digital twin platform can involve deploying a virtual simulation environment and configuring data interfaces to access the real-time status of the physical world. Furthermore, this operation can be achieved by building a 3D visualization digital twin platform based on Unity or WebGL, or by using a lightweight graph database (such as Neo4j) to build a logical-level digital twin platform, thereby creating a digital mirror of the physical transportation process. Mapping quality status trajectories and quality anomaly data to the digital twin platform can be done by binding analysis results to the corresponding virtual entities of the transportation task through a data synchronization mechanism. Further, this operation can be achieved through an event-driven architecture or a timed synchronization mechanism, thereby enabling the visualization of quality status in virtual space. Constructing a delivery route graph structure and a multi-dimensional quality association graph can be done within the digital twin platform by abstracting elements such as logistics nodes, routes, quality status, and environmental parameters into graph nodes and edges, establishing semantic relationships. Furthermore, this operation can be implemented by constructing a graph with the delivery route as the skeleton, quality status as node attributes, and environmental disturbances as edge weights, or by using ontology modeling methods to define three types of entities—quality, timeliness, and environment—and their relationships, thereby generating an RDF graph, thus forming a structured knowledge representation that supports multidimensional analysis.

[0034] Step S400: Based on the analysis of the multi-dimensional quality correlation graph, the constraint boundary between delivery timeliness and quality assurance is analyzed. Combined with cold chain resource scheduling data and delivery priority parameters, a dynamic delivery adjustment window set is output to generate an intelligent delivery optimization strategy.

[0035] The constraint boundary between delivery timeliness and quality assurance can be the maximum allowable delivery time or route delay limit under the premise of ensuring the quality and safety of fresh milk. This can be used to define the feasible scope of delivery adjustments and prevent sacrificing quality for efficiency. In one specific embodiment, this constraint boundary is derived from the quality decay model and timeliness sensitivity analysis in a multi-dimensional quality correlation graph. Cold chain resource scheduling data can be the status and capacity information of currently available logistics resources such as refrigerated vehicles, drivers, and cold storage spaces, which can be used as resource constraints for generating optimization strategies. In an exemplary embodiment, cold chain resource scheduling data is obtained in real time from a TMS (Transportation Management System) or WMS (Warehouse Management System).

[0036] Delivery priority parameters can be weighted indicators reflecting the delivery timeliness and service level requirements of different orders or customers, and can be used to guide the priority ranking of optimization strategies when resources are limited. In one specific embodiment, the delivery priority parameters are set by the business system based on customer type, order urgency, etc. The dynamic delivery adjustment window set can be a set of time and route feasible intervals for adjusting the delivery plan under the premise of meeting quality assurance and timeliness requirements. It can be used to provide decision space for intelligent delivery optimization strategies, supporting proactive control rather than passive response. In this embodiment, the dynamic delivery adjustment window set is calculated based on the constraint boundaries in the multi-dimensional quality correlation graph, combined with the current resource availability and task priority, to generate multiple executable adjustment options. Furthermore, the dynamic delivery adjustment window set is jointly generated by the multi-dimensional quality correlation graph, cold chain resource scheduling data, and delivery priority parameters, serving as the direct input for generating intelligent delivery optimization strategies. The intelligent delivery optimization strategy can be a dynamically generated optimal or suboptimal delivery execution plan under the conditions of meeting quality constraints and resource limitations, and can be used to achieve synergistic optimization of delivery efficiency and quality assurance. In one specific embodiment, the intelligent delivery optimization strategy is generated based on a dynamic delivery adjustment window set, combined with optimization algorithms (such as reinforcement learning and integer programming).

[0037] Analyzing the constraint boundary between delivery timeliness and quality assurance based on a multidimensional quality correlation graph can be achieved by traversing the quality decay paths in the graph and, combined with a fresh milk shelf-life model, calculating the maximum allowable delay time for each path segment. Furthermore, this operation can be implemented by fitting an Arrhenius model based on the quality decay rate in the graph to derive the timeliness upper limit, or by generating multiple quality evolution paths on the graph using Monte Carlo simulation and statistically analyzing the safe time window, thereby quantifying the feasibility boundary of delivery adjustments. Combining cold chain resource scheduling data and delivery priority parameters, the current available resource status and order priority can be introduced as constraints into the decision model. Further, this operation can be achieved by encoding resource capacity and priority weights as constraints in the optimization objective function, ensuring the optimization strategy is realistically executable.

[0038] Outputting a dynamic delivery adjustment window set can enumerate all feasible adjustment options (such as rerouting, delaying, and switching vehicles) while satisfying constraints and resource limitations. Furthermore, this operation can be achieved by searching for alternative sub-paths that meet quality constraints on the path graph based on a sliding time window, or by using a reinforcement learning agent to explore the feasible scheduling action space under resource reallocation, thus providing multiple optional intervention opportunities and methods. Generating an intelligent delivery optimization strategy can be achieved by selecting the optimal solution from the dynamic delivery adjustment window set to form an executable scheduling instruction. Furthermore, this operation can be achieved by using a multi-objective optimization algorithm (such as NSGA-II) to solve the trade-off between efficiency and quality, or by using a rule engine to automatically select adjustment actions based on priority and window urgency, thereby realizing closed-loop control from perception to decision-making.

[0039] Taking a sudden traffic jam during the morning rush hour for fresh milk delivery in a city as an example, the intelligent cold chain logistics delivery method based on real-time monitoring of fresh milk quality in this embodiment can be as follows: A refrigerated truck encounters severe traffic congestion during the morning rush hour, and the onboard sensors continuously upload temperature and vibration data. The system integrates the real-time data with the quality impact labels in the historical "urban short-distance high-frequency delivery scenario model," and through a spatiotemporal fusion analysis mechanism, it discovers that the current vibration intensity and temperature fluctuations have caused the quality status trajectory to deviate from the normal range, triggering a quality anomaly warning. This trajectory and abnormal data are mapped to a digital twin platform, highlighting risk sections on the delivery route map structure and updating the multi-dimensional quality correlation map. The system then analyzes the delivery timeliness and quality assurance constraints of the order, and, combined with the current idle status of other vehicles (cold chain resource scheduling data) and the high priority parameters of the hospital customer, identifies a dynamic adjustment window of "being able to call upon nearby vehicles for pick-up within 30 minutes." Finally, it generates an intelligent delivery optimization strategy of "the original vehicle suspends its advance, and a backup vehicle is dispatched to pick up the goods from the nearest node for direct delivery," preventing the fresh milk from spoiling due to prolonged storage.

[0040] In one embodiment, the process of acquiring historical cold chain data of the cold chain transportation system, identifying typical cold chain scenarios, and constructing a cold chain scenario model includes: By acquiring historical cold chain data generated during the historical delivery process of the cold chain transportation system, the historical cold chain data is constructed into a cold chain data matrix according to spatial location and time sequence; The cold chain data matrix can be a two-dimensional or three-dimensional structured data table that organizes historical cold chain data according to spatial location and temporal order. Rows represent time steps, and columns represent spatial nodes or sensor dimensions. It can provide structured input for subsequent spatiotemporal grid partitioning and support the precise extraction of local environmental segments. In this embodiment, the cold chain data matrix can be constructed by spatiotemporally aligning and resampling the original historical cold chain data to unify the temporal granularity and spatial coordinate system. Constructing the cold chain data matrix according to spatial location and temporal order can be achieved by spatiotemporally aligning the original historical cold chain data, using unified timestamps and geographic coordinates as indexes, and organizing it into a matrix form. In an exemplary embodiment, this operation can be achieved by constructing a sparse matrix with GPS trajectory points as columns and second-level timestamps as rows, and then filling in missing values ​​through interpolation; or by dividing the delivery route into fixed segments, each segment corresponding to a column, and filling in the environmental parameter mean at the minute granularity. This enables the structured integration of heterogeneous sensor data, laying the foundation for refined spatiotemporal slicing.

[0041] The cold chain data matrix is ​​divided into blocks based on the spatiotemporal grid partitioning mechanism to obtain multiple historical cold chain units, and the local environmental fragments corresponding to each historical cold chain unit are extracted. The spatiotemporal grid partitioning mechanism can be a method of processing cold chain data matrices by regularly or adaptively dividing them into blocks according to preset time windows and spatial regions. This can be used to achieve fine-grained slicing of complex transportation processes, ensuring that each historical cold chain unit has spatiotemporal consistency. For example, the spatiotemporal grid partitioning mechanism can employ sliding windows, quadtree partitioning, or slicing strategies based on transportation topology to divide a continuous spatiotemporal domain into discrete units. In a specific embodiment, the spatiotemporal grid partitioning mechanism may include, but is not limited to, an equidistant spatiotemporal grid partitioning mechanism, an event-triggered adaptive grid partitioning mechanism, and a semantic grid partitioning mechanism combined with road network structure. Furthermore, the spatiotemporal grid partitioning mechanism is applied to the cold chain data matrix to output historical cold chain units. A historical cold chain unit can be a subset of data with clear spatiotemporal boundaries obtained after spatiotemporal grid partitioning, representing a local transportation process. It can be used as the basic unit for extracting local environmental fragments, ensuring the contextual integrity of environmental features. In this embodiment, the historical cold chain units are segmented from the cold chain data matrix by the spatiotemporal grid partitioning mechanism.

