A logistics path planning method and system based on big data processing

Abstract

The application discloses a logistics path planning method and system based on big data processing and relates to the field of logistics scheduling, and comprises the following steps: acquiring agricultural product cold chain distribution dataset; constructing a cold chain distribution performance graph; performing vectorization coding on the graph node attributes and edge relationship attributes of the cold chain distribution performance graph through an input coding layer of an improved GraphGPS model; performing freshness time sequence enhancement processing on the cold chain distribution graph representation data through a freshness time sequence coding layer; performing graph update processing on the freshness time sequence graph representation data through a multi-layer GPS update layer, and fusing the influence of door opening disturbance and the influence of performance chain; identifying freshness risk nodes and determining the forward and backward movement relationship of distribution nodes to generate a logistics path planning result. The application uses big data processing to construct a cold chain performance graph, deduces the freshness life of agricultural products and optimizes the distribution path, and has the advantages of controllable freshness, predictable risk and accurate scheduling.

Description

Technical Field

[0001] This invention relates to the field of logistics scheduling, and in particular to a logistics route planning method and system based on big data processing. Background Technology

[0002] Existing logistics route planning technologies typically plan the sequence of delivery nodes and vehicle routes based on delivery orders, vehicle capacity, road conditions, customer time windows, and estimated travel time. The main goals are to shorten delivery distances, reduce transportation time, and improve vehicle turnaround efficiency. With the increasing demand for cold chain distribution of agricultural products, some methods are beginning to incorporate temperature and humidity monitoring and cold chain status recording to determine whether the environment during transportation meets preservation requirements.

[0003] However, existing methods mostly use the cold chain environment as the basis for verifying the status after transportation. They lack prior calculation of the freshness consumption relationship between picking time, loading time, expected arrival time and acceptance time window. They also fail to reflect the continuous impact of door opening, temperature and humidity recovery, proximity of loading location, waiting at previous nodes and unloading delays on subsequent delivery nodes. As a result, the route planning results may only meet the vehicle arrival requirements, but cannot guarantee that the agricultural products meet the freshness acceptance requirements upon arrival.

[0004] Therefore, how to provide a logistics route planning method and system based on big data processing is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a logistics route planning method and system based on big data processing. This invention utilizes big data processing to construct a cold chain fulfillment map, extrapolate the freshness lifespan of agricultural products, and optimize delivery routes, possessing the advantages of controllable freshness, predictable risks, and precise scheduling.

[0006] A logistics route planning method based on big data processing according to an embodiment of the present invention includes the following steps:

[0007] Acquire data on agricultural products to be delivered, cold chain environment data, vehicle loading data, delivery node data, road traffic data, and customer acceptance data, and perform standardized processing and object association to generate an agricultural product cold chain delivery dataset;

[0008] A cold chain delivery fulfillment map was constructed based on the agricultural product cold chain delivery dataset;

[0009] The cold chain delivery fulfillment map is input into the improved GraphGPS model. The graph node attributes and edge relationship attributes are vectorized and encoded using the input encoding layer to generate cold chain delivery graph representation data. The improved GraphGPS model includes an input encoding layer, a freshness time-series encoding layer, a multi-layer GPS update layer, and an output layer.

[0010] The freshness temporal coding layer is used to perform freshness temporal enhancement processing on the cold chain distribution map representation data to generate freshness temporal map representation data;

[0011] The freshness time series graph representation data is updated by using a multi-layer GPS update layer, and the effects of opening disturbance and fulfillment chain are integrated to generate freshness remaining life projection data.

[0012] Based on the remaining freshness life projection data, identify freshness risk nodes that are expected to fail to meet freshness acceptance requirements upon arrival. Then, determine the forward and backward movement relationships of delivery nodes based on freshness risk nodes and road traffic data, and generate logistics route planning results.

[0013] Optionally, the standardization process includes time unification, unit unification, status encoding, missing data completion, and anomaly marking; the object association includes identifier matching, location mapping, and node association.

[0014] Optionally, the construction of the cold chain delivery fulfillment map specifically includes:

[0015] Based on the agricultural products to be delivered and their loading locations in the cold chain agricultural product distribution dataset, establish nodes for agricultural products to be delivered, nodes for the area of ​​the vehicle, and the loading relationship between the two.

