Digital cold storage multi-objective path dynamic optimization method based on edge computing

Abstract

This invention relates to the field of edge computing technology, specifically to a dynamic optimization method for multi-objective paths in digital cold storage based on edge computing. The method acquires batch data of agricultural products to be delivered, outbound operation data from digital cold storage, and delivery network data. These are converted into quality degradation characteristics, outbound constraint characteristics, and path operation characteristics at the edge computing node side, respectively. A spatiotemporal constraint graph is constructed, including cold storage outbound time slices, road network path nodes, and delivery node time windows. Delivery time cost, quality degradation cost, temperature control energy consumption cost, and path risk cost are calculated for candidate paths. Through non-dominated sorting, local node replacement, outbound time slice shifting, and delivery node rearrangement, the primary delivery path and alternative delivery paths are determined. This method ensures that the dynamic path optimization of agricultural product cold chain logistics is simultaneously constrained by cold storage operations, quality maintenance, and road network operation, reducing the problem of inconsistencies between path results and actual outbound status.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, specifically to a dynamic optimization method for multi-objective paths in digital cold storage based on edge computing. Background Technology

[0002] Existing digital cold storage facilities typically record agricultural product batches, entry times, storage temperature zones, outbound tasks, and delivery orders through warehouse management programs, and correlate operational data within the cold storage with refrigerated vehicle delivery data. For dynamic route optimization in agricultural cold chain logistics, the conventional approach is to use delivery nodes, delivery time windows, road travel time, and vehicle current location as primary inputs, calculate delivery routes through a centralized server, and then send the calculation results to the cold storage dispatcher or delivery execution unit. Such solutions usually aim for the shortest transportation distance, the least estimated time, or the best delivery sequence. Cold storage-side data is often used as order generation conditions or outbound queuing conditions, while the freshness level of the agricultural products themselves, storage duration, target temperature zone, and allowable delivery time limits are often not converted into quality degradation constraints that can be directly used in route cost calculations.

[0003] In relatively similar conventional technical solutions, digital cold storage generates outbound plans based on order batches. The cold chain logistics platform establishes a route map based on delivery addresses, road conditions, and time windows, and searches the route map for delivery routes that meet traffic conditions. Some solutions incorporate cold storage outbound time, vehicle waiting time, or road congestion information into the route calculation, but their processing methods are mostly fixed weight superposition or manual priority setting. When cold storage outbound time slices change, batch queuing order is adjusted, delivery node time windows tighten, or road traffic times fluctuate, the route cost usually needs to be recalculated back to the central side. Because cold storage operation constraints and transportation route constraints belong to different data links, the nodes and edges in the route map cannot reflect batch quality status, outbound window conflicts, and temperature zone matching deviations in a timely manner.

[0004] The core technical problem arising from this is that existing dynamic path optimization for agricultural cold chain logistics fails to construct a unified spatiotemporal constraint graph at the edge computing side, integrating the batch quality degradation characteristics of agricultural products to be delivered, the constraints of digital cold storage outbound operations, and the operational characteristics of the delivery road network. This results in the multi-objective cost of candidate paths not simultaneously reflecting delivery timeliness, quality degradation, temperature control energy consumption, and path risks. When the status of cold storage outbound operations or the delivery road network changes, path calculation easily remains at a single dimension of distance or time. The generated delivery path may meet the passage conditions but not the agricultural product quality constraints or outbound operation constraints, causing inconsistencies between the path optimization results and the cold chain execution status. Summary of the Invention

[0005] The purpose of this invention is to provide a dynamic optimization method for multi-objective paths in digital cold storage based on edge computing, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A multi-objective path dynamic optimization method for digital cold storage based on edge computing includes: acquiring batch data of agricultural products to be delivered, outbound operation data of digital cold storage, and delivery road network data required for dynamic path optimization of agricultural cold chain logistics; On the edge computing node side, the batch data of agricultural products to be delivered is converted into quality degradation features, the outbound operation data of the digital cold storage is converted into outbound constraint features, and the delivery road network data is converted into path operation features; A spatiotemporal constraint graph is constructed based on the quality degradation characteristics, the outbound constraint characteristics, and the path operation characteristics. The calculation of candidate paths in the spatiotemporal constraint graph includes a multi-objective cost factor encompassing delivery time cost, quality degradation cost, temperature control energy consumption cost, and path risk cost. The method for selecting target delivery paths that meet the constraints of delivery time window, outbound operation, and agricultural product quality based on the multi-objective cost value.

[0007] Preferably, the batch data of agricultural products to be delivered is converted into quality degradation characteristics, including: extracting agricultural product category, storage duration, freshness grade, target temperature zone, allowable delivery time limit and outbound priority from the batch data of agricultural products to be delivered; The warehousing duration and the allowable delivery time limit are normalized according to the quality sensitivity range corresponding to the agricultural product category. Map the freshness rating, the target temperature zone, and the outbound priority to batch constraint codes; Based on the normalization result and the batch constraint encoding, a batch state vector is generated for participating in the quality degradation cost calculation.

[0008] Preferably, the digital cold storage outbound operation data is converted into outbound constraint features, including: extracting the current cold storage capacity occupancy status, outbound time slice, outbound batch queuing sequence, temperature zone scheduling status, and assigned delivery tasks from the digital cold storage outbound operation data; A set of available outbound windows is generated based on the outbound time slice and the queue sequence of the batches to be outbound. Generate outbound conflict constraints based on the temperature zone scheduling status and the allocated delivery tasks; Associate the set of available outbound windows and the outbound conflict constraints with the corresponding batches of agricultural products to be delivered.

[0009] Preferably, the delivery network data is converted into path operation characteristics, including: extracting delivery node time windows, estimated travel time of road segments, historical lateness probability, path stability records and delivery node succession relationships from the delivery network data; A set of directed edges for the road network is formed based on the connection relationship between the delivery nodes; Write the estimated travel time of the road segment and the historical lateness probability into the corresponding edges in the directed edge set of the road network; The delivery node time window is associated with the path stability record and then linked to the corresponding delivery node to form path operation characteristics for path feasibility determination.

[0010] Preferably, the construction of the spatiotemporal constraint graph includes: using the cold storage outbound time slice, the delivery node time window, and the road network path nodes as graph nodes, and using the feasible transfer relationship of agricultural product batches from the cold storage outbound time slice to the delivery node time window via the road network path nodes as graph edges; The quality degradation feature, the outbound constraint feature, and the path operation feature are respectively attached to the corresponding graph nodes or graph edges; Graph edges that do not meet the target temperature zone, outbound time slice, or delivery time window are marked as invalid, while graph edges that meet the constraints are retained for candidate path search.

[0011] Preferably, generating a batch state vector based on the normalization processing result and the batch constraint encoding includes: assigning feature weights to the warehousing duration, the allowable delivery time limit, the freshness grade, and the target temperature zone according to the preset quality sensitivity level of the agricultural product category; The normalization result and the batch constraint code are weighted and combined according to the feature weights; The weighted combination results of different batches of agricultural products to be delivered are sorted among batches to obtain a batch state vector carrying the batch priority relationship.

[0012] Preferably, associating the available outbound window set and the outbound conflict constraint with the corresponding batch of agricultural products to be delivered includes: determining the initial outbound time slice corresponding to each batch of agricultural products to be delivered based on the queuing sequence of the batches to be delivered; When the initial outbound time slice does not match the set of available outbound windows, an alternative outbound time slice that satisfies the constraint relationship with the allowed delivery time limit is selected from the set of available outbound windows. The alternative outbound time slice and the outbound conflict constraint are written together into the path search starting point constraint of the corresponding batch.

[0013] Preferably, calculating the multi-objective cost value for candidate paths in the spatiotemporal constraint graph includes: searching for candidate paths from the cold storage outbound time slice to the delivery node time window among the graph edges that satisfy the constraints; For each candidate path, the waiting time of graph nodes, the travel time of graph edges, the quality degradation increment, the temperature zone matching deviation, and the lateness risk record are accumulated respectively. The waiting time of the graph nodes and the travel time of the graph edges are used to form the delivery time cost, the quality decay increment is used to form the quality decay cost, the temperature zone matching deviation is used to form the temperature control energy consumption cost, and the lateness risk record is used to form the path risk cost.

[0014] Preferably, the selection of target delivery routes based on the multi-objective cost value includes: sorting the candidate routes in a non-dominated manner according to the delivery time cost, the quality degradation cost, the temperature control energy consumption cost, and the route risk cost to obtain a set of non-dominated candidate routes; In the set of non-dominated candidate paths, local node replacement, outbound time slice shifting, or delivery node rearrangement are performed on candidate paths that have conflicts in outbound operations, delivery time windows, or agricultural product quality constraints. Candidate routes that still meet the constraints after conflict resolution will be selected as target delivery routes.

[0015] Preferably, the step of taking the candidate path that still satisfies the constraints after conflict resolution as the target delivery path includes: determining the target weight combination of different agricultural product batches to be delivered based on the batch state vector carrying the batch priority relationship among the candidate paths that still satisfy the constraints after conflict resolution. The delivery time cost, quality degradation cost, temperature control energy consumption cost, and route risk cost are comprehensively ranked according to the target weight combination. Candidate paths that satisfy the outbound operation constraints, the delivery time window, and the agricultural product quality constraints from the comprehensive ranking results are determined as primary delivery paths, and candidate paths that are ranked adjacently and satisfy the constraints are determined as alternative delivery paths.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By acquiring batch data of agricultural products to be delivered, outbound operation data of digital cold storage, and delivery road network data at the edge computing node side, and converting them into quality degradation characteristics, outbound constraint characteristics, and path operation characteristics respectively, the dynamic path optimization of agricultural product cold chain logistics no longer relies solely on road distance or estimated travel time. By constructing a spatiotemporal constraint graph using cold storage outbound time slices, delivery node time windows, and road network path nodes, and using feasible transition relationships as graph edges, the path search process can be simultaneously constrained by agricultural product quality status, cold storage outbound window, temperature zone requirements, and delivery time window. By calculating delivery time cost, quality degradation cost, temperature control energy consumption cost, and path risk cost for candidate paths, the evaluation basis of candidate paths remains consistent with the key constraints in the cold chain execution process, reducing constraint omissions caused by a single path calculation dimension. After non-dominated sorting, local node replacement, outbound time slice translation, and delivery node rearrangement, the target delivery path can be generated under the condition of satisfying cold storage operation constraints, delivery time window, and agricultural product quality constraints, maintaining a correspondence between the path optimization results and the actual outbound status of the digital cold storage.