[0042] Local environmental fragments can be raw sensor data sequences extracted from a single historical cold chain unit, reflecting the environmental state within a specific spatiotemporal interval. These fragments can serve as a direct source for multidimensional environmental feature extraction, preserving original dynamic characteristics. In an exemplary embodiment, local environmental fragments are obtained by extracting and standardizing channel data such as temperature, humidity, and vibration from the historical cold chain unit. The cold chain data matrix is ​​segmented based on a spatiotemporal grid partitioning mechanism, which can be done by regular or adaptive segmentation according to a preset time window length and spatial region range. Furthermore, this operation can be achieved by using a fixed 5-minute × 1-kilometer grid for regular partitioning, or by dynamically adjusting the grid boundaries based on vehicle start / stop events or road segment types, thereby generating historical cold chain units with spatiotemporal consistency and avoiding cross-scene mixing. Obtaining multiple historical cold chain units can be achieved by outputting all sub-matrices or data blocks after spatiotemporal grid partitioning, thus forming independently processable local transportation process samples. Extracting local environmental fragments corresponding to each historical cold chain unit can be done by extracting the time series of raw sensor data from each historical cold chain unit, thereby preserving environmental dynamic details and providing raw materials for feature engineering.

[0043] Multidimensional environmental features are extracted for each local environmental segment. An environmental feature vector is formed by combining the multidimensional environmental features extracted from the same local environmental segment. An environmental feature vector set is constructed by combining the environmental feature vectors of all local environmental segments. The multidimensional environmental features can be statistical quantities or time-series indicators calculated from local environmental segments that characterize the properties of environmental disturbances. These are used to quantify the potential impact of the environment on fresh milk quality, such as temperature fluctuation variance, peak vibration frequency, and cumulative humidity exposure. The environmental feature vector can be a numerical vector composed of multidimensional environmental features extracted from the same local environmental segment in a fixed order. It can be used to transform heterogeneous sensor data into a unified, computable, structured representation, supporting clustering and similarity measurements. In this embodiment, the environmental feature vector is obtained by concatenating multidimensional environmental features into a fixed-length vector according to predefined dimensions. The environmental feature vector set can be a collection of environmental feature vectors corresponding to all historical cold chain units. It can be used as the input dataset for cluster analysis, supporting the automatic identification of typical cold chain scenarios. For example, the environmental feature vector set is obtained by summarizing all environmental feature vectors to form a matrix or list structure.

[0044] Extracting multidimensional environmental features for each local environmental segment can involve calculating statistics, frequency domain indices, or composite indices for channels such as temperature, humidity, and vibration. Further, this operation can be achieved by calculating the moving standard deviation of the temperature sequence to characterize fluctuation intensity, or by performing wavelet transform on the vibration signal to extract energy distribution features, thus transforming the raw data into a physically meaningful representation of environmental disturbances. Combining the multidimensional environmental features extracted from the same local environmental segment into an environmental feature vector can be achieved by concatenating all features from the same segment into a fixed-dimensional vector in a predefined order, enabling unified encoding of multi-source heterogeneous features and supporting machine learning processing. Constructing an environmental feature vector set from the environmental feature vectors of all local environmental segments can be achieved by aggregating all environmental feature vectors into a single dataset, thus forming a complete sample space suitable for cluster analysis.

[0045] Cluster analysis is performed on the set of environmental feature vectors to identify representative typical cold chain scenarios. Based on the multi-dimensional environmental features extracted from each typical cold chain scenario, a cold chain scenario model is constructed. Cluster analysis of the environmental feature vector set can be performed using unsupervised learning algorithms to group the environmental feature vectors and identify high-frequency or high-impact patterns. In an exemplary embodiment, this operation can be achieved by using the DBSCAN algorithm to identify density clusters, adapting to non-spherical distribution scenarios, or by using a Gaussian mixture model (GMM) for soft clustering, supporting scenario overlap modeling, thereby automatically discovering representative transportation environment patterns. Identifying representative typical cold chain scenarios can be done by defining clusters with a large number of samples or significant quality impacts in the clustering results as typical cold chain scenarios, thus extracting environmental pattern categories that are practically significant to fresh milk quality. Summarizing the multi-dimensional environmental features extracted from each typical cold chain scenario can be done by calculating the centroid or statistical distribution of the environmental feature vectors within each typical scenario. This operation can be achieved by calculating the mean of the feature vectors within a cluster as a scenario prototype, or by retaining the upper and lower quartiles of the feature distribution within a cluster to characterize the range of variation, thereby forming a standardized environmental feature description of the scenario. Constructing a cold chain scenario model can involve structurally storing typical cold chain scenarios and their summarized multidimensional environmental features to form a queryable knowledge model. This allows for the establishment of a prior knowledge base with physical meaning and statistical representativeness, supporting real-time matching and reasoning.

[0046] Taking the modeling of bumpy and humid mountain road sections as an example, the intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality in this embodiment can be as follows: The system extracts cold chain records of a certain mountain delivery route from historical data, constructs a cold chain data matrix, and identifies multiple road segment units containing continuous curves and high humidity through a spatiotemporal grid partitioning mechanism; for each unit, it extracts multi-dimensional environmental features such as vibration RMS value, average humidity, and duration to form an environmental feature vector; cluster analysis reveals that one type of vector appears frequently and corresponds to a high protein denaturation rate in historical quality inspections, which is identified as a typical cold chain scenario of bumpy and humid mountain roads; the features of this scenario are summarized and stored in the cold chain scenario model. In subsequent real-time transportation, when a vehicle enters a similar road section and the real-time feature vector is highly similar to that scenario, the system issues an early warning of quality risks and triggers an alternative route evaluation.

[0047] In one embodiment, the process of extracting environmental feature parameters of typical cold chain scenarios and generating corresponding quality impact labels includes: For various typical cold chain scenarios included in the cold chain scenario model, environmental feature parameters are extracted as quality impact descriptors based on statistically representative multidimensional environmental features in the environmental feature vector set. Each quality impact description item is input into a preset quality rule set. The quality rule set is constructed based on historical quality monitoring data of fresh milk and quality grade samples through multi-factor variance analysis or Bayesian network inference, and outputs the range of quality impact intensity coefficients and its quality risk level. By identifying the quality impact intensity coefficient ranges for various typical cold chain scenarios, quality risk levels are linked to corresponding typical cold chain scenarios, and quality impact labels are generated.

[0048] The quality impact description item can be a set of structured environmental feature parameters extracted from typical cold chain scenarios to characterize their potential impact on fresh milk quality. This set can be used to transform raw environmental data into descriptive indicators with quality semantics. In this embodiment, the quality impact description item can be screened and semantically named based on statistically representative multidimensional environmental features (such as temperature fluctuation variance, cumulative vibration energy, etc.) in the environmental feature vector set to form interpretable input variables. The preset quality rule set can be an inference model or rule base constructed based on historical fresh milk quality monitoring data and quality grade samples, used to map environmental features to quality impact results. This can be used to achieve semantic conversion from environmental features to quality risks, supporting automated label generation. For example, the preset quality rule set can identify the significant impact of environmental variables on quality indicators based on multi-factor variance analysis, or model the probabilistic dependency between environment and quality through Bayesian networks to form an executable inference mechanism. In an exemplary embodiment, the preset quality rule set can be one or more of the following: a rule set based on statistical significance testing, an inference rule set based on Bayesian causal graphs, or a fusion rule set combining expert rules and data-driven approaches. Furthermore, the preset quality rule set can accept quality impact descriptions as input and output the range of quality impact intensity coefficients and quality risk levels.

[0049] Historical quality monitoring data for fresh milk can be measured data on physicochemical indicators, microbiological indicators, and sensory evaluations collected during past transportation or storage. This data can be used as training or validation data for constructing a pre-defined quality rule set. In one specific embodiment, historical quality monitoring data can originate from laboratory test reports, online quality sensors, or quality inspection record systems. Quality grade samples can be a set of historical cases with labeled quality risk levels, including corresponding environmental conditions and final quality judgment results. This data can be used for supervised learning or rule calibration to ensure the operational consistency of the quality rule set. In this embodiment, quality grade samples can be generated by experts through retrospective labeling based on quality inspection results and consumer feedback. Multifactor ANOVA can be a statistical method used to assess the significant impact of multiple environmental factors and their interactions on fresh milk quality indicators. It can be used to identify key environmental variables and support the construction of main effects and interaction terms in the rule set. For example, multifactor ANOVA can group historical quality monitoring data for fresh milk according to different environmental factors and calculate the F-statistic to determine significance.