[0016] Based on the delivery order, estimated unloading time, and node waiting status in the delivery node data, establish delivery nodes and establish a fulfillment priority relationship between adjacent delivery nodes;

[0017] Based on the road segments, road traffic conditions, and estimated travel time between delivery nodes, road segment nodes are established, and road segment nodes are connected to associated delivery nodes to form traffic relationships;

[0018] Based on the acceptance time window and freshness acceptance requirements of the agricultural products to be delivered at the delivery nodes, an acceptance constraint relationship is established between the agricultural product nodes to be delivered and the delivery nodes. A graph structure is then arranged for the agricultural product nodes to be delivered, the vehicle area nodes, the delivery nodes, the road segment nodes, the loading relationship, the order of performance, the passage relationship, and the acceptance constraint relationship to generate a cold chain delivery performance diagram.

[0019] Optionally, the generation of the cold chain distribution map representation data specifically includes:

[0020] The cold chain delivery fulfillment map is input into the improved GraphGPS model, and the node attributes of each graph are extracted by the input encoding layer according to the node type and a node attribute encoding queue is established.

[0021] The improvements of the improved GraphGPS model compared to the existing GraphGPS model are as follows: the original location structure encoding used to represent the topological location of nodes is replaced with freshness time sequence encoding oriented towards picking time, loading time, expected arrival time and acceptance time window; the original local message passing used for information transmission between adjacent nodes is replaced with door opening disturbance branch oriented towards the influence of door opening and closing, temperature and humidity changes and proximity of loading location; and the original global attention used for modeling the association of nodes in the whole graph is replaced with fulfillment chain attention branch oriented towards the influence of delivery sequence, node waiting and unloading connection.

[0022] The category field, time field, status field, and numerical field in the node attribute encoding queue are vectorized and mapped separately to generate the initial representation data of the graph nodes;

[0023] Extract the start and end nodes, relationship types, relationship connection order, and attribute fields of loading relationships, performance sequence relationships, passage relationships, and acceptance constraint relationships. Convert them into edge relationship vectors and align them with the initial representation data of the graph nodes to generate cold chain distribution graph representation data.

[0024] Optionally, the generation of the freshness time series representation data specifically includes:

[0025] Input the cold chain distribution map representation data into the freshness time-series coding layer, extract the picking time, loading time, estimated arrival time and acceptance time window, and encode the time position of each time field to generate freshness time-series coded data;

[0026] Based on the chronological relationship between harvest time, loading time and expected arrival time, the freshness consumption stage of the agricultural products to be delivered during the cold chain distribution process is coded to generate freshness stage coding data.

[0027] Based on the loading time, estimated arrival time, and the end time of the acceptance time window, the delivered time and the remaining deliverable time are encoded to generate time constraint coding data.

[0028] Based on the positional relationship between the expected arrival time and the start and end times of the acceptance time window, the status of the acceptance time window matching is encoded to generate acceptance matching code data.

[0029] Freshness time-series coding data, freshness stage coding data, duration constraint coding data, and acceptance matching coding data are overlaid onto the cold chain distribution map representation data to generate freshness time-series map representation data.

[0030] Optionally, the generation of the freshness remaining life projection data specifically includes:

[0031] The freshness time series graph representation data is input into the multi-layer GPS update layer to extract cold chain environment data, cargo compartment loading data and estimated unloading time, and door opening disturbance input data is generated by aggregating the door opening and closing time and loading location.

[0032] The door opening disturbance branch is used to transmit the door opening disturbance input data according to the loading relationship, the car area adjacency relationship and the loading position adjacency relationship, and generate the door opening disturbance local update data;

[0033] By utilizing the fulfillment chain attention branches along the fulfillment sequence, the order of fulfillment impact between delivery nodes is organized, generating full-path update data for the fulfillment chain;

[0034] The local update data of the door opening disturbance and the full path update data of the performance chain are input into the feedforward update unit to perform layer-by-layer fusion and update of the freshness time series graph representation data, and generate freshness update graph representation data.