[0017] 2. By characterizing agricultural product categories, storage duration, preservation level, target temperature zone, allowable delivery time limit, and outbound priority, a batch state vector carrying batch priority relationships is formed, providing a calculable quality constraint basis for different batches of agricultural products to be delivered during route selection. By extracting storage capacity occupancy status, outbound time slices, outbound batch queuing sequences, temperature zone scheduling status, and assigned delivery tasks from digital cold storage outbound operation data, and generating a set of available outbound windows and outbound conflict constraints, the path search starting point can correspond to the actual outbound resource occupancy status. By converting delivery node time windows, estimated road segment travel time, historical lateness probability, path stability records, and delivery node succession relationships into path operation features, the travel cost and risk cost of candidate paths have a unified data source. Through the generation of primary and alternative delivery paths, subsequent path switching has a candidate basis, reducing the computational burden caused by repeated mapping and screening. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of a multi-objective path dynamic optimization method for digital cold storage based on edge computing; Figure 2 Flowchart for generating batch quality degradation characteristics and batch state vectors for agricultural products; Figure 3 Flowchart for generating outbound constraint features and matching outbound time slices for digital cold storage; Figure 4 Flowchart for constructing spatiotemporal constraint graphs and marking graph edges for feasibility; Figure 5 A flowchart is generated for multi-objective evaluation of candidate paths, constraint repair, and primary / backup path generation. Detailed Implementation

[0019] In one embodiment, as shown in the appendix Figure 1 A dynamic optimization method for multi-objective paths in digital cold storage based on edge computing is executed on the edge computing node side of the digital cold storage. The edge computing node is a computing execution end deployed in the local data processing environment of the cold storage, used to receive cold storage operation data, agricultural product batch data, and delivery road network data, and to complete feature transformation, constraint graph construction, candidate path evaluation, and target path selection locally. This method is designed for dynamic path optimization scenarios in agricultural product cold chain logistics, processing agricultural product batches with established delivery tasks, and calculating feasible paths between cold storage outbound time slices, delivery node time windows, and road network path nodes. Its working principle is to convert the originally separately recorded cold storage outbound operation status, agricultural product quality status, and external delivery road network status into constraint graph elements in the same computational space. This ensures that the path search process is simultaneously constrained by delivery timeliness, quality degradation, temperature control energy consumption, and path risk, rather than generating delivery paths solely based on distance or estimated travel time.

[0020] In this embodiment, the batch data of agricultural products to be delivered includes the agricultural product category, storage duration, preservation level, target temperature zone, allowable delivery time limit, and outbound priority. The digital cold storage outbound operation data includes the current storage capacity occupancy status, outbound time slice, queue sequence of batches to be delivered, temperature zone scheduling status, and assigned delivery tasks. The delivery road network data includes delivery node time windows, estimated travel time for road segments, historical lateness probability, path stability records, and delivery node continuity relationships. Before entering path calculation, the above data needs to undergo field validation and time-series alignment. Field validation is used to eliminate data anomalies such as missing batch numbers, missing delivery nodes, and incorrect relationships between time windows; time-series alignment is used to map the storage time, outbound time slice, road segment travel time, and delivery time window to the same time base. Specifically, the edge computing node uses the current cold storage operation time as a reference to uniformly convert the remaining deliverable time, optional outbound time slice, and reachable delivery node time window of the batch of agricultural products to be delivered into discrete time slice expressions, facilitating subsequent graph node matching and graph edge reachability judgment in the spatiotemporal constraint graph.

[0021] Table 1 shows the correspondence between the three types of input data and feature representations in this embodiment. The contents of the table are used to explain how different data objects are converted into computational elements that can participate in path search after entering the edge computing node. In actual implementation, equivalent mapping can be performed based on the naming of cold storage data fields.

[0022] Table 1. Correspondence between the three types of input data and feature representations Batch data of agricultural products to be delivered Agricultural product category, storage duration, preservation level, target temperature zone, allowable delivery time limit, and outbound priority. Quality degradation features and batch state vector Quality degradation cost calculation, batch priority relationship determination, and target weight combination generation. Digital cold storage outbound operation data Current warehouse capacity occupancy status, outbound time slots, waiting batch queues, temperature zone scheduling status, and assigned delivery tasks. Outbound constraint characteristics, available outbound window set, outbound conflict constraints Path search starting point constraints, outbound time slice shifting, and outbound operation conflict handling. Delivery network data Delivery node time window, estimated travel time for road segment, historical lateness probability, route stability records, and delivery node succession relationships. Path running characteristics and the set of directed edges in the road network Graph edge generation, path feasibility assessment, delivery time cost and path risk cost calculation In one embodiment, as shown in the appendix Figure 2When edge computing nodes convert batch data of agricultural products awaiting delivery into quality degradation characteristics, they first establish a category benchmark table according to the agricultural product category. This table stores the corresponding agricultural product's shelf-life reference duration, delivery time limit reference value, and quality sensitivity level. For each batch of agricultural products awaiting delivery, the edge computing node extracts the storage duration and allowable delivery time limit, and converts them into dimensionless feature values. The shelf-life level can be converted into a numerical value according to existing level codes in cold storage management, the target temperature zone can be converted into a temperature zone category code, and the outbound priority can be converted into a batch sorting factor. Through the above processing, agricultural product batches are transformed from raw business records into batch status expressions that can participate in mathematical operations.

[0023] The quality degradation characteristic can be calculated using the following formula: in, Indicates batch The quality degradation characteristic value, the larger the value, the tighter the quality constraint that the batch is subjected to in the path calculation; Indicates batch Agricultural product category; Indicates batch The duration of the inbound process; Indicates product category Corresponding preservation reference time; Indicates batch The remaining allowable delivery time; Indicates product category Corresponding delivery time reference values; Indicates the value for conversion of preservation level; , and These represent the weights of the warehousing duration, remaining allowable delivery time, and freshness grade within the product category. For example, if a batch has a warehousing duration of 12, a category-specific freshness reference duration of 48, a remaining allowable delivery time of 20, a category-specific delivery time reference value of 40, and a freshness grade conversion value of 0.4, with weights of 0.5, 0.3, and 0.2 respectively, the calculation result is 0.5 multiplied by 12 divided by 48, plus 0.3 multiplied by the 1 in parentheses minus 20 divided by 40, plus 0.2 multiplied by 0.4, resulting in 0.355. This result serves as the fundamental characteristic for calculating the quality degradation cost of the batch.

[0024] Furthermore, when edge computing nodes generate batch state vectors, they do not use only a single quality decay feature value. Instead, they form vector components from agricultural product category, storage duration, freshness grade, target temperature zone, allowable delivery time limit, and outbound priority. Specifically, agricultural product category is used to find the quality sensitivity level and benchmark parameters; storage duration is used to represent the storage time pressure; freshness grade is used to represent the quality boundary of the current batch; target temperature zone is used to limit the temperature zone matching of subsequent transportation links; allowable delivery time limit is used to constrain the reachability time of the path endpoint; and outbound priority is used to handle the situation where multiple batches compete for the same outbound time slice simultaneously. In this embodiment, the batch state vector is written into the path search task and subsequently referenced in outbound window matching, candidate path cost calculation, and primary / backup path selection to maintain the consistency of batch features in different computing stages.

[0025] In a preferred embodiment, edge computing nodes assign feature weights to storage duration, allowable delivery time, freshness grade, and target temperature zone according to a preset quality sensitivity level for each agricultural product category. The quality sensitivity level can be determined by existing category-specific freshness rules in the cold storage. Categories more susceptible to storage time and delivery delays have higher time-related weights, while categories more affected by temperature zone deviations have higher temperature-related weights. After weighted combination, the edge computing nodes sort all batches within the same delivery batch set, and the sorting results form batch priority relationships. These batch priority relationships do not directly replace path cost calculations but are used to determine the target weight combination for different batches when multiple candidate paths meet the basic constraints. Therefore, the path selection process can distinguish the constraint strength of different agricultural product batches on timeliness, quality, and temperature zone, avoiding treating all batches with the same cost function.

[0026] In one embodiment, as shown in the appendix Figure 3 When edge computing nodes convert digital cold storage outbound operation data into outbound constraint features, they first read the current storage capacity occupancy status and the queue sequence of batches awaiting outbound processing. The current storage capacity occupancy status is used to determine whether outbound resources are occupied in a specific temperature zone or outbound area; the queue sequence of batches awaiting outbound processing is used to determine the initial outbound order of the batches to be delivered; the outbound time slice describes the discrete time intervals within the cold storage operation sequence during which batches can complete outbound preparation and handover; the temperature zone scheduling status is used to determine whether batches with the same or similar target temperature zones can be outbound in adjacent time slices; and the assigned delivery tasks are used to identify the delivery resources already occupied or the succession relationships of delivery nodes within the same outbound time slice. Using this data, the edge computing nodes generate a set of available outbound windows and mark the batches, time slices, and temperature zone statuses that may cause outbound conflicts.

[0027] In this embodiment, the initial outbound time slice for a batch of agricultural products to be delivered is determined by the queuing sequence of the batches to be delivered. If the initial outbound time slice matches the set of available outbound windows, it is directly used as a candidate starting point for path search. If the initial outbound time slice is already occupied by an assigned delivery task, or if an outbound conflict occurs due to temperature zone scheduling status, the edge computing node selects an alternative outbound time slice from the set of available outbound windows. The selection of the alternative outbound time slice is not simply choosing the earliest time slice, but is calculated by combining the allowed delivery time limit and batch quality decay characteristics. The matching cost of the candidate outbound time slice can be determined using the following formula: in, Indicates batch Select candidate outbound time slices Matching cost at that time; This represents the initial outbound time slice determined by the queuing sequence; This represents the discrete distance between the candidate outbound time slice and the initial outbound time slice; Indicates batch Candidate outbound time slice The outbound conflict flag is set to 1 if a conflict exists, and 0 if no conflict exists. This indicates the amount of compression that the allowed delivery time limit is reduced after selecting a candidate outbound time slice; Indicates the weight of the outbound conflict flag; This indicates the weight of the delivery time limit reduction. For example, if the initial outbound time slice for a batch is 3, the candidate outbound time slice is 4, the outbound conflict flag is 0, and the delivery time limit reduction is 0.2. Take 3. If we choose 2, the matching cost is 1 plus 0 plus 0.4, resulting in 1.4. If another candidate outbound time slice is 3 but there is an outbound conflict, the matching cost includes a conflict penalty of 3. The edge computing node will prioritize retaining the candidate time slice with the lower matching cost and that meets the basic constraints.