[0050] For various typical cold chain scenarios included in the cold chain scenario model, environmental feature parameters are extracted as quality impact descriptors based on statistically representative multidimensional environmental features from the environmental feature vector set. This can be achieved by selecting dimensions with high discriminative power or high relevance from the environmental feature vectors corresponding to each typical cold chain scenario, assigning semantic names, and standardizing them. Furthermore, this operation can be further refined by using feature importance ranking (such as SHAP values) to select the most quality-sensitive environmental features as descriptors, or by manually defining key parameters (such as "duration of high temperature") based on domain knowledge and synthesizing them from the original features. This transforms abstract numerical values ​​into interpretable quality impact factors, supporting subsequent rule-based reasoning. Each type of quality impact descriptor is input into a pre-defined quality rule set, which can be done by passing structured quality impact descriptors as variables into the pre-defined quality rule set for reasoning. This operation can further trigger a semantic mapping process from environment to quality impact.

[0051] The quality rule set is constructed based on historical quality monitoring data and quality grade samples of fresh milk through multi-factor ANOVA or Bayesian network inference. It can utilize statistical or probabilistic methods to learn the environment-quality correlation patterns from historical data, forming a reusable inference model. Furthermore, this operation can use multi-factor ANOVA to determine the significance of the main effects and second-order interaction terms of temperature, humidity, and vibration, constructing linear rules; or it can use Bayesian networks to learn the conditional probability distribution of each environmental variable on "protein denaturation rate," supporting uncertainty inference. This allows for the establishment of a quality impact assessment mechanism with interpretability and generalization capabilities.

[0052] The output of the quality impact intensity coefficient range and its quality risk level can be achieved by generating continuous impact intensity ranges and discrete risk classifications based on the reasoning results of the rule set. This operation can provide both quantitative assessment and qualitative judgment, meeting the needs of different decision-making levels. By binding the quality risk level with the corresponding typical cold chain scenario through the quality impact intensity coefficient ranges conforming to various typical cold chain scenarios, the reasoned quality risk level can be appended as an attribute to the metadata of the corresponding typical cold chain scenario. Furthermore, this operation can achieve a structured association between scenario and risk semantics. The generation of quality impact tags can be achieved by encoding the bound risk level into a standardized tag (such as "high risk - heat exposure") and storing it in the cold chain scenario model. Furthermore, this operation can form prior knowledge units that can be called by the real-time system, supporting rapid matching and early warning.

[0053] The quality impact intensity coefficient range can be a numerical range that quantifies the impact of a specific environmental combination on the degree of fresh milk quality deterioration, reflecting the strength of the impact. It can be used to provide a continuous measure of impact intensity, supporting fine-grained risk assessment and threshold setting. In this embodiment, the quality impact intensity coefficient range can be calculated by a preset quality rule set based on the input quality impact description, usually expressed as a confidence interval or membership interval. The quality risk level can be the result of discretizing the severity of quality problems that may occur in fresh milk under a specific cold chain scenario, and can be used to provide an intuitive and operable risk label for scheduling decisions. In a specific embodiment, the quality risk level can be divided into several levels (such as low, medium, and high) based on the quality impact intensity coefficient range, combined with business tolerance or safety standards. In an exemplary embodiment, the quality risk level can be bound to a typical cold chain scenario to generate a quality impact label.

[0054] Taking the generation of labels for high-temperature exposure scenarios in summer urban delivery as an example, the intelligent cold chain logistics delivery method based on real-time monitoring of fresh milk quality in this embodiment can be as follows: The system extracts "average daytime temperature > 25℃ and refrigerated truck door opening frequency > 8 times / hour" as a quality impact description item from the typical cold chain scenario of "short-distance high-frequency urban delivery"; this description item is input into a preset quality rule set, which, based on the total number of colonies and sensory scores in historical data, uses Bayesian network inference to derive the "microbial growth acceleration coefficient range of [1.8, 2.4]" corresponding to this combination, and determines it as "high spoilage risk level"; this risk level is then bound to the typical scenario to generate a "high risk - heat exposure" quality impact label. In subsequent real-time delivery, if the vehicle enters a similar high-temperature, high-door-opening-frequency road section, the system immediately matches this label, triggering an optimization strategy of early unloading or activating a backup refrigerated box.

[0055] In one embodiment, the process of taking quality impact labels and acquired real-time monitoring data of fresh milk as input, processing them based on a spatiotemporal fusion analysis mechanism, and outputting the quality status trajectory of fresh milk in different cold chain stages and quality anomaly data is as follows: The real-time monitoring data of fresh milk acquired by multiple sensors during cold chain transportation is continuously collected and spatiotemporally registered. The real-time monitoring data includes fresh milk temperature data, humidity data, vibration data, gas composition data, and location trajectory records. The gas composition data can be monitoring data reflecting the concentration of volatile gases inside the fresh milk packaging or in the vehicle environment. This data can serve as a sensitive indicator of early microbial metabolism or packaging damage, supplementing the limitations of traditional temperature, humidity, and vibration parameters. In one exemplary embodiment, the gas composition data can be acquired in real time using electrochemical or optical gas sensors. Spatiotemporal registration processing can be a preprocessing operation that aligns the time and spatial coordinates of multi-source asynchronously acquired sensor data. This can address the issues of asynchronous and heterogeneous multi-source sensor data, ensuring the accuracy of subsequent fusion analysis. Furthermore, spatiotemporal registration processing can employ interpolation, resampling, or event synchronization mechanisms to align data from different sampling frequencies, such as temperature, vibration, and gas, to a unified spatiotemporal grid. Continuous acquisition and spatiotemporal registration processing of real-time monitoring data of fresh milk obtained from multiple sensors during cold chain transportation can involve continuously receiving raw data streams from temperature, humidity, vibration, gas, and positioning sensors, and synchronizing the data through timestamp alignment and spatial coordinate mapping. Furthermore, this operation can be achieved by linearly interpolating and resampling all sensor data at a reference frequency of 1Hz, or by using vehicle CAN bus events (such as door opening / closing and start / stop) as synchronization anchors for semantic alignment, thereby eliminating multi-source asynchronous errors and forming high-quality input data that is consistent in time and space.

[0056] Based on the spatiotemporal fusion analysis mechanism, real-time monitoring data is processed for spatiotemporal correlation to form quality monitoring segments with spatiotemporal coupling characteristics; each quality monitoring segment is associated and bound with the corresponding quality impact label to construct a quality evolution sample set. The quality monitoring segment can be a multi-dimensional environmental data unit with a unified timestamp and spatial location semantics, formed after spatiotemporal correlation processing. It represents a local quality impact context in a continuous transportation process and can be used as a basic unit for quality state modeling. It retains the spatiotemporal coupling characteristics of environmental disturbances and supports accurate matching with quality impact labels. In a specific embodiment, the quality monitoring segment can be spatiotemporally registered with real-time fresh milk monitoring data, then segmented according to a preset time window or event boundary (such as route switching), and fused with multi-source sensor values ​​such as temperature, humidity, vibration, and gas composition. Furthermore, the quality monitoring segment can be bound to a quality evolution sample set to form a quality evolution sample set; multiple segments are sequentially fused to generate quality feature nodes. The quality evolution sample set can be a structured dataset formed by binding multiple quality monitoring segments to their corresponding quality impact labels. It is used to characterize the dynamic response patterns of quality under different transportation scenarios and can provide training and inference basis for trend analysis and deep learning models, realizing the mapping from environmental input to quality state. In one exemplary embodiment, the quality evolution sample set can match each quality monitoring segment to the most similar typical cold chain scenario through spatial location and temporal context, inherit its quality impact label, and form a labeled sample.

[0057] Based on a spatiotemporal fusion analysis mechanism, real-time monitoring data is spatiotemporally correlated to form quality monitoring segments with spatiotemporal coupling characteristics. This can be achieved by dividing the registered data into preset spatiotemporal windows (e.g., every 5 minutes or per road segment) and fusing multi-dimensional sensor values ​​into unified segments. Furthermore, this operation can utilize sliding time windows to generate overlapping segments to capture gradual changes, or automatically divide segment boundaries based on GIS road network nodes to match typical cold chain links, thereby generating analysis units with contextual integrity and supporting semantic matching with historical scenarios. Each quality monitoring segment is associated with a corresponding quality impact label, which can be achieved by similarity matching between the segment's spatial location, environmental feature vectors, and historical cold chain scenario models, inheriting the quality impact label from the matched scenario. This operation imbues real-time data with quality semantics, enabling a leap from environmental perception to risk awareness. A quality evolution sample set is constructed, which can be achieved by summarizing all labeled quality monitoring segments to form a structured dataset. This operation provides a foundation of labeled data for subsequent trend analysis and deep learning.