[0035] Based on the freshness update map representation data, the remaining freshness lifespan of the agricultural products to be delivered at each delivery node is extrapolated to generate node freshness lifespan data.

[0036] Based on the node freshness lifetime data and freshness acceptance requirements, acceptance satisfaction is determined, and the determination results are converted into freshness risk status and freshness deliverable status, and then freshness remaining lifetime projection data is generated. The freshness risk status includes freshness normal status, freshness critical delivery status and freshness non-compliance status; the freshness deliverable status includes deliverable status, critical deliverable status and non-deliverable status.

[0037] Optionally, the generation of the logistics route planning results specifically includes:

[0038] Based on the freshness remaining life projection data and acceptance constraints, the nodes of agricultural products to be delivered that are expected to not meet the freshness acceptance requirements upon arrival, as well as their associated delivery nodes, are screened to generate freshness risk node data.

[0039] The life gap value of freshness risk nodes is calculated based on the freshness risk node data, and the adjustment priority of freshness risk nodes is arranged in combination with the acceptance time window to generate risk node adjustment order data.

[0040] Based on the risk node adjustment order data, the forward position is retrieved between delivery nodes, and the estimated arrival time after the forward movement is verified in combination with road traffic conditions and estimated travel time to generate the forward movement relationship of delivery nodes;

[0041] Based on the forward movement relationship of delivery nodes, the associated delivery nodes are inserted into the candidate forward movement positions, and the crossed delivery nodes are extended in the original relative order. The passage relationship and freshness delivery status are re-verified, and the logistics route planning result is generated. The logistics route planning result includes the adjusted delivery node order, associated road segments, adjusted estimated arrival time, and adjusted freshness delivery status.

[0042] A logistics route planning system based on big data processing according to an embodiment of the present invention includes:

[0043] The dataset generation module is used to acquire data on agricultural products to be delivered, cold chain environment data, vehicle loading data, delivery node data, road traffic data, and customer acceptance data, and to perform standardized processing and object association to generate agricultural product cold chain delivery datasets.

[0044] The fulfillment graph construction module is used to construct a cold chain delivery fulfillment graph based on the agricultural product cold chain delivery dataset.

[0045] The graph representation encoding module is used to input the cold chain delivery fulfillment graph into the improved GraphGPS model, and use the input encoding layer to vectorize the graph node attributes and edge relationship attributes to generate cold chain delivery graph representation data.

[0046] The freshness temporal enhancement module is used to perform freshness temporal enhancement processing on the cold chain distribution map representation data using the freshness temporal coding layer, and generate freshness temporal map representation data;

[0047] The freshness lifetime projection module is used to perform graph update processing on the freshness time series graph representation data using multiple GPS update layers, and to integrate the effects of door opening disturbance and fulfillment chain to generate freshness remaining lifetime projection data.

[0048] The route planning generation module is used to identify freshness risk nodes based on freshness remaining life projection data, and determine the forward and backward movement relationship of delivery nodes based on freshness risk nodes and road traffic data, thereby generating logistics route planning results.

[0049] The beneficial effects of this invention are:

[0050] This invention acquires and processes data on agricultural products awaiting delivery, cold chain environment, vehicle loading, delivery nodes, road traffic, and customer acceptance, and then standardizes and associates these data. This allows for the unified organization of data previously scattered across orders, vehicles, truck compartments, roads, and customer acceptance stages into a single agricultural product cold chain delivery dataset. Further, by constructing a cold chain delivery fulfillment map, the loading relationships, fulfillment sequence relationships, traffic relationships, and acceptance constraints between agricultural products awaiting delivery, truck compartment areas, delivery nodes, and road segments can be uniformly expressed. This enables route planning to go beyond simply considering vehicle travel distance and arrival order; it simultaneously considers the location of agricultural products within the truck compartment, the fulfillment connections at delivery nodes, and the impact of the customer acceptance time window on freshness delivery.