[0028] Furthermore, edge computing nodes will replace the outbound time slice and outbound conflict constraints in the path search starting point constraints for the corresponding batch. The path search starting point constraints include at least the batch number, available outbound time slices, target temperature zone, prohibited time slices, conflicted task identifiers, and the latest allowed arrival time at the delivery node. This starting point constraint corresponds to the cold storage outbound time slice node during the construction of the spatiotemporal constraint graph; only graph edges satisfying the starting point constraint are allowed to connect to subsequent road network path nodes. In this embodiment, the setting of outbound constraint features transforms the internal operational state of the cold storage into boundary conditions in the path graph, avoiding situations where an external path is feasible but the cold storage cannot complete the outbound process according to the specified time slice.

[0029] In one embodiment, when edge computing nodes convert delivery network data into path operation characteristics, they first form a set of directed edges for the network based on the connection relationships between delivery nodes. The start and end points of the directed edges can be cold storage outbound nodes, network path nodes, or delivery nodes; the direction of the directed edges represents the feasible flow direction of delivery vehicles or delivery tasks in time. Each directed edge is accompanied by the estimated travel time of the road segment, historical lateness probability, and path stability record. Delivery node time windows are associated with the corresponding delivery nodes to constrain the arrival time of candidate paths. Path stability records can be formed based on historical fluctuations in travel time of the same road segment, temporary unreachability records, or frequent replanning records; the lower the value, the easier it is for the road segment to be subsequently corrected in dynamic path optimization.

[0030] The route operation characteristics can be used to form the segment operation cost using the following formula: in, Represents a directed edge The operational value of the road section; Represents a directed edge The estimated travel time; Represents a directed edge Historical reference travel time for the corresponding road section; Represents a directed edge The historical probability of being late; Represents a directed edge The path stability value is taken as follows: the closer the value is to 1, the more stable the historical execution is. , and These are the weights for travel time, probability of lateness, and stability deviation, respectively. For example, if a directed edge has an estimated travel time of 30, a historical reference travel time of 25, a historical probability of lateness of 0.2, and a path stability of 0.7, with weights of 0.5, 0.3, and 0.2 respectively, then the road segment's operating cost is 0.5 × 30 ÷ 25 + 0.3 × 0.2 + 0.2 × 0.3 = 0.72. This value is written into the directed edge for subsequent joint assessment of path risk cost and delivery time cost.

[0031] In this embodiment, delivery node time windows and path stability records are associated with the corresponding delivery nodes. If a candidate path arrives at a delivery node earlier than the start time of the time window, the edge computing node writes the waiting time into the graph node waiting cost; if the arrival time is later than the end time of the time window, the candidate path is marked as a delivery time window conflict, and subsequently enters constraint repair or is eliminated. Delivery node succession relationships are used to restrict the reachability order between delivery nodes. For example, before completing delivery at a certain delivery node, the same delivery batch is not allowed to jump to a node with no succession relationship. Thus, the delivery network data does not participate in the calculation in the form of a single distance matrix, but forms a set of directed edges that includes travel time, lateness probability, stability, and time window constraints.

[0032] In one embodiment, as shown in the appendix Figure 4 Edge computing nodes construct a spatiotemporal constraint graph based on quality degradation characteristics, outbound constraint characteristics, and path operation characteristics. This spatiotemporal constraint graph uses cold storage outbound time slices, delivery node time windows, and road network path nodes as graph nodes, and the feasible transfer relationships of agricultural product batches from the cold storage outbound time slice to the delivery node time window via road network path nodes as graph edges. Cold storage outbound time slice nodes record available outbound window sets, outbound conflict constraints, and target temperature zones; road network path nodes record reachable road segments and travel times; delivery node time windows record allowed reachable intervals, node continuity relationships, and delivery task constraints. Feasibility assessment is required before connecting graph edges; only transfers that satisfy time, temperature zone, outbound conflict, and node continuity relationships become searchable graph edges.

[0033] The feasibility of a graph edge can be expressed by the following formula: in, Indicates batch From graph nodes Transfer to graph node The feasibility is indicated by a value of 1, which represents feasibility, and a value of 0, which represents infeasibility. This indicates an indicator operation; the value is 1 if the condition inside the parentheses is true, and 0 if it is false. Represents a node To the node The available time span between; Represents a node To the node Expected transfer time between them; Indicates arrival at node The expected time; Represents a node The set of allowed arrival time windows; Indicates batch The target temperature range; Represents a node The set of temperature zones that are allowed to match; Indicates batch At the node The outbound conflict flag is displayed at the location. For example, if the available time span is 18, the estimated transfer time is 15, the arrival time is within the allowable time window, the batch target temperature zone belongs to the node's allowable temperature zone set, and the outbound conflict flag is 0, then all four indication results will be 1. The result is equal to 1; if any condition is not met, the product is 0, and the edge of the graph is marked as invalid.

[0034] Table 2 shows the different graph elements in the spatiotemporal constraint graph. In the table, "invalid marker" indicates that the graph edge does not participate in the candidate path search, but the relevant data can be retained for subsequent conflict analysis and path repair.

[0035] Table 2. Composition methods of different graph elements in the spatiotemporal constraint graph. Cold storage outbound time segment node Available outbound window set, outbound batch queuing sequence, temperature zone scheduling status Is the outbound time slice available? Are there any outbound conflicts? Is the target temperature zone matched? If the constraints are met, the path can be selected as a candidate; otherwise, an invalid flag is added. Road network path nodes Delivery node connection relationship, estimated travel time of road segment, and route stability record. Whether a node is reachable, whether subsequent edges exist, and whether path stability are included in the cost calculation. Connect directed edges that satisfy the feasibility requirement, and record the waiting time of each node. Delivery node time window Delivery node time window, historical lateness probability, delivery task constraints Does the arrival time fall within the time window? Does it cause a delivery node conflict? Participate in candidate route screening as the end point or intermediate delivery node of the route. Image edge Directed edge set, transfer duration, temperature zone matching relationship, outbound conflict marker Does the transfer time meet the requirements? Are the temperature zones compatible? Do the starting points conflict? If feasible, proceed to candidate path search; if not feasible, retain as a conflict record. Furthermore, when constructing the spatiotemporal constraint graph, edge computing nodes attach quality degradation features to the cold storage outbound time slice nodes and delivery node time window nodes related to the corresponding batch, attach outbound constraint features to the cold storage outbound time slice nodes and graph edges originating from the cold storage outbound nodes, and attach path operation features to the directed edges of the road network and delivery nodes. For graph edges that do not meet the target temperature zone, outbound time slice, or delivery time window, edge computing nodes do not directly delete them, but instead mark them as invalid and record the reasons for invalidity. Invalidity reasons may include outbound time slice conflicts, temperature zone mismatches, delivery time window out-of-bounds, insufficient road segment transfer time, and invalid node continuation relationships. This method of retaining invalidity reasons facilitates subsequent local node replacement, outbound time slice shifting, or delivery node rearrangement when there are insufficient candidate paths, without requiring re-parse of all the original data.

[0036] In one embodiment, as shown in the appendix Figure 5 Candidate path search is performed on graph edges that satisfy the constraints. Edge computing nodes start from the cold storage outbound time slice node and search along feasible graph edges to the delivery node time window. During the search, each candidate path contains a set of graph nodes and a set of graph edges. Graph nodes provide waiting time, delivery time window, and outbound constraints; graph edges provide travel time, quality degradation increment, temperature zone matching deviation, and late arrival risk records. Since there may be multiple available outbound time slices and multiple reachable delivery paths for the same batch, edge computing nodes retain multiple candidate paths during the search, rather than directly determining a unique result in local selection. This allows for sufficient candidate space for subsequent multi-objective evaluation.

[0037] The multi-objective cost of a candidate path can be represented by the following vector: in, Indicates batch In candidate paths The multi-objective cost vector on; The delivery time cost is represented by the sum of the waiting time of each graph node and the travel time of each graph edge in the candidate path. This represents the cost of quality degradation, calculated from the candidate path and batch. The related quality degradation increments accumulate to form; This represents the energy consumption cost of temperature control, which is formed by the cumulative temperature zone matching deviation in the candidate paths; The path risk cost is represented by the sum of the late arrival risk records of each graph edge in the candidate path. Indicate candidate path The set of graph nodes contained therein; Indicate candidate path The set of edges contained in the graph; Represents graph nodes The corresponding waiting time; Represents graph edges Travel time; Indicates batch After the graph edge The resulting increase in quality degradation; Indicates batch After the graph edge The resulting temperature range matching deviation; Represents graph edges The lateness risk record is recorded. For example, a candidate path contains 2 waiting nodes with waiting times of 3 and 2 respectively, and contains 3 graph edges with travel times of 10, 8 and 12 respectively. Then the delivery time cost is 35; if the quality decay increments corresponding to the 3 graph edges are 0.04, 0.03 and 0.05, the quality decay cost is 0.12; if the temperature zone matching deviation is 1, 0 and 2, the temperature control energy consumption cost is 3; if the lateness risk record is 0.1, 0.2 and 0.1, the path risk cost is 0.4.

[0038] In this embodiment, delivery timeliness cost, quality degradation cost, temperature control energy consumption cost, and route risk cost are retained separately and not directly merged into a single score in the initial evaluation stage. This is because different agricultural product batches have different constraints, and the suitability of the same candidate route may vary for different batches. For batches with a longer warehousing duration and a shorter remaining allowable delivery time, quality degradation cost and delivery timeliness cost need to account for a higher proportion in subsequent ranking; for batches with strong target temperature zone constraints, temperature control energy consumption cost needs to be retained in the candidate route comparison. Edge computing nodes use a vector method to store cost values, which can adjust the comparison rules according to the batch state vector in subsequent non-dominated ranking and comprehensive ranking.