[0058] Based on the quality evolution sample set and each typical cold chain link, multiple quality monitoring segments are fused and trend analyzed in spatiotemporal order. The dynamic change characteristics of quality parameters in each cold chain link are extracted, quality feature nodes are constructed, and the links are correlated in spatiotemporal order to form a quality change trajectory. The quality change trajectory is then analyzed based on deep learning combined with multidimensional environmental features.

[0059] Typical cold chain links can be standardized stages with clearly defined functional boundaries and environmental characteristics during cold chain transportation, such as loading, trunk transportation, transshipment, and last-mile delivery. These serve as logical units for organizing quality monitoring segments and constructing quality feature nodes. The dynamic change characteristics of quality parameters can be statistical or model-driven indicators extracted from time-series monitoring data within a single cold chain link, reflecting the trend of quality evolution. These can be used to construct quality feature nodes, characterizing the actual intensity and direction of the link's impact on fresh milk quality. For example, the dynamic change characteristics of quality parameters may include cumulative temperature deviation, vibration spectrum centroid drift, and gas concentration change rate. Quality feature nodes can be core feature vectors extracted from multiple quality monitoring segments within a specific cold chain link through fusion and trend analysis, characterizing the dynamic changes in quality at that link. These can be used as constituent units of the quality change trajectory, compressing redundant data and retaining key evolutionary information. In a specific embodiment, quality feature nodes can perform time-series aggregation (e.g., moving average, trend slope, fluctuation entropy, etc.) on all quality monitoring segments within the same cold chain link (e.g., urban distribution segment) to generate comprehensive features. Furthermore, quality characteristic nodes can be connected in the spatiotemporal sequence of the cold chain to form a quality change trajectory. For example, quality characteristic nodes may include temperature accumulation offset characteristic nodes, vibration energy attenuation characteristic nodes, gas evaporation acceleration characteristic nodes, etc.

[0060] The quality change trajectory can be a structured path reflecting the continuous evolution of fresh milk quality, formed by connecting multiple quality feature nodes in the spatiotemporal sequence of cold chain transportation. This path can support global quality status tracking, anomaly location, and future trend prediction, providing a basis for early warning and scheduling. In an exemplary embodiment, the quality change trajectory can be linked sequentially with quality feature nodes at each stage based on location trajectory records and logistics node information, forming an end-to-end quality evolution view. Based on the quality evolution sample set and each typical cold chain stage, multiple quality monitoring segments are fused and trend analyzed in spatiotemporal order. This can involve temporal aggregation and pattern extraction of all segments belonging to the same cold chain stage. Furthermore, this operation can use moving averages and first-order differences to extract trend and fluctuation features, or apply an LSTM autoencoder to learn the latent representation of the segment sequence, thereby refining stage-level quality impact features and reducing data noise.

[0061] Extracting the dynamic change characteristics of quality parameters within each cold chain link can involve calculating indicators such as cumulative temperature deviation, vibration energy integral, and gas concentration slope. This operation quantifies the actual disturbance intensity of each link on quality. Constructing quality feature nodes can be achieved by encapsulating dynamic change characteristics into fixed-dimensional vectors, serving as a quality state representation for that link. This operation forms the basic unit for trajectory construction. Based on the spatiotemporal sequence of cold chain links, a quality change trajectory is formed, which can be achieved by sequentially connecting the quality feature nodes of each link according to the temporal order and spatial topology of the transportation path. This operation can generate an end-to-end quality evolution path, supporting global monitoring. Analyzing the quality change trajectory based on deep learning combined with multi-dimensional environmental features can be achieved by inputting the quality change trajectory into a time-series model (such as Transformer or TCN) and combining it with multi-dimensional environmental features for anomaly detection and trend prediction. Furthermore, this operation can use Graph Attention Networks (GAT) to model inter-link dependencies and identify abnormal propagation paths, or use Variational Autoencoders (VAE) to learn the normal trajectory distribution and detect anomalies by reconstructing errors, thereby enabling early warning of quality deterioration and future state prediction.

[0062] For example, in a scenario where gas anomalies trigger early warnings during community delivery, the intelligent cold chain logistics delivery method based on real-time monitoring of fresh milk quality in this embodiment can involve refrigerated trucks continuously uploading temperature, vibration, and CO2 concentration data within the packaging via multi-source sensors during community delivery. After spatiotemporal registration, the system divides the data into quality monitoring segments every 2 minutes. A segment showing a rapid increase in CO2 concentration accompanied by slight temperature fluctuations is matched and tagged with a "high-risk - active microorganisms" quality impact label. This segment is added to the quality evolution sample set. Within a typical cold chain link of "community delivery," the system performs trend analysis on nearly 10 segments, extracting "gas concentration change rate" and "temperature fluctuation entropy" as dynamic change features to construct quality feature nodes. When the quality change trajectory formed by this node and its preceding nodes triggers a high anomaly score in the deep learning model, the system immediately issues an early warning that "fresh milk may have begun to ferment" and pushes the alert to the digital twin platform, triggering the dispatch center to assess whether to terminate delivery early or replace the product.

[0063] In one embodiment, the process of analyzing the trajectory of quality change based on deep learning and multi-dimensional environmental features includes: Identify the spatiotemporal intervals of quality anomalies, extract the change gradient, fluctuation amplitude, and critical threshold points of environmental characteristic parameters for the spatiotemporal intervals of quality anomalies, and extract the remaining quality maintenance time in the cold chain corresponding to the spatiotemporal intervals of quality anomalies, which is recorded as the quality safety period. The change gradient, fluctuation amplitude, critical threshold points, and quality safety period are recorded as quality anomaly data, and the quality change trajectory corresponding to the spatiotemporal intervals of quality anomalies is recorded as the quality status trajectory.

[0064] The quality anomaly spatiotemporal interval can be a continuous temporal and spatial range with a significant quality deterioration trend identified in the quality change trajectory. Corresponding to specific transportation segments and time periods, it can be used to pinpoint the specific spatiotemporal location of quality risks, providing targets for refined intervention. In this embodiment, the quality anomaly spatiotemporal interval can be scanned using a deep learning model (such as a temporal anomaly detection network) to identify local areas deviating from the normal evolutionary pattern. The gradient of environmental characteristic parameters can be the rate or derivative characteristics of changes in environmental parameters such as temperature, humidity, or vibration over time within the quality anomaly spatiotemporal interval. This can be used to reflect the dynamic acceleration of quality deterioration and assess the urgency of the risk. In an exemplary embodiment, the gradient of environmental characteristic parameters can be obtained by numerical differentiation or sliding slope calculation of registered real-time monitoring data within the anomaly interval. The fluctuation amplitude can be the maximum deviation range of environmental characteristic parameters from their local mean within the quality anomaly spatiotemporal interval. This can be used to characterize the instability intensity of the transportation environment and assist in determining the type of disturbance (e.g., frequent starts and stops vs. sustained high temperatures). For example, the fluctuation range can be calculated by measuring the standard deviation, peak-to-peak value, or interquartile range of environmental parameters within the abnormal interval.

[0065] The critical threshold point can be the time-space coordinate point at which environmental or quality parameters first exceed a preset safety boundary within the spatiotemporal interval of quality anomalies. It can be used to mark the starting moment of quality assurance failure, serving as a key basis for triggering early warnings. In a specific embodiment, the critical threshold point can be determined by comparing real-time parameters with a dynamic threshold set based on a fresh milk preservation model. The remaining quality maintenance time can be the estimated remaining duration for fresh milk to maintain an acceptable quality level, starting from the current quality state without changing existing transportation conditions. It can be used to quantify the time margin under the current state, providing time constraints for scheduling decisions. Furthermore, the remaining quality maintenance time can be obtained by calculating the time required to reach an unacceptable quality level based on current quality characteristic nodes and historical quality decay models (such as the Arrhenius equation or a data-driven prediction model).

[0066] The quality safety period can be a standardized naming convention for the remaining quality maintenance time, emphasizing its safety window attribute in delivery decisions. It can be used as a core input for calculating dynamic delivery adjustment windows, directly affecting whether delays or rerouting are permissible. In this embodiment, the quality safety period, along with the change gradient, fluctuation amplitude, and critical threshold point, constitutes quality anomaly data. Quality anomaly data can be a structured set of anomaly descriptions composed of the change gradient, fluctuation amplitude, critical threshold point, and quality safety period. This can be used to transform fuzzy anomalies into computable, comparable, and decision-making multidimensional indicators, supporting intelligent early warning and strategy generation. The quality state trajectory can specifically refer to the complete quality change trajectory containing the spatiotemporal interval of quality anomalies. As the backbone representation of the quality evolution of the current transportation task, it can be used as a core input for digital twin platform mapping and multidimensional quality correlation graph construction. In a specific embodiment, there is a synergistic relationship between the quality state trajectory and the quality anomaly data generated after the extraction of its anomaly sub-intervals.