[0051] This invention applies an improved GraphGPS model to cold chain delivery fulfillment graph processing. It replaces the original location structure encoding with freshness temporal encoding, replaces local message passing with door-opening perturbation branches, and replaces global attention with fulfillment chain attention branches. This allows the model to extrapolate the freshness consumption process around harvesting time, loading time, estimated arrival time, and acceptance time windows. Furthermore, it incorporates factors such as door opening, temperature and humidity changes, proximity of loading locations, waiting at preceding nodes, and unloading delays into the graph update process. This enables more accurate generation of remaining freshness life projection data and early identification of freshness risk nodes that do not meet freshness acceptance requirements upon estimated arrival, avoiding the problem in traditional route planning where "vehicles can reach the destination, but agricultural products cannot be accepted."

[0052] This invention arranges delivery nodes forward and backward based on freshness risk nodes and road traffic data. It proactively adjusts the order of delivery nodes prone to freshness issues during the route planning stage and re-verifies the adjusted estimated arrival time, road traffic conditions, and freshness delivery status. Through this process, the logistics route planning results not only meet road traffic and delivery fulfillment requirements but also take into account the remaining shelf life of agricultural products and customer acceptance requirements, improving the precision of cold chain delivery route planning, the ability to predict freshness risks, and the reliability of actual delivery. Attached Figure Description

[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0054] Figure 1 This is a flowchart of a logistics route planning method based on big data processing proposed in this invention;

[0055] Figure 2 This is a comparison chart of the freshness lifespan of delivery nodes in a logistics route planning method based on big data processing proposed in this invention. Detailed Implementation

[0056] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0057] refer to Figures 1-2 A logistics route planning method based on big data processing includes the following steps:

[0058] Acquire data on agricultural products to be delivered, cold chain environment data, vehicle loading data, delivery node data, road traffic data, and customer acceptance data, and perform standardized processing and object association to generate an agricultural product cold chain delivery dataset;

[0059] A cold chain delivery fulfillment map was constructed based on the agricultural product cold chain delivery dataset;

[0060] The cold chain delivery fulfillment map is input into the improved GraphGPS model. The graph node attributes and edge relationship attributes are vectorized and encoded using the input encoding layer to generate cold chain delivery graph representation data. The improved GraphGPS model includes an input encoding layer, a freshness time-series encoding layer, a multi-layer GPS update layer, and an output layer.

[0061] The freshness temporal coding layer is used to perform freshness temporal enhancement processing on the cold chain distribution map representation data to generate freshness temporal map representation data;

[0062] The freshness time series graph representation data is updated by using a multi-layer GPS update layer, and the effects of opening disturbance and fulfillment chain are integrated to generate freshness remaining life projection data.

[0063] Based on the remaining freshness life projection data, identify freshness risk nodes that are expected to fail to meet freshness acceptance requirements upon arrival. Then, determine the forward and backward movement relationships of delivery nodes based on freshness risk nodes and road traffic data, and generate logistics route planning results.

[0064] In this embodiment, the data for agricultural products to be delivered includes agricultural product identification, agricultural product category, harvesting time, loading time, and initial freshness status; cold chain environment data includes compartment temperature, compartment humidity, external ambient temperature, door opening time, door closing time, and temperature and humidity recovery status; compartment loading data includes compartment area and loading location; delivery node data includes delivery node identification, delivery sequence, estimated arrival time, estimated unloading time, and node waiting status; road traffic data includes road segment, road traffic status, and estimated travel time; and customer acceptance data includes acceptance time window and freshness acceptance requirements.

[0065] Standardized processing includes time unification, unit unification, status coding, missing data completion, and anomaly marking. Time unification involves standardizing picking time, loading time, door opening time, door closing time, estimated arrival time, and acceptance time window to a year-month-day-hour-minute-second format, converting the cumulative seconds to a base start time of 0 seconds. The time merging cycle is 60 seconds; data with a time deviation of no more than 30 seconds is included in the current cycle, data with a time deviation greater than 30 seconds but no more than 120 seconds is included in the adjacent cycle, and data with a time deviation greater than 120 seconds is marked as abnormal time data.

[0066] Unit standardization includes standardizing the temperature of the cargo compartment and the external ambient temperature to degrees Celsius, the humidity of the cargo compartment to a relative humidity percentage, and the estimated unloading time and estimated travel time to seconds; temperature data is retained to one decimal place, humidity data is retained to an integer, and time data is retained to an integer.