[0039] In one embodiment, edge computing nodes filter target delivery routes based on multi-objective cost. The filtering process first performs non-dominated ranking of candidate routes. If a candidate route is no higher than another candidate route in terms of delivery timeliness cost, quality degradation cost, temperature control energy consumption cost, and route risk cost, and at least one cost is lower than the other candidate route, then the former candidate route dominates the latter candidate route in the current batch. This comparison rule can be expressed by the following formula: in, Indicate candidate path In batch Next dominant candidate path For example, candidate paths The multi-objective costs are 35, 0.12, 3, and 0.4, and the candidate paths are... The multi-objective costs are 38, 0.15, 3, and 0.5, then the candidate paths are... It is no higher than the candidate path in all four costs. Among them, three have lower costs, and these are candidate paths. Dominant candidate path If candidate path If the delivery time cost is 32 but the quality degradation cost is 0.2, then there is no unidirectional dominance relationship between the two paths, and both need to be retained to enter the non-dominated candidate path set.

[0040] In this embodiment, after the set of non-dominated candidate paths is formed, edge computing nodes perform constraint repair on candidate paths that have conflicts such as outbound operation conflicts, delivery time window conflicts, or agricultural product quality constraint conflicts. For outbound operation conflicts, edge computing nodes prioritize selecting adjacent alternative time slices from the available outbound window set in the same batch and recalculate feasible graph edges from the alternative time slice to subsequent road network path nodes. For delivery time window conflicts, edge computing nodes search for alternative road network path nodes or adjust the delivery node sequence without changing the cold storage outbound time slice, so that the arrival time falls back into the delivery time window. For agricultural product quality constraint conflicts, edge computing nodes reduce the quality degradation cost in candidate paths by shortening waiting nodes, replacing high-risk graph edges, or adjusting outbound time slices. After constraint repair is completed, candidate paths that still do not meet the basic constraints are eliminated, and candidate paths that meet the constraints are entered into the comprehensive ranking.

[0041] Table 3 shows the correspondence between candidate path conflict types and processing methods. This table illustrates that constraint repair is not a uniform recalculation, but rather a localized processing based on the location of the conflict.

[0042] Table 3. Correspondence between candidate path conflict types and handling methods Outbound operation conflict The cold storage outbound time slot has been occupied, or the temperature zone scheduling status does not allow the corresponding batch to be outbound. Perform outbound time slice shifting within the available outbound window set and update the path search start point constraint. The relocation does not violate the permitted delivery time limit and matches the target temperature zone. Delivery time window conflict The time when the candidate path arrives at the delivery node is later than the end of the allowed time window. Perform local node replacement or delivery node rearrangement, and recalculate the feasibility of subsequent graph edges. After rearrangement, the arrival times of each delivery node are located within the corresponding time window. Conflicts in agricultural product quality constraints The cumulative quality degradation cost of candidate paths exceeds the batch allowable boundary. Shorten waiting nodes, replace high-risk graph edges, or select earlier outbound time slices. The adjusted quality degradation cost satisfies the batch state constraint. Path risk conflict The cumulative number of edges in the high-latest-risk graph along the path exceeds the acceptable range for the current batch. Replace graph edges with a higher risk of being late, and retain the original paths as conflict records. The replacement will not create new outbound operation conflicts or delivery time window conflicts. Furthermore, in this embodiment, local node replacement only applies to the node where the conflict occurred and its adjacent graph edges, without changing the path prefix that already satisfies the constraints; the outbound time slice translation only changes the cold storage outbound time slice node corresponding to the path start point and reconnects the graph edges related to that start point; the delivery node rearrangement is only performed within the allowed range of delivery node succession relationships, without disrupting the node constraints of the original delivery task. The above processing methods keep candidate path repair within the affected local range, reduce the computational cost of re-searching the entire path due to a single conflict, and preserve path segments that already satisfy the constraints.

[0043] In one embodiment, candidate paths that still meet the constraints after conflict resolution need to determine the primary delivery path and alternative delivery paths. Edge computing nodes determine target weight combinations for different batches of agricultural products to be delivered based on batch state vectors carrying batch priority relationships. The target weight combination includes at least delivery time weight, quality degradation weight, temperature control energy consumption weight, and path risk weight. For batches with shorter remaining allowable delivery time limits, the delivery time weight and quality degradation weight are relatively high; for batches with strong target temperature zone constraints, the temperature control energy consumption weight is relatively high; for batches with tight delivery node time windows or passing through highly volatile road sections, the path risk weight is relatively high. The target weight combination is generated from the batch state vector and changes with batch characteristics.

[0044] The overall ranking value can be calculated using the following formula: in, Indicates batch In candidate paths The overall ranking value; Indicates batch The corresponding delivery timeliness weight; Indicates batch The corresponding quality degradation weight; Indicates batch The corresponding temperature control energy consumption weight; Indicates batch The corresponding path risk weight; , , and The meaning is consistent with the corresponding component in the aforementioned multi-objective cost vector. For example, if the multi-objective costs of a batch of candidate paths are 35, 0.12, 3, and 0.4, with corresponding weights of 0.4, 0.3, 0.2, and 0.1, then the comprehensive ranking value is 0.4×35+0.3×0.12+0.2×3+0.1×0.4=14.676. Edge computing nodes arrange candidate paths from low to high according to the comprehensive ranking value, and determine the main delivery path from the candidate paths that meet the constraints of outbound operation, delivery time window, and agricultural product quality. Candidate paths that are ranked adjacently and meet the constraints are determined as alternative delivery paths.

[0045] In this embodiment, both the primary delivery path and the alternative delivery paths include a cold storage outbound time slice, the path nodes along the route, the delivery node time window, the corresponding multi-objective cost, and conflict records. The primary delivery path is used for the current delivery task execution, while the alternative delivery paths are used for switching judgments when the cold storage outbound operation status or the delivery road network status changes. The alternative delivery paths do not directly replace the primary delivery path; their significance lies in the fact that the edge computing nodes have already completed the feasibility check and multi-objective cost calculation within the spatiotemporal constraint graph. When local disturbances occur subsequently, a path that meets the constraints can be selected from the alternative delivery paths first, without needing to reconstruct the entire candidate space from the original data. This processing can maintain the correspondence between the dynamic adjustment of the paths and the actual outbound status of the cold storage.

[0046] In a preferred embodiment, before performing feature transformation on the batch data of agricultural products to be delivered, the edge computing node also performs batch integrity verification. The batch integrity verification includes category field verification, time field verification, temperature zone field verification, and delivery task field verification. Category field verification confirms that the batch matches the category benchmark table; time field verification confirms a computable relationship between the warehousing time, the allowed delivery time limit, and the delivery node time window; temperature zone field verification confirms that the target temperature zone can be mapped to the existing temperature zone code in the digital cold storage; and delivery task field verification confirms that the batch is associated with at least one delivery node. For data that fails verification, the edge computing node does not include it in the candidate path search but instead writes the reason for the anomaly into the pending processing record. For data with complete fields but slight time discrepancies, such as the warehousing time and the outbound queuing time slice boundary not completely coinciding, the edge computing node merges the data according to a unified time slice rule, ensuring that the batch data has a consistent time benchmark when entering subsequent calculations.

[0047] In this embodiment, the freshness grade conversion value can be discretely mapped according to cold storage management rules. For example, the freshness grade can be converted into values ​​such as 0.2, 0.4, 0.6, and 0.8, with larger values ​​indicating higher preservation pressure. This conversion does not change the original business meaning of the freshness grade, but rather enables it to participate in the calculation of quality degradation characteristics. After the target temperature zone is converted into a temperature zone category code, it participates in both the temperature zone matching judgment in the spatiotemporal constraint graph and the calculation of temperature control energy consumption costs. The allowed delivery time limit is converted into a remaining time limit proportion to reflect the urgency between the current cold storage operation time and the latest delivery time. Through the above conversions, the original batch data no longer remains in a text or enumeration state, but becomes a numerical feature or set feature that can be referenced by path search and cost calculation.

[0048] In a preferred embodiment, when edge computing nodes process digital cold storage outbound operation data, they use outbound time slices as the basic time sequence unit. Each outbound time slice corresponds to a cold storage operation period that can be occupied by a delivery task. The queuing sequence of batches to be outbound is used to determine the initial distribution of batches in time slices, the assigned delivery tasks are used to mark time slices that have already been occupied, and the temperature zone scheduling status is used to restrict the arrangement relationship of batches with different target temperature zones in adjacent time slices. If the target temperature zone of a batch does not match the temperature zone status of the current outbound area, the edge computing node writes the outbound time slice into the prohibited occupation set. If the initial outbound time slice of a batch has been occupied by another assigned delivery task, the edge computing node will not directly adjust the assigned task, but will instead find an alternative outbound time slice for the current batch and write the alternative result into the path search starting point constraint.

[0049] Furthermore, the available outbound window set consists of multiple discrete outbound time slices, each with an availability status, temperature zone matching status, and conflict flag. When selecting alternative outbound time slices, edge computing nodes traverse the process from low to high matching costs and check whether candidate time slices cause the allowable delivery time limit to be compressed to an unacceptable state. If multiple alternative outbound time slices satisfy the constraints, multiple candidate starting points are retained, allowing the subsequent spatiotemporal constraint graph to generate multiple candidate paths; if only one time slice satisfies the constraints, that time slice is used as the unique feasible starting point; if no time slice satisfies the constraints, the batch enters the conflict record and does not participate in the current round's target path output. This approach ensures that cold storage operation constraints are limited before path search begins, reducing the contradiction of subsequent paths being reachable but unable to be outbound.

[0050] In a preferred embodiment, the path stability record of the delivery network data is formed by historical path execution deviations. For the same road segment, edge computing nodes can statistically analyze the difference between the estimated travel time and the actual travel time in multiple executions, assigning higher stability values ​​to road segments with smaller differences and lower stability values ​​to road segments with larger differences. The historical lateness probability can be calculated based on the historical arrival records of delivery nodes. If a road segment frequently causes the time window of subsequent delivery nodes to be compressed, then the lateness risk record for that road segment is high. The succession relationship of delivery nodes can be determined by the node order of the current delivery task, cold chain delivery constraints, and reachability relationships. Edge computing nodes only generate corresponding directed edges when the succession relationship is established. Through this processing, the set of directed edges in the road network includes not only travel relationships but also the impact of execution reliability and time windows.