[0067] Identifying spatiotemporal intervals of quality anomalies can be achieved by using deep learning models (such as LSTM-AE and Transformer anomaly detectors) to evaluate quality change trajectories point-by-point or through a sliding window, marking continuous spatiotemporal regions that significantly deviate from the normal pattern. Furthermore, identifying these intervals can be achieved by using continuous segments with reconstruction errors exceeding a dynamic threshold as anomaly intervals, or by using attention mechanisms to focus on the periods with the greatest impact on quality and aggregating them into intervals, thus enabling precise spatiotemporal localization of quality degradation events. Extracting the gradient, fluctuation amplitude, and critical threshold points of environmental characteristic parameters for these quality anomaly intervals can be done by calculating the derivative, statistical dispersion, and intersection points with safety thresholds for the original environmental parameter sequences within the identified anomaly intervals. Further, this operation can be achieved by smoothing with a Savitzky-Golay filter and then differentiating to obtain robust gradient changes, or by dynamically setting critical threshold points through quantile regression to adapt to different seasonal baselines, thereby quantifying the dynamic characteristics and severity of the anomalies.

[0068] Extracting the remaining quality maintenance time within the cold chain corresponding to the spatiotemporal interval of quality anomalies can be achieved by calling a fresh milk quality degradation prediction model based on the current quality feature node status to estimate the time required to reach an unacceptable quality level. Furthermore, this operation can be implemented through online prediction using a historical degradation model based on Bayesian updates, or by analogy estimation using the remaining time distribution of similar historical scenarios, thus providing crucial time margin information to support the judgment of whether delivery can be delayed. Recording the change gradient, fluctuation amplitude, critical threshold point, and quality safety period as quality anomaly data can be done by structurally encapsulating these four types of indicators into JSON or vector form and attaching them to the corresponding transportation task metadata, thereby forming a standardized anomaly description for easy subsequent system calls and analysis. Recording the quality change trajectory corresponding to the spatiotemporal interval of quality anomalies as a quality state trajectory can be done by marking the entire quality change trajectory (including normal and abnormal segments) as the main quality state representation of the current task and associating it with the transportation order, thus providing a complete and highly semantic input source for digital twin mapping.

[0069] Taking the gradual deterioration caused by refrigeration equipment malfunction during trunk transportation as an example, the intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality in this embodiment can be as follows: During cross-provincial trunk transportation, the refrigeration system efficiency of a cold chain transport vehicle decreases, and the temperature rises slowly. A deep learning model identifies the "quality anomaly spatiotemporal interval" from the 3rd to the 5th hour in the quality change trajectory. The system extracts the temperature change gradient within this interval as +0.8℃ / h, with a fluctuation range of ±2.1℃, and reaches the critical threshold point (7℃) at the 4.2th hour. At the same time, based on the current protein stability model, the remaining quality maintenance time is calculated to be 1.5 hours, i.e., the "quality safety period". These four indicators constitute "quality anomaly data", and the entire trajectory is marked as "quality status trajectory" and pushed to the digital twin platform. Based on this, the dispatch center judges that if no intervention is made, the product will deteriorate after 6.7 hours; and since the nearest backup cold storage is within a 1-hour drive, an optimized strategy of "unloading and temporary storage nearby" is immediately generated to avoid the entire order being scrapped.

[0070] In one embodiment, the process of setting up a digital twin platform and mapping quality status trajectories and quality anomaly data to the digital twin platform to construct a delivery route map structure and a multi-dimensional quality correlation map includes: The quality status trajectory and quality anomaly data are correlated to each virtual cold chain node in the digital twin platform. Based on the spatiotemporal sequence of each cold chain link, a multi-node status mapping chain for the fresh milk delivery process is established, and the multi-node status mapping chain is used as the basic spatiotemporal axis. The multi-node state mapping chain can be a chain structure formed in a digital twin platform, mapping quality state trajectories to virtual nodes one-to-one according to the spatiotemporal sequence of cold chain links. It can serve as the skeleton for constructing the basic spatiotemporal axis, supporting the organization of subsequent delivery nodes and path edges. In this embodiment, the multi-node state mapping chain can map each cold chain link (such as loading, trunk line, and delivery) in the physical transportation process to virtual nodes and link them sequentially to form the main trunk of state evolution. Furthermore, the multi-node state mapping chain can form a delivery path diagram through path edge connections; binding quality attribute markers, quality labels, and delivery timeliness boundaries. The basic spatiotemporal axis can be a logical coordinate system abstracted from the multi-node state mapping chain, representing the spatiotemporal evolution of the entire fresh milk delivery process. It can be used to provide temporal and topological benchmarks for delivery node aggregation, path edge construction, and attribute binding. In an exemplary embodiment, the basic spatiotemporal axis can establish a unified spatiotemporal reference framework based on the timestamps and spatial locations of the multi-node state mapping chain.

[0071] Associating quality status trajectories and quality anomaly data with virtual cold chain nodes in the digital twin platform can be achieved by binding trajectories and anomaly data to pre-defined virtual nodes in the digital twin platform based on spatiotemporal matching principles. This operation can achieve node binding through dual matching of timestamps and GPS coordinates, or semantic alignment using road segment IDs and event logs, thus achieving a precise mapping from physical states to virtual entities. Based on the spatiotemporal sequence of each cold chain link, a multi-node state mapping chain for the fresh milk delivery process can be established, which can be done by sequentially linking each virtual cold chain node according to the transportation process, forming a chain structure. Furthermore, this operation can be achieved by automatically constructing the mapping chain based on logistics WMS / TMS event flows or by initializing the mapping chain through the link sequence output by the path planning engine, thus constructing the logical backbone of state evolution. Using the multi-node state mapping chain as a basic spatiotemporal axis, it can be abstracted into a unified spatiotemporal reference system for subsequent node aggregation and edge construction, thereby providing a globally consistent spatiotemporal benchmark.

[0072] Based on location trajectory records, the corresponding quality feature nodes in each cold chain link are aggregated to construct delivery nodes that correspond one-to-one with the cold chain links, and quality anomaly data is bound to the corresponding delivery nodes. In this context, delivery nodes can be virtual nodes in a digital twin platform, corresponding one-to-one with physical cold chain links. They carry multi-dimensional attributes such as quality and timeliness, and can serve as the basic unit of a multi-dimensional quality association graph, integrating environmental, quality, and timeliness information. In one specific embodiment, delivery nodes can aggregate multiple quality feature nodes within the same cold chain link into a single logical node based on location trajectory records. Furthermore, delivery nodes can form a delivery path graph by connecting path edges; binding quality attribute markers, quality tags, and delivery timeliness boundaries. Aggregating corresponding quality feature nodes within each cold chain link based on location trajectory records to construct delivery nodes corresponding one-to-one with the cold chain link can be achieved through statistical aggregation (such as mean, maximum value) or feature fusion of multiple quality feature nodes within the same physical link. Further, this operation can be achieved by using time-weighted average aggregation of quality features or using cluster centers to represent the overall state of the link, thereby compressing data granularity and improving graph readability and computational efficiency. Binding quality anomaly data to the corresponding delivery node can be done by attaching the quality anomaly data generated in the preceding steps as metadata to the matching delivery node, thereby assigning risk semantics to the node.

[0073] The quality anomaly data and the corresponding change parameters in each delivery node in the quality status trajectory are recorded as the quality attribute markers of the delivery node; the quality safety period in the quality anomaly data is used as the quality label in the current cold chain link and marked to the corresponding delivery node. The quality attribute marker can be a set of structured parameters reflecting the impact of the current environment on the quality of fresh milk at the delivery node. Originating from the changing parameters in the quality anomaly data, it can provide an interpretable description of the intensity of quality disturbances at the node level, supporting risk visualization and comparison. In an exemplary embodiment, the quality attribute marker can extract and standardize changing parameters such as temperature gradient and vibration amplitude that match the node in the quality anomaly data. The quality label can be a time-sensitive quality status identifier marked on the delivery node, centered on the quality safety period. It can be used to intuitively express the remaining acceptable delivery time of the current node, guiding the judgment of scheduling urgency. In a specific embodiment, the quality label can directly assign the quality safety period from the quality anomaly data as a dynamic label for the node. Changing parameters that match the quality anomaly data and the quality status trajectory within each delivery node are recorded as the quality attribute markers of the delivery node. These can include parameters such as temperature gradient and vibration amplitude that are consistent with the spatiotemporal range of the node, forming structured attributes, thereby enabling the quantification of environmental disturbances at the node level. Using the quality safety period from the abnormal quality data as a quality label within the current cold chain process and marking it to the corresponding delivery node can be achieved by directly writing the quality safety period value into the node's quality label field, thus providing an intuitive timeliness risk warning.