[0067] The status coding uses a 4-level field: 0 represents a missing or unknown status, 1 represents a normal status, 2 represents a status of concern, and 3 represents an abnormal status. In road traffic status, smooth traffic is coded as 1, slow traffic as 2, congestion as 3, and impassable traffic is marked as impassable. In node waiting status, no waiting is coded as 1, short-term waiting as 2, and long-term waiting as 3. In temperature and humidity recovery status, recovered as coded as 1, recovering as 2, and not recovered as 3. In initial freshness status, sufficient freshness is coded as 1, critical freshness as 2, and insufficient freshness as 3.

[0068] Missing data completion includes determining consecutive missing values ​​for numeric fields. If consecutive missing values ​​are less than 3 merging periods, linear completion is performed using the preceding and following valid data. If consecutive missing values ​​are greater than 3 but less than 10 merging periods, the average of the 10 most recent valid data for the same field is used for completion. If consecutive missing values ​​are greater than 10 merging periods, the missing value marker is retained. For status fields, if consecutive missing values ​​are less than 3 merging periods, the previous valid status code is used. If consecutive missing values ​​are greater than 3 merging periods, the unknown status code is retained.

[0069] Anomaly marking includes marking data whose end time is earlier than the start time as abnormal time data, data whose duration is less than 0 as abnormal duration data, data whose values ​​are the same for three consecutive merging periods of the same numeric field as data to be reviewed, and data whose current value deviates from the average of the last 20 valid data of the same field by more than 3 times the standard deviation as abnormal numerical data.

[0070] Object association includes identifier matching, location mapping, and node association; specifically: identifier matching is performed based on agricultural product identifiers, delivery node identifiers, and road segments, merging data under the same identifier into the same object record; location mapping is performed according to the compartment area and loading position to determine the location of the agricultural products to be delivered within the compartment; and node association is performed according to the delivery sequence, estimated arrival time, estimated travel time, acceptance time window, and freshness acceptance requirements for the data of agricultural products to be delivered, delivery node data, road traffic data, and customer acceptance data.

[0071] In this embodiment, the construction of the cold chain delivery fulfillment map specifically includes:

[0072] Based on the agricultural products to be delivered and the loading location of the vehicle compartment in the agricultural cold chain distribution dataset, establish the nodes of agricultural products to be delivered, the nodes of the vehicle compartment area, and the loading relationship between the two; the loading relationship uses the nodes of agricultural products to be delivered and the nodes of the vehicle compartment area as associated objects, and writes the loading location into the loading relationship attribute;

[0073] Based on the delivery order, estimated unloading time, and node waiting status in the delivery node data, delivery nodes are established, and a fulfillment priority relationship is established between adjacent delivery nodes. The fulfillment priority relationship is formed according to the delivery order and is used to characterize the fulfillment connection between the preceding delivery node and the subsequent delivery node.

[0074] Based on the road segments, road traffic conditions, and estimated travel time between delivery nodes, road segment nodes are established, and road segment nodes are connected to associated delivery nodes to form traffic relationships; traffic relationships are used to characterize the vehicle travel connections formed between delivery nodes via road segments;

[0075] Based on the acceptance time window and freshness acceptance requirements of the agricultural products to be delivered at the delivery nodes, an acceptance constraint relationship is established between the agricultural product nodes to be delivered and the delivery nodes. A graph structure is arranged for the agricultural product nodes to be delivered, the vehicle area nodes, the delivery nodes, the road section nodes, the loading relationship, the order of performance, the passage relationship and the acceptance constraint relationship to generate a cold chain delivery performance diagram.

[0076] The diagram structure arrangement includes connecting the agricultural product nodes to be delivered and the vehicle area nodes according to loading relationships, connecting adjacent delivery nodes according to the order of fulfillment, connecting delivery nodes and road segment nodes according to traffic relationships, connecting agricultural product nodes to be delivered and delivery nodes according to acceptance constraints, and merging duplicate nodes and duplicate relationships.

[0077] In this embodiment, the generation of cold chain distribution map representation data specifically includes:

[0078] The cold chain delivery fulfillment map is input into the improved GraphGPS model, and the node attributes of each graph are extracted by the input encoding layer according to the node type and a node attribute encoding queue is established.