[0051] In this embodiment, the writing positions of path operation features correspond to the graph structure. The estimated travel time for a road segment is written to a graph edge for calculating delivery timeliness costs; historical lateness probabilities are written to graph edges for calculating path risk costs; path stability records can be written to graph edges or stored together with related delivery nodes to determine whether the road segment is prone to causing subsequent time window conflicts; delivery node time windows are written to delivery nodes for arrival time determination; and delivery node continuity relationships are used to generate or prohibit directed edges. All types of path operation features have clear positions in the spatiotemporal constraint graph, allowing for direct reading during subsequent searches, repairs, and sorting without requiring repeated access to the original road network data.

[0052] In one embodiment, after the spatiotemporal constraint graph is constructed, the edge computing node performs invalidation checks on all graph edges. Invalidation checks include target temperature zone checks, outbound time slice checks, delivery time window checks, and continuity checks. The target temperature zone check determines whether the batch's target temperature zone belongs to the node's allowed temperature zone set; the outbound time slice check determines whether there is a conflict with the corresponding outbound starting point; the delivery time window check determines whether the expected arrival time falls within the allowed time window; and the continuity check determines whether the preceding and following nodes connected by the graph edge have a sequential relationship in the delivery task. For graph edges that do not meet any of these conditions, the edge computing node marks them as unsearchable graph edges and records the conditions that triggered the marking.

[0053] In this embodiment, invalidation markers differ from deletion. Deletion removes the source of conflict for subsequent constraint repair, while invalidation markers preserve the original connectivity and conflict causes. For example, if a candidate path is infeasible because its outbound time slice is later than the allowed delivery time limit, the edge computing node can try to shift the outbound time slice forward based on the invalidation cause; if a candidate path is infeasible because the delivery node's time window is out of bounds, the edge computing node can try to replace intermediate road network path nodes or adjust the delivery node connection relationship. If graph edges are directly deleted, the above local repair requires re-retrieval of the original data. By preserving invalidation markers, edge computing nodes can complete candidate path expansion and constraint repair within the same spatiotemporal constraint graph, keeping the computation process continuous.

[0054] In a preferred embodiment, the candidate path search employs a hierarchical expansion approach. Edge computing nodes first expand from cold storage outbound time-slice nodes that satisfy outbound constraints, obtaining the first batch of feasible graph edges connected to these nodes. Then, based on edge travel time and path stability records, they expand to road network path nodes. Subsequently, based on delivery node time windows and node continuity relationships, they expand to delivery nodes. Each time a new node is expanded, the edge computing node synchronously updates its cumulative waiting time, cumulative travel time, cumulative quality degradation increment, cumulative temperature zone matching deviation, and cumulative lateness risk records. If an expansion branch violates the delivery time window or quality constraint at an intermediate node, that branch is stopped from expanding, and the reason for stopping is written into the candidate path record. This approach avoids obviously infeasible branches from continuing to occupy computing resources.

[0055] In this embodiment, the quality degradation increment can be determined jointly based on batch quality degradation characteristics and path dwell time. The more waiting nodes in the candidate path, or the longer the travel time, the easier it is for the quality degradation increment to increase. Temperature zone matching deviation can be formed by the difference between the target temperature zone and the corresponding temperature zone category of the path. If the transportation or handover conditions traversed by the candidate path are consistent with the target temperature zone, the deviation is low; if there is a complete mismatch in temperature zone categories, the deviation is written into the temperature control energy consumption cost. The late arrival risk record is formed by accumulating the historical late arrival probabilities of the graph edges. The more graph edges with high late arrival risk that the path passes through, the higher the path risk cost. By accumulating the above costs separately, the candidate path can retain constraint information in different dimensions.

[0056] In one embodiment, after the non-dominated ranking is completed, the edge computing nodes form one or more sets of non-dominated candidate paths. During ranking, different cost dimensions are not directly and forcibly merged, allowing paths with better delivery timeliness but higher quality degradation costs, or paths with lower quality degradation costs but higher path risks, to enter subsequent comparisons. For agricultural cold chain logistics scenarios, this process avoids prematurely eliminating feasible candidate paths based on a single comprehensive score. The edge computing nodes perform conflict detection on the non-dominated candidate path sets, with the detection content consistent with the constraint types recorded in the spatiotemporal constraint graph. After conflict detection, only conflicting local path segments are repaired; graph nodes and edges that already satisfy the constraints are not changed.

[0057] In this embodiment, the processing logic for local node replacement is as follows: Identify the conflicting graph node, read its predecessor and successor nodes, and search for a replacement node in the spatiotemporal constraint graph that is connectable to both the predecessor and successor nodes. Replacement nodes must satisfy delivery time window, temperature zone matching, and node continuity relationships. If multiple replacement nodes exist, edge computing nodes are sorted according to the combined changes of new travel time, quality degradation increment, and late arrival risk records, and the replacement node with the lowest combined change is selected. The processing logic for shifting outbound time slices is as follows: Search for candidate time slices adjacent to the original outbound time slice in the set of available outbound windows, and recalculate the graph edge feasibility from the new starting point to the original subsequent node. The processing logic for rearranging delivery nodes is as follows: Adjust the order of delivery nodes within the allowable range of delivery node continuity relationships, and re-determine whether the arrival time of each node falls within the corresponding time window. All the above repairs are completed within the original spatiotemporal constraint graph.

[0058] In a preferred embodiment, when multiple batches of agricultural products to be delivered share the same cold storage outbound time slot or the same delivery node, the edge computing node performs inter-batch conflict handling. This handling is based on the batch priority relationship in the batch state vector. If two batches compete for the same outbound time slot, and one batch has a higher quality decay characteristic value and a shorter remaining allowable delivery time, that batch is given priority to retain its current time slot, while the other batch enters the search for an alternative outbound time slot. If two batches compete for the same delivery node time window, the edge computing node selects the delivery order according to the node succession relationship and the remaining width of the delivery time window, and recalculates the multi-objective cost value of the candidate paths for the adjusted batches. This process ensures that the path calculation results for multiple batches remain consistent at the cold storage operation level, avoiding conflicts in the overall outbound arrangement even if individual batch paths are feasible.

[0059] In one embodiment, the generation and saving of the primary delivery route and alternative delivery routes includes route content, cost vectors, and switching conditions. The route content includes the cold storage outbound time slice, the path nodes along the route, the delivery nodes, and the arrival time. The cost vector includes delivery timeliness cost, quality degradation cost, temperature control energy consumption cost, and route risk cost. Switching conditions include conflicts in the original primary delivery route such as outbound time slice conflicts, delivery time window conflicts, quality constraint conflicts, or route risk conflicts. Alternative delivery routes do not change the execution state of the primary delivery route when saved; they are only used as candidates for subsequent dynamic processing by edge computing nodes. If the cold storage outbound operation data or delivery network data changes, the edge computing node can first check whether the alternative delivery routes still meet the constraints before deciding whether to search for new candidate routes.

[0060] In this embodiment, the primary and backup paths can be stored using a path record table. Each record in the path record table includes a batch number, path number, path type, outbound time slice, delivery node sequence, path operation feature summary, multi-objective cost vector, and constraint satisfaction status. The path type of the primary delivery path is marked as currently executing, while the path type of the alternative delivery paths is marked as candidate reserved. If subsequent dynamic changes cause the primary delivery path to no longer meet the constraints, the edge computing node reads the alternative delivery paths under the same batch and re-verifies the outbound operation constraints, delivery time window, and agricultural product quality constraints. If the verification passes, the alternative delivery path can be converted into a new currently executing path; if the verification fails, it will proceed to local repair or re-search. In this way, a closed processing link is formed between dynamic path optimization and changes in cold storage outbound data.

[0061] In a preferred embodiment, the edge computing nodes can also version the intermediate results during the path calculation process. Each time a spatiotemporal constraint graph is generated, a corresponding data version is recorded, including batch data versions, cold storage outbound operation data versions, and delivery road network data versions. Candidate paths, primary delivery paths, and alternative delivery paths are all associated with their corresponding data versions. If subsequent data changes, the edge computing nodes can compare the differences between the old and new versions to determine whether only local graph edges or local graph nodes need to be updated. For example, if only the expected travel time for a certain road segment changes, only the running cost of the directed edges containing that road segment and the delivery time cost of the relevant candidate paths are updated; if a cold storage outbound time slice is occupied, only the graph edges connecting that outbound time slice and the starting constraints of the relevant batches are updated. Versioning reduces the impact of irrelevant data changes on path calculation, ensuring traceability of edge-side processing.

[0062] In this embodiment, data versioning does not change the basic logic of path optimization, but rather serves as a way for edge computing nodes to manage dynamic data. For dynamic path optimization in agricultural cold chain logistics, the cold storage outbound operation status and road network operation status are characterized by continuous changes. If the entire spatiotemporal constraint graph is reconstructed for each change, it will result in redundant calculations. By recording versions and differences, edge computing nodes can update only the affected data objects and reuse the unchanged data objects and path segments. This process, in conjunction with the aforementioned invalidation marking, local repair, and primary / backup path saving, enables edge computing nodes to complete relatively stable dynamic path adjustments locally.

[0063] In one embodiment, the method can also be combined with a human-computer interaction interface to display the route calculation results to cold storage dispatchers. However, the displayed content is merely a way of expressing the calculation results and does not change the route optimization process. The displayed content may include the batch of agricultural products to be delivered, the main delivery route, alternative delivery routes, outbound time slices, delivery node arrival times, route risk warnings, and conflict resolution records. If dispatchers need to see the reason why a candidate route is eliminated, the edge computing node can output the corresponding reason based on the invalidation flag and conflict record, such as outbound time slice conflict, delivery time window exceeding the limit, or quality constraints not being met. This display method allows the source of constraints in the route optimization calculation process to be verified, making it easier to discover abnormal records in cold storage operation data or delivery network data.