[0074] Based on the quality and safety period, the cold chain process is analyzed to identify the upper and lower thresholds of the timeliness constraints of the delivery nodes, and to form the delivery timeliness boundary. The timeliness constraint upper threshold can be the maximum allowable delivery completion time for the current cold chain stage, derived from the quality and safety period. It can be used to define the upper time limit for scheduling adjustments, preventing quality control issues caused by timeouts. In one embodiment, the timeliness constraint upper threshold can be calculated by adding the quality and safety period to the current time, representing the latest acceptable delivery time. The timeliness constraint lower threshold can be the earliest time limit that the current cold chain stage must complete, set based on business requirements or customer commitments. It can be used to ensure that high-priority orders are not delivered prematurely due to over-optimization, resulting in resource idleness. In an exemplary embodiment, the timeliness constraint lower threshold can be determined by delivery priority parameters or SLA agreements, typically earlier than the upper threshold. The delivery timeliness boundary can be a feasible time window interval for the current delivery node, composed of the upper and lower timeliness constraint thresholds. It can be used to define the operable time flexibility range for scheduling and is the basis for generating dynamic adjustment windows. Furthermore, the delivery timeliness boundary can be bound to the delivery node and participate in the generation of the dynamic delivery adjustment window set. Based on the quality and safety period, the cold chain process is analyzed to identify the upper and lower thresholds of timeliness constraints for delivery nodes, thus forming the delivery timeliness boundary. This can be achieved by calculating the upper threshold as the current time plus the quality and safety period, and combining this with the business SLA to obtain the lower threshold, forming a time window. Furthermore, this operation can be implemented by using the formula: upper threshold = current time + quality and safety period × 0.9 (with buffer allowance) or lower threshold = customer appointment time - fixed lead time, thereby clearly defining the feasible scheduling domain.

[0075] Based on the basic spatiotemporal axis, the delivery nodes are arranged in spatiotemporal order, and path edges are constructed between the delivery nodes. The delivery nodes are connected through the path edges to construct a delivery path graph. Quality attribute markers, quality labels, and delivery timeliness boundaries are bound to the corresponding delivery nodes and edges, and mapped to the virtual cold chain model in the digital twin platform to construct a multi-dimensional quality association graph.

[0076] In this system, path edges can be directed edges connecting adjacent delivery nodes, representing the logical mapping of physical transportation paths. They can be used to construct the topological structure of the delivery path graph, supporting path-level analysis and optimization. In one specific embodiment, path edges can be automatically generated based on the sequential order of nodes in the basic spatiotemporal axis, and attributes such as distance and estimated travel time can be added. The delivery path graph can be a directed graph composed of delivery nodes and path edges, fully expressing the spatiotemporal topological structure of fresh milk delivery. It can be used as the skeleton of a multidimensional quality association graph, supporting path-level joint optimization of quality and resources. In one exemplary embodiment, the delivery path graph can be formed by sequentially connecting all delivery nodes on the basic spatiotemporal axis. Furthermore, the delivery path graph can carry quality attribute markers, quality labels, and delivery timeliness boundaries, constituting a multidimensional quality association graph. The virtual cold chain model can be a digital mirror of the physical cold chain transportation system (vehicles, warehouses, routes, etc.) in a digital twin platform, providing an operating environment and visualization carrier for the multidimensional quality association graph. In one specific embodiment, the virtual cold chain model can be constructed based on GIS, vehicle IoT data, and logistics network topology.

[0077] Based on a fundamental spatiotemporal axis, delivery nodes are arranged in spatiotemporal order to construct path edges between them. This can be achieved by traversing the node sequence in the fundamental spatiotemporal axis and creating directed edges sequentially, thus establishing path topology relationships. Delivery nodes are then connected via these path edges to construct a delivery path graph. This can be achieved by integrating all nodes and edges into a graph data structure (such as an adjacency list or graph database records), thereby forming a computable delivery network model. Quality attribute markers, quality labels, and delivery timeliness boundaries are bound to the corresponding delivery nodes and edges. This can be achieved by writing these three types of attributes as node attributes into the graph structure, with some timeliness boundaries extended to edges (such as road segment travel time constraints). Furthermore, this operation can be further enhanced by using an attribute graph model (such as Neo4j) to store multidimensional attributes of nodes and edges or by serializing the graph using JSON-LD format to support semantic interoperability, thereby constructing a multidimensional semantic graph. Mapping to the virtual cold chain model in the digital twin platform, a multi-dimensional quality correlation map can be constructed. This can be achieved by overlaying the delivery route map onto the 3D or logical view of the digital twin platform, enabling visualization and interaction, thereby forming a unified decision-making view that integrates physical, quality, and timeliness.

[0078] Taking multi-node risk visualization in inter-provincial trunk line transportation as an example, the intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality in this embodiment can be as follows: A fresh milk order is shipped from an East China factory to a South China store. The system constructs a path map in the digital twin platform that includes four distribution nodes: "loading - highway trunk line - transit warehouse - city distribution". Among them, the "high-speed trunk line" node is marked with a quality attribute of "temperature gradient +0.7℃ / h, fluctuation ±1.8℃" due to refrigeration fluctuations, and the quality label is "1.2-hour safety period". Based on this, the upper threshold of the timeliness constraint is calculated as 14:30, and the lower threshold is 12:00, forming a delivery timeliness boundary of [12:00, 14:30]. This node is highlighted in orange in the digital twin platform, and the dispatcher can see intuitively that if unloading is not completed before 14:30, the fresh milk will spoil; and the current time is 11:00, with a 3.5-hour operation window. The system then assesses whether other high-priority orders can be inserted or whether the original plan can be maintained. The entire multidimensional quality correlation map makes the previously hidden quality-time coupling relationship explicit, supporting refined decision-making.

[0079] In one embodiment, the process of analyzing the constraint boundary between delivery timeliness and quality assurance based on a multi-dimensional quality correlation graph, combining cold chain resource scheduling data and delivery priority parameters, outputting a dynamic delivery adjustment window set, and generating an intelligent delivery optimization strategy includes: Obtain the set of delivery nodes, the set of quality tags, and the delivery time boundary in the multidimensional quality correlation graph; identify the delivery node group that has a risk of conflict between the delivery time boundary and the quality status trajectory; and determine the corresponding quality risk period. The conflict-risk delivery node group can be a set of delivery nodes identified in the multi-dimensional quality correlation map whose planned delivery time falls within the quality risk period and who have conflicts between quality assurance and timeliness. This set can be used to locate high-risk delivery tasks requiring intervention, serving as the starting point for adjustability analysis. In this embodiment, the conflict-risk delivery node group can be filtered by comparing the planned delivery time of each delivery node with the quality risk period derived from its quality safety period (e.g., the current time + the time period after the remaining safety period) to identify nodes with overlapping times. For example, the conflict-risk delivery node group can include, but is not limited to, one or more of the following: single-point critical conflict node group, continuous road segment cumulative conflict node group, and multi-order concurrent conflict node group. The quality risk period can be a time interval derived from the quality safety period, where failure to complete delivery within this period will result in unacceptable fresh milk quality. This can be used to define the time boundary of quality assurance failure for conflict detection. In an exemplary embodiment, the quality risk period can be defined by extending from the current time to the current time + the quality safety period; the interval after this extension constitutes the quality risk period. Obtaining the set of delivery nodes, the set of quality tags, and the delivery time boundary from a multidimensional quality association graph can be achieved by extracting structured attribute data from the constructed multidimensional quality association graph. Furthermore, this operation can be implemented by batch retrieving node attributes through graph database queries or by subscribing to graph change events to obtain the latest status in real time, thus providing a complete input for conflict identification.

[0080] Identifying delivery node groups where there is a risk of conflict between delivery timeliness boundaries and quality status trajectories can be achieved by comparing the planned times of each node with the quality risk periods and marking overlaps. Furthermore, this operation can be implemented by using a time interval intersection algorithm to detect conflicts or by predicting future conflict probabilities based on Monte Carlo simulation and setting thresholds, thereby achieving the technical effect of proactively discovering potential quality control failure points. Determining the corresponding quality risk period can be achieved by calculating the risk initiation time of each conflict node based on the quality safety period, thus quantifying the time window for risk occurrence.

[0081] By combining delivery priority parameters, delivery nodes are sorted, and delivery nodes with lower priority that are in the quality risk period are marked as adjustable delivery groups. Optimizable time segments in their delivery time boundary are extracted to construct an adjustable time window set. The adjustable delivery group can be a subset of low-priority delivery nodes that have been prioritized and marked as suitable for scheduling adjustments within a group of delivery nodes with conflict risk. This subset can be used to limit the scope of optimization intervention and protect the experience of high-priority customers. In one specific embodiment, the adjustable delivery group can be formed by combining delivery priority parameters, sorting conflicting nodes in ascending order of priority, and selecting nodes below a threshold or with relatively low priority. Furthermore, the adjustable delivery group can include, but is not limited to, one or more of the following: an adjustable group for ordinary retail customers, an adjustable group for non-urgent replenishment, and an adjustable group for next-day delivery flexible orders. For example, the delivery time boundary of the adjustable delivery group is used to extract optimizable time segments.

[0082] Optimizable time segments can be time intervals extracted from the delivery timeliness boundaries corresponding to adjustable delivery groups, which can be used to reschedule deliveries. These segments can form a preliminary adjustment space for resource matching verification. In this embodiment, optimizable time segments can be idle periods within the delivery timeliness boundaries or windows that do not conflict with adjacent tasks. The adjustable time window set can be a preliminary candidate time window set composed of optimizable time segments from all adjustable delivery groups, which can be used to provide input for resource constraint filtering. In an exemplary embodiment, the adjustable time window set can be formed by aggregating the optimizable time segments of each adjustable delivery group to create an initial window pool that has not undergone resource verification. Exemplarily, the adjustable time window set can include, but is not limited to, one or more of the following: a single-node independent window set, a multi-node collaborative window set, and a cross-regional aggregated window set.

[0083] By combining delivery priority parameters, delivery nodes are sorted, either in ascending or descending order of priority value. This operation can be achieved using weighted priority (combining customer type and time sensitivity) or hierarchical sorting: first grouping and then sorting within groups, thus supporting differentiated intervention strategies. Delivery nodes with lower priority and high quality risk periods are prioritized as adjustable delivery groups. This can be achieved by selecting nodes with lower priority and lower priority that are in high-risk periods and adding them to the adjustable groups, thereby releasing scheduling flexibility while ensuring key customers are served. Optimizable time segments are extracted from the delivery time boundaries, identifying unoccupied or negotiable time intervals within the time boundaries of the adjustable delivery groups. Further, this operation can be achieved by using a sliding window to scan continuous idle periods within the boundaries or predicting compressible / extended intervals based on historical scheduling patterns, thus generating preliminary adjustment options. A set of adjustable time windows is constructed, which can be achieved by aggregating all optimizable time segments into a unified set, thus forming an adjustment space pool to be validated.

[0084] Based on the cold chain resource scheduling data, identify the time segments that meet the quality assurance conditions in the adjustable time window set, output the dynamic delivery adjustment window set as candidates, and send the adjustable delivery group, the optimizable time segment, and the dynamic delivery adjustment window set to the delivery scheduling center. Combine the adjustable delivery group, the optimizable time segment, and the dynamic delivery adjustment window set to generate the corresponding intelligent delivery optimization plan.

[0085] The time segments that meet the quality assurance conditions can be those within the adjustable time window set, verified by cold chain resource scheduling data, that can be delivered without compromising the quality of fresh milk. This ensures the physical feasibility and quality safety of the adjustment plan. In this embodiment, the time segments meeting the quality assurance conditions can be determined by combining resource status such as refrigerated truck availability, cold storage pre-cooling capacity, and driver scheduling to determine whether the required temperature control and operating conditions can be maintained during that period. The delivery scheduling center can be a logistics decision-making hub system that receives optimization suggestions and generates final execution instructions, serving as the receiving end and execution initiation point for intelligent delivery optimization plans. In one embodiment, the delivery scheduling center can receive a dynamic delivery adjustment window set and related parameters via API or message queue. Identifying time segments that meet the quality assurance conditions within the adjustable time window set based on cold chain resource scheduling data can involve querying the availability and capacity matching degree of resources such as refrigerated trucks, cold storage, and personnel during candidate time periods. Furthermore, this operation can be achieved by calling the resource scheduling engine to solve the constraint satisfaction problem (CSP) or by checking resource-time matching conditions based on a rule engine, thereby filtering infeasible windows and ensuring the feasibility of the plan.

[0086] The output, serving as a set of candidate dynamic delivery adjustment windows, can be achieved by encapsulating resource-verified time segments into structured output, thus providing high-confidence adjustment options. Sending the adjustable delivery group, optimizable time segments, and the set of dynamic delivery adjustment windows to the delivery scheduling center can be done via an internal message bus or API push of these three types of data to the scheduling system, triggering the optimization scheme generation process. Combining the adjustable delivery group, optimizable time segments, and the set of dynamic delivery adjustment windows, a corresponding intelligent delivery optimization scheme is generated. This can be achieved within the scheduling center by replanning routes, allocating resources, or adjusting timing based on the aforementioned inputs. Furthermore, this operation can be implemented by employing a reinforcement learning agent to select the optimal action sequence from the window set or by using a mixed-integer programming model to solve for the minimum-cost adjustment scheme, thereby achieving a closed-loop control from perception to execution.

[0087] Taking the coordination of multiple order conflicts during peak urban delivery periods as an example, the intelligent cold chain logistics delivery method based on real-time monitoring of fresh milk quality in this embodiment can be as follows: At 10:00 AM on a certain day, the system identifies 5 delivery nodes with conflict risks from a multi-dimensional quality correlation graph, of which 3 are community convenience stores (low priority) and 2 are hospitals (high priority). The quality risk period is after 11:30 AM. The system prioritizes marking the 3 convenience store nodes as 'adjustable delivery groups', with their delivery time boundaries being [10:30, 12:00]. Optimizable time segments such as [10:30–11:00] and [11:45–12:00] are extracted from these. Combined with cold chain resource scheduling data, it is found that a spare vehicle can depart from a nearby cold storage during the 11:45–12:00 time period, and the temperature inside the vehicle is stable. Therefore, '11:45–12:00' is identified as a 'time segment that meets the quality assurance conditions' and included in the dynamic delivery adjustment window set. Based on this, the dispatch center generated an optimization plan: the original vehicle continued to serve the hospital orders, while the backup vehicle took over the delivery of 3 convenience stores at 11:45, ensuring that all orders were completed within the quality and safety period, while avoiding delays for the main vehicle.

[0088] In addition, refer to Figure 2 To achieve the above objectives, the present invention also provides an intelligent cold chain logistics and distribution system based on real-time monitoring of fresh milk quality, the system comprising: The scenario modeling module 10 is used to acquire historical cold chain data of the cold chain transportation system, identify typical cold chain scenarios, construct cold chain scenario models, extract environmental feature parameters of typical cold chain scenarios, and generate corresponding quality impact labels. The quality tracking module 20 is used to take the quality impact label and the acquired real-time monitoring data of fresh milk as input, process them based on the spatiotemporal fusion analysis mechanism, and output the quality status trajectory of fresh milk in different cold chain links and quality anomaly data. The digital mapping module 30 is used to set up the digital twin platform, map the quality status trajectory and quality anomaly data to the digital twin platform, and construct the delivery route map structure and multi-dimensional quality correlation map. The intelligent scheduling module 40 is used to analyze the constraint boundary between delivery timeliness and quality assurance based on the multi-dimensional quality correlation graph, and output a dynamic delivery adjustment window set by combining cold chain resource scheduling data and delivery priority parameters to generate intelligent delivery optimization strategies.

[0089] Other embodiments or specific implementations of the intelligent cold chain logistics and distribution system based on real-time monitoring of fresh milk quality described in this invention can be referred to the above-mentioned method embodiments, and will not be repeated here.

[0090] Furthermore, to achieve the above objectives, the present invention also provides an intelligent cold chain logistics distribution device based on real-time monitoring of fresh milk quality. The device includes: a memory, a processor, and an intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality stored in the memory and executable on the processor. The intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality is configured to implement the steps of the intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality as described above.

[0091] In addition, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing an intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality. When the intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality is executed by a processor, it implements the steps of the intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality as described above.

[0092] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A smart cold chain logistics distribution method based on real-time monitoring of fresh milk quality, characterized in that, The method includes: Acquire historical cold chain data from the cold chain transportation system, identify typical cold chain scenarios, construct cold chain scenario models, extract environmental characteristic parameters of typical cold chain scenarios, and generate corresponding quality impact labels. The quality impact label and the acquired real-time monitoring data of fresh milk are used as inputs, and the data are processed based on the spatiotemporal fusion analysis mechanism to output the quality status trajectory of fresh milk in different cold chain links and quality anomaly data. Set up a digital twin platform to map the quality status trajectory and quality anomaly data into the digital twin platform, and construct a delivery route map structure and a multi-dimensional quality correlation map. Based on the analysis of the constraint boundary between delivery timeliness and quality assurance using a multidimensional quality correlation graph, and combined with cold chain resource scheduling data and delivery priority parameters, a dynamic delivery adjustment window set is output to generate an intelligent delivery optimization strategy. The process of acquiring historical cold chain data from the cold chain transportation system, identifying typical cold chain scenarios, and constructing a cold chain scenario model includes: By acquiring historical cold chain data generated during the historical delivery process of the cold chain transportation system, the historical cold chain data is constructed into a cold chain data matrix according to spatial location and time sequence; The cold chain data matrix is ​​divided into blocks based on the spatiotemporal grid partitioning mechanism to obtain multiple historical cold chain units, and the local environmental fragments corresponding to each historical cold chain unit are extracted. Multidimensional environmental features are extracted for each local environmental segment. An environmental feature vector is formed by combining the multidimensional environmental features extracted from the same local environmental segment. An environmental feature vector set is constructed by combining the environmental feature vectors of all local environmental segments. Cluster analysis is performed on the set of environmental feature vectors to identify representative typical cold chain scenarios. Based on the multi-dimensional environmental features extracted from each typical cold chain scenario, a cold chain scenario model is constructed. The process of extracting environmental feature parameters of typical cold chain scenarios and generating corresponding quality impact labels includes: For various typical cold chain scenarios included in the cold chain scenario model, environmental feature parameters are extracted as quality impact descriptors based on statistically representative multidimensional environmental features in the environmental feature vector set. Each quality impact description item is input into a preset quality rule set. The quality rule set is constructed based on historical quality monitoring data of fresh milk and quality grade samples through multi-factor variance analysis or Bayesian network inference, and outputs the range of quality impact intensity coefficients and its quality risk level. By identifying the quality impact intensity coefficient ranges for various typical cold chain scenarios, quality risk levels are linked to corresponding typical cold chain scenarios, and quality impact labels are generated.