[0064] In this embodiment, if a scheduler manually locks a specific outbound time slice, the edge computing node writes the locking result into the outbound conflict constraint and uses it as a condition for prohibiting or prioritizing occupancy in the next path calculation, participating in the spatiotemporal constraint graph construction. If the scheduler adjusts the time window of a delivery task, the edge computing node writes the new time window into the delivery node's time window and reassesses the feasibility of relevant graph edges. Data changes generated by human-computer interaction still enter path optimization through the constraint graph logic of the edge computing node, without bypassing the unified calculation process of quality degradation characteristics, outbound constraint characteristics, and path operation characteristics. Thus, data structure consistency is maintained between manual adjustments and automatic calculations.

[0065] In a preferred embodiment, when processing paths for multiple delivery nodes, edge computing nodes can categorize delivery nodes into mandatory nodes and adjustable-order nodes. Mandatory nodes are determined by the delivery task and cannot be deleted; adjustable-order nodes are nodes whose order can be changed within the permissible range of delivery node succession relationships. During path search, mandatory nodes are written into the spatiotemporal constraint graph as endpoints or critical intermediate nodes, while adjustable-order nodes form multiple successive branches in the candidate path expansion. If the delivery node time window is narrow, the edge computing node will determine the feasibility of arrival in advance during the candidate path search, avoiding the formation of a large number of path branches that will inevitably conflict later. If the delivery node time window is wide, more candidate successive relationships are retained to compare the quality degradation cost and path risk cost in subsequent non-dominated sorting.

[0066] In this embodiment, the reordering of delivery nodes must adhere to the sequential relationship between them, and a node order that contradicts the delivery task is not allowed. For example, if a delivery node can only arrive after another delivery node has completed its turn, the edge computing node will not advance its arrival. If there is no mandatory order between two delivery nodes, the edge computing node can adjust their order based on arrival time, path risk, and quality degradation increment. After reordering, the multi-objective cost vector is recalculated for each candidate path, and the delivery time window is rechecked. This process ensures that the reordering of delivery nodes becomes a constrained local repair, rather than an unbounded path rearrangement.

[0067] In one embodiment, the edge computing node processes temperature control energy consumption costs based on the matching deviation between the target temperature zone and the path temperature zone. Each batch of agricultural products has a target temperature zone, and cold storage outbound time slices, some path nodes, or delivery handover nodes have a set of permissible temperature zones. When the target temperature zone belongs to the permissible temperature zone set, the temperature zone matching deviation is low; when the target temperature zone does not belong to the permissible temperature zone set but can be temporarily connected through adjacent temperature zones, the temperature zone matching deviation is recorded as a high value; when the target temperature zone is completely mismatched and would violate quality constraints, the corresponding graph edge is directly marked as unsearchable. The temperature zone matching deviation, as a component of temperature control energy consumption costs, is distinguished from the quality degradation cost. The former represents the path cost of maintaining or switching temperature zone conditions, while the latter represents the quality risk of agricultural products arising from changes in time and conditions along candidate paths.

[0068] In this embodiment, the recording location of temperature zone matching deviation can be a graph edge or a graph node related to delivery handover. If the deviation originates from a transportation route segment, it is written to a graph edge; if the deviation originates from cold storage outbound or delivery handover waiting, it is written to a graph node and converted into the corresponding temperature control energy consumption cost during path accumulation. When comparing candidate paths, edge computing nodes do not mix temperature control energy consumption cost and quality degradation cost into the same index to avoid incorrectly evaluating paths with small temperature zone deviations but long waiting times. By separating the records, the multi-objective cost vector can retain different sources of constraints in cold chain execution.

[0069] In a preferred embodiment, edge computing nodes can reference both historical lateness probabilities and path stability records when processing path risk costs. Historical lateness probabilities directly reflect the historical occurrence of delivery node delays caused by a directed edge, while path stability records reflect fluctuations in the travel time of that directed edge. If a graph edge has a low historical lateness probability but low stability, it indicates that it may still cause subsequent time window compression under the current road network conditions; if it has a high historical lateness probability and low stability, then that graph edge will make a significant contribution to the path risk cost. The lateness risk records of all graph edges in the candidate path are accumulated to form the path risk cost, which is used to complement the delivery time cost. The delivery time cost reflects the expected time consumption, while the path risk cost reflects the probability of execution fluctuations; both are entered into non-dominated sorting and comprehensive sorting, respectively.

[0070] In this embodiment, the path risk cost setting enables edge computing nodes to distinguish between paths with "short but highly volatile estimated travel times" and paths with "slightly longer but more stable execution times." For agricultural product batches with tight remaining allowable delivery deadlines, the path risk weight can be increased based on the batch state vector, making the overall ranking more biased towards paths with low lateness risk. For batches with wide allowable delivery deadlines and low quality degradation characteristic values, the path risk weight can be kept low, allowing path selection to retain more timeliness and temperature zone factors. This processing allows the path selection rules to change with batch state.

[0071] In one embodiment, edge computing nodes hierarchically label the constraint satisfaction status of candidate paths. Basic constraints include outbound operation constraints, delivery time windows, and agricultural product quality constraints; extended constraints include temperature zone matching deviation, path risk cost, and path stability records. Candidate paths that do not meet basic constraints cannot be used as target delivery paths, but can undergo local repair; candidate paths with higher extended constraints can be retained in non-dominated ranking, but may be ranked lower in comprehensive ranking. Hierarchical labeling avoids premature loss of candidate paths due to treating all constraints equally, and also prevents paths that clearly violate basic constraints from entering the target output.

[0072] In this embodiment, the determination of the target delivery route requires the simultaneous fulfillment of three conditions. First, the cold storage outbound time slice corresponding to the route's starting point satisfies the outbound operation constraints and there are no outbound conflicts that are prohibited from being occupied. Second, the arrival times of the route at each delivery node are within the corresponding delivery time window, or have been restored to meet the time window after partial repair. Third, the cumulative quality degradation cost of the route does not exceed the quality constraint boundary defined by the batch state vector. Only candidate routes that simultaneously meet the above conditions can be determined as the primary delivery route or alternative delivery route after comprehensive sorting. This process ensures that the output route remains consistent with the actual outbound status of the digital cold storage, the delivery node time window, and the quality status of the agricultural products.

[0073] In a preferred embodiment, when outputting the main delivery route, the edge computing node also outputs corresponding constraint source records. These constraint source records include quality degradation feature sources, outbound constraint feature sources, route operation feature sources, and candidate route repair records. Quality degradation feature sources can point to agricultural product category, inbound duration, freshness grade, target temperature zone, and allowable delivery time limit; outbound constraint feature sources can point to the available outbound window set and outbound conflict constraints; and route operation feature sources can point to delivery node time windows, estimated travel time for road segments, historical lateness probability, and route stability records. Through these records, when subsequently reviewing the route results, it is possible to trace each constraint and determine which type of data object it originates from.

[0074] In this embodiment, the constraint source record can also be used for anomaly handling. If a primary delivery route is determined to no longer meet the delivery time window before execution, the edge computing node can determine whether the change is due to a change in the estimated travel time of the road segment, a change in the delivery node time window, or a change in the outbound time slice based on the constraint source record. If the change originates from the estimated travel time of the road segment, the edge computing node prioritizes updating the relevant graph edges and checking alternative delivery routes; if the change originates from the outbound time slice, the edge computing node prioritizes performing an outbound time slice shift; if the change originates from a change in batch quality status, the edge computing node prioritizes recalculating the quality degradation cost. This processing provides a clear data trigger basis for dynamic route adjustments.

[0075] In one embodiment, to ensure sufficient disclosure, the following provides an exemplary description of a complete path optimization process. Assume there are several batches of agricultural products awaiting delivery in a digital cold storage facility. Each batch has a category, storage duration, preservation grade, target temperature zone, allowable delivery time limit, and outbound priority. Edge computing nodes read the current cold storage operation time and convert this batch data into quality degradation features and batch state vectors. Subsequently, they read the cold storage outbound operation data, generating a set of available outbound windows and outbound conflict constraints. For a given batch, if the initial outbound time slice given by the queuing sequence is occupied by an already assigned delivery task, the edge computing node selects an adjacent available time slice based on the candidate outbound time slice matching cost and writes this time slice to the path search starting point. Then, they read the delivery road network data, generate a set of directed edges according to the delivery node succession relationship, and record the expected travel time, historical lateness probability, and path stability in the corresponding directed edges.

[0076] After the data transformation is completed, the edge computing nodes construct a spatiotemporal constraint graph. The graph contains a starting node corresponding to the cold storage outbound time slice, intermediate nodes corresponding to road network path nodes, and an end node corresponding to the delivery node time window. Each graph edge originating from the starting node undergoes a feasibility assessment; only edges that meet the expected transfer time, arrive within the allowed time window, have a target temperature zone belonging to the allowed temperature zone set, and do not have outbound conflicts are considered searchable edges. Candidate path search starts from the starting node that satisfies the outbound constraints and extends along the searchable graph edges to the delivery node time window. For each candidate path formed, the edge computing nodes accumulate waiting time, travel time, quality degradation increment, temperature zone matching deviation, and lateness risk records, forming a multi-objective cost vector.

[0077] After the candidate path set is formed, edge computing nodes perform non-dominated sorting, retaining candidate paths that are not dominated by other paths across multiple cost dimensions. For paths with outbound operation conflicts, edge computing nodes shift the outbound time slice; for paths with delivery time window conflicts, edge computing nodes perform local node replacement or delivery node rearrangement; for paths with agricultural product quality constraint conflicts, edge computing nodes shorten waiting nodes or replace high-risk graph edges. Candidate paths that still do not meet the basic constraints after repair are excluded. The remaining candidate paths generate target weight combinations based on the batch state vector and calculate a comprehensive ranking value. Candidate paths that satisfy outbound operation constraints, delivery time windows, and agricultural product quality constraints in the ranking results are determined as primary delivery paths, and candidate paths that are ranked adjacent and satisfy the same constraints are determined as alternative delivery paths.

[0078] In another embodiment, when multiple batches participate in path calculation simultaneously, the edge computing node does not process each batch completely independently. For multiple batches sharing the same cold storage outbound time slice, the edge computing node compares their quality degradation characteristic values, remaining allowable delivery time limits, and outbound priorities based on the batch state vector. If two batches conflict in the same time slice, the batch with tighter constraints is retained to occupy that time slice, and an alternative outbound time slice is found for the other batch. After the alternative time slice is determined, the edge computing node reconstructs the path search starting point for that batch without changing the path segments of other batches that have already been determined. For multiple batches sharing the same delivery node time window, the edge computing node calculates and rearranges the order based on the delivery node succession relationship and arrival time, and recalculates the multi-objective cost vector for the affected batches. Through inter-batch conflict handling, multiple delivery tasks can share the same spatiotemporal constraint graph for coordinated calculation.

[0079] In this embodiment, the path results of multiple batches can also share the directed edge operation characteristics of the road network. If the estimated travel time or historical lateness probability of a certain road segment changes, the edge computing node updates the road segment operation cost of that directed edge and searches for all candidate paths, primary delivery paths, and alternative delivery paths that contain that directed edge. After recalculating the delivery time cost and path risk cost for the affected paths, the edge computing node determines whether the primary delivery path still meets the constraints; if it still meets the constraints, the current path is retained; if it does not meet the constraints, alternative delivery paths are checked first. This implementation allows dynamic changes in the road network to be propagated to relevant paths at the edge side without causing irrelevant updates to batch paths that do not pass through the road segment.

[0080] In a preferred embodiment, the edge computing node can set an upper limit on the number of candidate paths to save. However, this upper limit is not a hardware parameter but rather part of the data processing rules. When retaining candidate paths, paths from the non-dominant candidate path set are retained first, followed by paths that meet the basic constraints after local repair. If multiple candidate paths with very similar costs exist in the same batch, the edge computing node can filter them based on the path node difference to avoid high overlap between candidate delivery paths and the main delivery path. The path node difference can be obtained by comparing the number of different road network path nodes and different outbound time slices in two paths. This processing allows candidate delivery paths to have alternative space when they change dynamically in the future, rather than just being repeated records of the main delivery path.

[0081] In this embodiment, the path node difference degree is only used for saving candidate paths and not for changing the basic constraint judgment. That is, if a path differs significantly from the main delivery path but does not meet the delivery time window or quality constraints, it cannot be used as a candidate delivery path; if a path meets all basic constraints but is exactly the same as the main delivery path, it does not need to be saved as a candidate delivery path again. The edge computing node establishes a calculation rule between the number of candidate paths to retain and the substitutability of the alternatives through this process, so that there are available candidates for subsequent path switching.

[0082] In one embodiment, the method dynamically updates the status of cold storage outbound operations when the status changes. Status changes can include a new task occupying an outbound time slice, a change in the queuing order of a batch awaiting outbound operations, a change in the scheduling status of a temperature zone, or an update of a delivery node's time window. Upon receiving the changed data, the edge computing node first locates the graph node or edge corresponding to the changed object, and then updates the relevant outbound constraint features or path operation features. If the changed object corresponds to a cold storage outbound time slice, the feasibility of connecting the graph edges to that time slice is recalculated; if the changed object corresponds to a delivery node's time window, the feasibility of candidate paths to that delivery node is reassessed; if the changed object corresponds to a road segment's travel time, the delivery time cost and path risk cost of the relevant path are recalculated. After the update is complete, the edge computing node performs constraint checks and necessary local repairs again.

[0083] In this embodiment, dynamic updates do not alter the basic processing flow of the method. Regardless of whether the data comes from timed refreshes, cold storage operation record updates, or delivery execution feedback, it is converted into quality degradation features, outbound constraint features, or path operation features, and the corresponding position is found in the spatiotemporal constraint graph. The multi-objective cost vector of the candidate path is updated as the graph node or edge data changes, and the target delivery path is re-verified based on the new constraint state. This processing method enables dynamic path optimization to continuously revolve around a unified constraint graph, avoiding inconsistencies in path results caused by processing different data sources separately.

[0084] In a preferred embodiment, after path calculation is completed, the edge computing node performs a consistency check on the path results for each batch. The consistency check includes checking whether the same outbound time slice is occupied by multiple mutually exclusive batches, whether the same delivery node time window is occupied by multiple paths that cannot arrive simultaneously, whether there are multiple primary delivery paths for the same batch, and whether the primary delivery path and alternative delivery paths are all associated with the same batch status version. If a consistency conflict is found, the edge computing node reallocates the outbound time slice or rearranges the delivery node order according to batch priority, and recalculates the multi-objective cost of the relevant candidate paths. If the conflict cannot be resolved through local processing, the corresponding batch is marked as a batch to be recalculated, while the path results of other batches that already meet the constraints are retained.

[0085] In this embodiment, consistency checks are used to ensure that there are no internal contradictions between the outputs of multiple batch paths. A single path satisfying constraints does not necessarily mean the entire delivery plan satisfies constraints, as multiple paths may share cold storage outbound resources or delivery node time windows. By performing consistency checks at the edge computing node side, the path results can maintain correspondence with the overall outbound arrangement of the digital cold storage. This process, combined with batch state vectors, outbound window sets, and delivery node succession relationships, provides an executable data foundation for the dynamic path optimization results.

[0086] In one embodiment, the calculation records of the method can form a path optimization log. The path optimization log includes an input data summary, feature transformation results, spatiotemporal constraint graph version, number of candidate paths, reasons for invalidation, non-dominated sorting results, local repair records, primary delivery path, and alternative delivery paths. The input data summary records the versions of the batch data, cold storage outbound operation data, and delivery network data used in this calculation; the feature transformation results record quality degradation features, outbound constraint features, and path operation features; the reasons for invalidation explain the specific reasons why graph edges do not participate in the search; and the local repair records explain the location of outbound time slice shifts, local node replacements, or delivery node rearrangements. The path optimization log provides data support for subsequent review and anomaly localization.

[0087] In this embodiment, the route optimization log does not participate in the actual calculation of route selection, but its recorded content comes from the actual data during the route optimization process. If the delivery results of a certain batch are found to be inconsistent with expectations, the quality degradation characteristics, outbound constraint characteristics, route operation characteristics, and candidate route cost values ​​corresponding to that batch can be read from the log to determine whether the problem originates from batch data, cold storage operation data, road network data, or the constraint repair process. In this way, the route optimization process of the edge computing node is verifiable, reducing repeated debugging caused by uninterpretable route results.

[0088] In a preferred embodiment, edge computing nodes can collaborate with a cloud data platform, but the critical calculations for dynamic route optimization are still performed at the edge. The cloud data platform can provide category benchmark tables, historical route statistics, or long-term delivery records, which the edge computing nodes read locally to generate the quality degradation characteristics and route operation characteristics required for the current calculation. If cloud data is unavailable, the edge computing nodes can still complete the route calculation based on the most recent local synchronization data and current cold storage operation data. This approach is suitable for scenarios where dynamic route optimization in cold chain logistics requires local response, enabling changes in cold storage outbound status and delivery network status to be processed at the edge.

[0089] In this embodiment, cloud data serves only as a historical reference or rule source and does not replace the construction of real-time constraint graphs on edge computing nodes. The current status of agricultural product batches, cold storage outbound time slices, outbound conflict constraints, and the current time window of delivery nodes are all calculated on the edge side. Through this edge-cloud division of labor, long-term data and current operational data are each placed in appropriate processing positions, avoiding the need to transmit the entire current outbound status back before waiting for path results. This processing aligns with the data processing characteristics of edge computing in the dynamic path optimization of agricultural cold chain logistics.

[0090] In one embodiment, the method can be implemented by a computer program. When the computer program is executed by a processor, it reads agricultural product batch data, digital cold storage outbound operation data, and distribution road network data according to the logic described in this embodiment, and completes feature transformation, spatiotemporal constraint graph construction, multi-objective cost calculation, non-dominated sorting, constraint repair, comprehensive sorting, and primary / backup distribution path output. The data structures in the program may include batch status records, outbound window records, road network directed edge records, graph node records, graph edge records, candidate path records, and target path records. The data structures are associated with each other through batch number, time slice number, node number, and path number. The focus of this computer program implementation is on the correspondence between data structures and computational logic, rather than specific hardware structures.

[0091] In this embodiment, the batch status record stores the agricultural product category, storage duration, freshness grade, target temperature zone, allowable delivery time limit, outbound priority, and quality degradation characteristics; the outbound window record stores the outbound time slice, availability status, temperature zone status, occupied task, and conflict flag; the road network directed edge record stores the origin, destination, estimated travel time, historical lateness probability, and path stability records; the graph node record stores the node type, time window, and temperature zone set; the graph edge record stores the feasibility flag, estimated transfer time, and invalidation reason; the candidate path record stores the graph node sequence, graph edge sequence, and multi-objective cost vector; and the target path record stores the main delivery path, alternative delivery paths, and comprehensive ranking value. Through the above data structure, the program can fully support the calculation process of this method.

[0092] In a preferred embodiment, if there are multiple delivery nodes for the same batch of agricultural products, the edge computing node can generate a separate time window constraint for each delivery node and embed it into the same spatiotemporal constraint graph. Candidate paths need to pass through these delivery nodes sequentially or in an allowed order. Upon reaching each delivery node, a time window determination and quality degradation cost update are performed. If an intermediate delivery node causes subsequent nodes to become unreachable within their time windows, the edge computing node will stop branching at that node and record the time window propagation conflict. If the order of delivery nodes can be adjusted, the time window propagation conflict is attempted to be resolved by rearranging the delivery nodes. This process is suitable for situations where the same cold storage batch needs to be distributed to multiple delivery nodes, enabling path calculation to reflect the cumulative impact of continuous multi-node delivery on quality degradation and time windows.

[0093] In this embodiment, the quality degradation cost in multi-node delivery is not calculated only at the final node, but is continuously accumulated as the product passes through each graph edge and waiting node. This processing reflects the complete cold chain path status of agricultural products from warehouse to the last delivery node. If the waiting time at the previous delivery node is too long, subsequent nodes may incur a high quality degradation cost even if the travel distance is short; if the risk of delay is high for a certain segment, the delivery time windows of multiple subsequent nodes may be affected. Edge computing nodes record these transmission relationships through multi-objective cost vectors, enabling candidate path selection to cover the actual constraints of multi-node delivery.

[0094] In one embodiment, edge computing nodes can also perform scale unification on the cost vector during comprehensive ranking. Since delivery timeliness cost, quality degradation cost, temperature control energy consumption cost, and route risk cost have different dimensions, each cost component needs to be converted to a comparable range before direct weighting. Scale unification can be based on the minimum and maximum values ​​in the current candidate path set, or on a reference range in the category benchmark table. The scale-unified cost components retain their original physical meaning; they are simply used for comprehensive ranking calculations. This process prevents a cost component with a larger dimension from having an unreasonable advantage in the comprehensive ranking.