2. The intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality as described in claim 1, characterized in that, The process of taking quality impact labels and the acquired real-time monitoring data of fresh milk as input, processing them based on a spatiotemporal fusion analysis mechanism, and outputting the quality status trajectory of fresh milk in different cold chain links and quality anomaly data is as follows: The real-time monitoring data of fresh milk acquired by multiple sensors during cold chain transportation is continuously collected and spatiotemporally registered. The real-time monitoring data includes fresh milk temperature data, humidity data, vibration data, gas composition data, and location trajectory records. Based on the spatiotemporal fusion analysis mechanism, real-time monitoring data is processed for spatiotemporal correlation to form quality monitoring segments with spatiotemporal coupling characteristics; each quality monitoring segment is associated and bound with the corresponding quality impact label to construct a quality evolution sample set. Based on the quality evolution sample set and each typical cold chain link, multiple quality monitoring segments are fused and trend analyzed in spatiotemporal order. The dynamic change characteristics of quality parameters in each cold chain link are extracted, quality feature nodes are constructed, and the links are correlated in spatiotemporal order to form a quality change trajectory. The quality change trajectory is then analyzed based on deep learning combined with multidimensional environmental features.

3. The intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality as described in claim 2, characterized in that, The process of analyzing the trajectory of quality change based on deep learning and multi-dimensional environmental features includes: Identify the spatiotemporal intervals of quality anomalies, extract the change gradient, fluctuation amplitude, and critical threshold points of environmental characteristic parameters for the spatiotemporal intervals of quality anomalies, and extract the remaining quality maintenance time in the cold chain corresponding to the spatiotemporal intervals of quality anomalies, which is recorded as the quality safety period. The change gradient, fluctuation amplitude, critical threshold points, and quality safety period are recorded as quality anomaly data, and the quality change trajectory corresponding to the spatiotemporal intervals of quality anomalies is recorded as the quality status trajectory.

4. The intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality as described in claim 3, characterized in that, The process of setting up a digital twin platform, mapping quality status trajectories and quality anomaly data to the digital twin platform, and constructing a delivery route map structure and a multi-dimensional quality correlation map includes: The quality status trajectory and quality anomaly data are correlated to each virtual cold chain node in the digital twin platform. Based on the spatiotemporal sequence of each cold chain link, a multi-node status mapping chain for the fresh milk delivery process is established, and the multi-node status mapping chain is used as the basic spatiotemporal axis. Based on location trajectory records, the corresponding quality feature nodes in each cold chain link are aggregated to construct delivery nodes that correspond one-to-one with the cold chain links, and quality anomaly data is bound to the corresponding delivery nodes. The quality anomaly data and the corresponding change parameters in each delivery node in the quality status trajectory are recorded as the quality attribute markers of the delivery node; the quality safety period in the quality anomaly data is used as the quality label in the current cold chain link and marked to the corresponding delivery node. Based on the quality and safety period, the cold chain links are analyzed to identify the upper and lower thresholds of the time constraints of the delivery nodes, and to form the delivery time boundary. The upper threshold of the time constraint is the maximum allowable delivery completion time of the current cold chain link, derived from the quality and safety period, and the lower threshold of the time constraint is the earliest time limit that the current cold chain link must complete, set based on business requirements or customer commitments. Based on the basic spatiotemporal axis, the delivery nodes are arranged in spatiotemporal order, and path edges are constructed between the delivery nodes. The delivery nodes are connected through the path edges to construct a delivery path graph. Quality attribute markers, quality labels, and delivery timeliness boundaries are bound to the corresponding delivery nodes and edges, and mapped to the virtual cold chain model in the digital twin platform to construct a multi-dimensional quality association graph.

5. The intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality as described in claim 4, characterized in that, The process of analyzing the constraint boundary between delivery timeliness and quality assurance based on multi-dimensional quality correlation graphs, combining cold chain resource scheduling data and delivery priority parameters, outputting a dynamic delivery adjustment window set, and generating an intelligent delivery optimization strategy includes: Obtain the set of delivery nodes, the set of quality tags, and the delivery time boundary in the multidimensional quality correlation graph; identify the delivery node group that has a risk of conflict between the delivery time boundary and the quality status trajectory; and determine the corresponding quality risk period. By combining delivery priority parameters, delivery nodes are sorted, and delivery nodes with lower priority that are in the quality risk period are marked as adjustable delivery groups. Optimizable time segments in their delivery time boundary are extracted to construct an adjustable time window set. Based on the cold chain resource scheduling data, identify the time segments that meet the quality assurance conditions in the adjustable time window set, output the dynamic delivery adjustment window set as candidates, and send the adjustable delivery group, the optimizable time segment, and the dynamic delivery adjustment window set to the delivery scheduling center. Combine the adjustable delivery group, the optimizable time segment, and the dynamic delivery adjustment window set to generate the corresponding intelligent delivery optimization plan.

6. An intelligent cold chain logistics and distribution system based on real-time monitoring of fresh milk quality, characterized in that, The system includes: The scenario modeling module is used to acquire historical cold chain data of the cold chain transportation system, identify typical cold chain scenarios, build cold chain scenario models, extract environmental feature parameters of typical cold chain scenarios, and generate corresponding quality impact labels. The quality tracking module takes quality impact labels and real-time monitoring data of fresh milk as input, processes them based on a spatiotemporal fusion analysis mechanism, and outputs the quality status trajectory of fresh milk in different cold chain links and quality anomaly data. The digital mapping module is used to set up the digital twin platform, map the quality status trajectory and quality anomaly data to the digital twin platform, and construct the delivery route map structure and multi-dimensional quality correlation map. The intelligent scheduling module is used to analyze the constraint boundary between delivery timeliness and quality assurance based on multi-dimensional quality correlation graph analysis. It combines cold chain resource scheduling data and delivery priority parameters to output a set of dynamic delivery adjustment windows and generate intelligent delivery optimization strategies. The process of acquiring historical cold chain data from the cold chain transportation system, identifying typical cold chain scenarios, and constructing a cold chain scenario model includes: By acquiring historical cold chain data generated during the historical delivery process of the cold chain transportation system, the historical cold chain data is constructed into a cold chain data matrix according to spatial location and time sequence; The cold chain data matrix is ​​divided into blocks based on the spatiotemporal grid partitioning mechanism to obtain multiple historical cold chain units, and the local environmental fragments corresponding to each historical cold chain unit are extracted. Multidimensional environmental features are extracted for each local environmental segment. An environmental feature vector is formed by combining the multidimensional environmental features extracted from the same local environmental segment. An environmental feature vector set is constructed by combining the environmental feature vectors of all local environmental segments. Cluster analysis is performed on the set of environmental feature vectors to identify representative typical cold chain scenarios. Based on the multi-dimensional environmental features extracted from each typical cold chain scenario, a cold chain scenario model is constructed. The process of extracting environmental feature parameters of typical cold chain scenarios and generating corresponding quality impact labels includes: For various typical cold chain scenarios included in the cold chain scenario model, environmental feature parameters are extracted as quality impact descriptors based on statistically representative multidimensional environmental features in the environmental feature vector set. Each quality impact description item is input into a preset quality rule set. The quality rule set is constructed based on historical quality monitoring data of fresh milk and quality grade samples through multi-factor variance analysis or Bayesian network inference, and outputs the range of quality impact intensity coefficients and its quality risk level. By identifying the quality impact intensity coefficient ranges for various typical cold chain scenarios, quality risk levels are linked to corresponding typical cold chain scenarios, and quality impact labels are generated.

7. An intelligent cold chain logistics distribution device based on real-time monitoring of fresh milk quality, characterized in that, The device includes: a memory, a processor, and an intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality stored in the memory and executable on the processor, wherein the intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality is configured to implement the steps of the intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality. When the intelligent cold chain logistics distribution program based on real-time monitoring of fresh milk quality is executed by a processor, it implements the steps of the intelligent cold chain logistics distribution method based on real-time monitoring of fresh milk quality as described in any one of claims 1 to 5.

Images