[0095] In this embodiment, scale unification is only used for comprehensive ranking and does not affect non-dominated ranking or basic constraint judgment. Non-dominated ranking directly uses the original cost components for comparison, and basic constraint judgment is directly based on outbound operation constraints, delivery time windows, and agricultural product quality constraints. The use of scale-unified cost components in the comprehensive ranking stage ensures that the target weight combination and different cost dimensions maintain a correspondence. For the same batch, candidate paths with lower comprehensive ranking values ​​have higher priority under the current weight combination; for different batches, since the target weight combinations are different, the comprehensive ranking value is only used for ranking within the same batch and is not directly compared across batches.

[0096] In a preferred embodiment, edge computing nodes set batch-specific quality constraint boundaries for agricultural products. These boundaries can be generated jointly from a category benchmark table, freshness grade, and allowable delivery time limit. For batches with higher freshness grades and wider remaining allowable delivery time limits, the quality constraint boundaries can be relatively loose; for batches with longer storage durations and shorter remaining allowable delivery time limits, the quality constraint boundaries are tighter. The quality degradation cost of a candidate path is compared with the corresponding batch's quality constraint boundary to determine whether the path should be included in the target delivery path candidate. This process avoids misjudgments caused by using a uniform quality boundary for all agricultural product batches.

[0097] In this embodiment, there is a correspondence between quality constraint boundaries and quality decay feature values. Quality decay feature values ​​reflect the current batch's state pressure, while quality constraint boundaries determine whether candidate paths still meet the batch's delivery requirements. Edge computing nodes can use quality constraint boundaries for branch pruning during candidate path search. After non-dominated sorting and comprehensive sorting, the quality constraint boundaries must be verified again to ensure that the locally repaired paths do not generate new quality constraint conflicts due to waiting time, graph edge replacement, or delivery node rearrangement. Through these two verifications, the quality status of candidate paths is constrained in both the search and output stages.

[0098] In one embodiment, the complete technical process of the method can be summarized as follows: edge computing nodes read three types of data and complete verification; agricultural product batch data are converted into quality decay features and batch state vectors; cold storage outbound operation data are converted into a set of available outbound windows and outbound conflict constraints; delivery network data are converted into a set of directed edges of the network and path operation features; a spatiotemporal constraint graph containing cold storage outbound time slices, network path nodes, and delivery node time windows is constructed; feasibility judgment and invalidation marking are performed on graph edges; candidate paths are searched and multi-objective cost vectors are accumulated; non-dominated sorting and local constraint repair are performed; target weight combinations are generated based on batch state vectors; and the main delivery path and alternative delivery paths are output. Each processing step in the above process revolves around edge computing, dynamic path optimization of agricultural product cold chain logistics, and digital cold storage operation constraints.

[0099] In this embodiment, the technical focus of the method is not on adding new cold storage hardware structures, nor on relying on specific vehicles or equipment, but on converting existing batch data, outbound operation data, and delivery network data of the digital cold storage into a constraint graph and cost vector that can be uniformly calculated. Through this processing, dynamic path optimization can be completed in a local computing environment, and the output results simultaneously satisfy cold storage outbound operation constraints, delivery time windows, and agricultural product quality constraints. The batch state vector, available outbound window set, outbound conflict constraints, road network directed edge set, spatiotemporal constraint graph, multi-objective cost vector, non-dominated sorting, local repair, and primary / backup path generation described in various embodiments constitute an interconnected data processing chain, which can cover the main computational links of dynamic path optimization for agricultural product cold chain logistics.

Claims

1. A multi-objective path dynamic optimization method for digital cold storage based on edge computing, characterized in that, include: To obtain batch data of agricultural products to be delivered, outbound operation data of digital cold storage, and delivery road network data required for dynamic route optimization of cold chain logistics for agricultural products; On the edge computing node side, the batch data of agricultural products to be delivered is converted into quality degradation features, the outbound operation data of the digital cold storage is converted into outbound constraint features, and the delivery road network data is converted into path operation features. A spatiotemporal constraint graph is constructed based on the quality degradation characteristics, the outbound constraint characteristics, and the path operation characteristics. The calculation of candidate paths in the spatiotemporal constraint graph includes a multi-objective cost factor encompassing delivery time cost, quality degradation cost, temperature control energy consumption cost, and path risk cost. The method for selecting target delivery paths that meet the constraints of delivery time window, outbound operation, and agricultural product quality based on the multi-objective cost value.

2. The multi-objective path dynamic optimization method for digital cold storage based on edge computing according to claim 1, characterized in that, Converting the batch data of agricultural products to be delivered into quality degradation characteristics includes: extracting agricultural product category, storage duration, freshness grade, target temperature zone, allowable delivery time limit and outbound priority from the batch data of agricultural products to be delivered; The warehousing duration and the allowable delivery time limit are normalized according to the quality sensitivity range corresponding to the agricultural product category. Map the freshness rating, the target temperature zone, and the outbound priority to batch constraint codes; Based on the normalization result and the batch constraint encoding, a batch state vector is generated for participating in the quality degradation cost calculation.

3. The multi-objective path dynamic optimization method for digital cold storage based on edge computing according to claim 2, characterized in that, Converting the digital cold storage outbound operation data into outbound constraint features includes: extracting the current cold storage capacity occupancy status, outbound time slice, waiting batch queue sequence, temperature zone scheduling status, and assigned delivery tasks from the digital cold storage outbound operation data. A set of available outbound windows is generated based on the outbound time slice and the queue sequence of the batches to be outbound. Generate outbound conflict constraints based on the temperature zone scheduling status and the allocated delivery tasks; Associate the set of available outbound windows and the outbound conflict constraints with the corresponding batches of agricultural products to be delivered.

4. The multi-objective path dynamic optimization method for digital cold storage based on edge computing according to claim 1, characterized in that, The delivery network data is converted into path operation characteristics, including: extracting delivery node time windows, estimated travel time of road segments, historical lateness probability, path stability records and delivery node succession relationships from the delivery network data; A set of directed edges for the road network is formed based on the connection relationship between the delivery nodes; Write the estimated travel time of the road segment and the historical lateness probability into the corresponding edges in the directed edge set of the road network; The delivery node time window is associated with the path stability record and then linked to the corresponding delivery node to form path operation characteristics for path feasibility determination.

5. The multi-objective path dynamic optimization method for digital cold storage based on edge computing according to claim 3, characterized in that, The construction of the spatiotemporal constraint graph includes: using the cold storage outbound time slice, the delivery node time window, and the road network path nodes as graph nodes, and using the feasible transfer relationship of agricultural product batches from the cold storage outbound time slice to the delivery node time window via the road network path nodes as graph edges; The quality degradation feature, the outbound constraint feature, and the path operation feature are respectively attached to the corresponding graph nodes or graph edges; Graph edges that do not meet the target temperature zone, outbound time slice, or delivery time window are marked as invalid, while graph edges that meet the constraints are retained for candidate path search.

6. The multi-objective path dynamic optimization method for digital cold storage based on edge computing according to claim 5, characterized in that, Based on the normalization processing result and the batch constraint encoding, a batch state vector is generated, including: assigning feature weights to the warehousing duration, the allowable delivery time limit, the freshness grade, and the target temperature zone according to the preset quality sensitivity level of the agricultural product category; The normalization result and the batch constraint code are weighted and combined according to the feature weights; The weighted combination results of different batches of agricultural products to be delivered are sorted among batches to obtain a batch state vector carrying the batch priority relationship.

7. The multi-objective path dynamic optimization method for digital cold storage based on edge computing according to claim 6, characterized in that, Associating the set of available outbound windows and the outbound conflict constraints with the corresponding batches of agricultural products to be delivered includes: determining the initial outbound time slice for each batch of agricultural products to be delivered based on the queuing sequence of the batches to be delivered. When the initial outbound time slice does not match the set of available outbound windows, an alternative outbound time slice that satisfies the constraint relationship with the allowed delivery time limit is selected from the set of available outbound windows. The alternative outbound time slice and the outbound conflict constraint are written together into the path search starting point constraint of the corresponding batch.

8. The multi-objective path dynamic optimization method for digital cold storage based on edge computing according to claim 6, characterized in that, Calculating the multi-objective cost value for candidate paths in the spatiotemporal constraint graph includes: searching for candidate paths from the cold storage outbound time slice to the delivery node time window in the graph edges that satisfy the constraints; For each candidate path, the waiting time of graph nodes, the travel time of graph edges, the quality degradation increment, the temperature zone matching deviation, and the lateness risk record are accumulated. The waiting time of the graph nodes and the travel time of the graph edges are used to form the delivery time cost, the quality decay increment is used to form the quality decay cost, the temperature zone matching deviation is used to form the temperature control energy consumption cost, and the lateness risk record is used to form the path risk cost.

9. The multi-objective path dynamic optimization method for digital cold storage based on edge computing according to claim 8, characterized in that, The method for selecting target delivery routes based on the multi-objective cost-value includes: sorting the candidate routes in a non-dominated manner according to the delivery time cost, the quality degradation cost, the temperature control energy consumption cost, and the route risk cost, to obtain a set of non-dominated candidate routes; In the set of non-dominated candidate paths, local node replacement, outbound time slice shifting, or delivery node rearrangement are performed on candidate paths that have conflicts in outbound operations, delivery time windows, or agricultural product quality constraints. Candidate routes that still meet the constraints after conflict resolution will be selected as target delivery routes.

10. The multi-objective path dynamic optimization method for digital cold storage based on edge computing according to claim 9, characterized in that, The step of using candidate paths that still satisfy the constraints after conflict resolution as target delivery paths includes: determining target weight combinations for different agricultural product batches to be delivered based on the batch state vector carrying batch priority relationships among the candidate paths that still satisfy the constraints after conflict resolution. The delivery time cost, quality degradation cost, temperature control energy consumption cost, and route risk cost are comprehensively ranked according to the target weight combination. Candidate paths that satisfy the outbound operation constraints, the delivery time window, and the agricultural product quality constraints from the comprehensive ranking results are determined as primary delivery paths, and candidate paths that are ranked adjacently and satisfy the constraints are determined as alternative delivery paths.

Images