[0079] The improvements of the improved GraphGPS model compared to the existing GraphGPS model are as follows: the original location structure encoding used to represent the topological location of nodes is replaced with freshness time sequence encoding oriented towards picking time, loading time, expected arrival time and acceptance time window; the original local message passing used for information transmission between adjacent nodes is replaced with door opening disturbance branch oriented towards the influence of door opening and closing, temperature and humidity changes and proximity of loading location; and the original global attention used for modeling the association of nodes in the whole graph is replaced with fulfillment chain attention branch oriented towards the influence of delivery sequence, node waiting and unloading connection.

[0080] The improved GraphGPS model consists of an input encoding layer, a freshness temporal encoding layer, a multi-layer GPS update layer, and an output layer connected sequentially. Within the multi-layer GPS update layer, the door opening perturbation branch and the fulfillment chain attention branch are connected in parallel to the feedforward update unit. During model training, the weighted sum of the freshness remaining lifetime error, freshness deliverable state error, and path constraint error is used as the total loss function. The number of training epochs is set to 100, the number of training samples per batch is set to 16, the initial learning rate is set to 0.001, and the weight decay coefficient is set to 0.0001. Training is stopped when the total loss decreases by less than 0.001 for 10 consecutive epochs, or when the freshness deliverable state recognition accuracy on the validation set no longer improves for 5 consecutive epochs, or when the number of training epochs reaches 100.

[0081] The node attribute coding queue is formed by aggregating the attributes of agricultural products to be delivered, the attributes of the vehicle compartment area, the attributes of the delivery node, and the attributes of the road segment. A single node attribute record is set to 32 fields, of which the node identifier, node type, and graph connection position each occupy 1 field, and the remaining 29 fields are written to the corresponding graph node attributes. Fields that have not obtained attribute values ​​are written with 0.

[0082] The category field, time field, status field, and numerical field in the node attribute encoding queue are vectorized and mapped separately to generate the initial representation data of the graph nodes;

[0083] The category field is discretely mapped according to agricultural product category, carriage area, loading location, and road segment, with an upper limit of 256 category numbers. Unregistered categories are written as 0, and a 32-dimensional category representation is output. The time field is mapped according to the cumulative number of seconds, with an upper limit of 259200 cumulative seconds. If the number of seconds exceeds this limit, it is written as 259200, and a 32-dimensional time representation is output. The status field is output as a 16-dimensional status representation according to the status code. The numerical field is mapped to the numerical range of 0 to 1 after unifying the units, and a 32-dimensional numerical representation is output. Then, it is concatenated with the 8-dimensional node type representation and the 8-dimensional graph connection position representation to generate a 128-dimensional graph node initial representation vector.

[0084] Extract the start and end nodes, relationship types, relationship connection order, and attribute fields of loading relationships, performance order relationships, passage relationships, and acceptance constraint relationships. Convert them into edge relationship vectors and align them with the initial representation data of graph nodes to generate cold chain distribution graph representation data.

[0085] The edge relationship vector is jointly encoded by the start and end nodes of the relationship, the relationship type, the relationship connection order, and the relationship attribute fields. The relationship attribute fields are written with corresponding attributes according to the relationship type. Among them, the loading relationship is written with the loading location, the delivery order relationship is written with the delivery order and the estimated unloading time, the passage relationship is written with the road passage status and the estimated travel time, and the acceptance constraint relationship is written with the acceptance time window and the freshness acceptance requirements. The relationship attribute fields output a 32-dimensional relationship attribute representation, which is concatenated with the 8-dimensional start node representation, 8-dimensional end node representation, 8-dimensional relationship type representation, and 8-dimensional relationship connection order representation to form a 64-dimensional edge relationship vector.

[0086] The alignment and arrangement are configured by placing the edge relationship vectors between the corresponding graph nodes according to the node connection relationship in the cold chain delivery fulfillment graph.

[0087] In this embodiment, the generation of freshness time series graph representation data specifically includes:

[0088] Input the cold chain distribution map representation data into the freshness time-series coding layer, extract the picking time, loading time, estimated arrival time and acceptance time window, and encode the time position of each time field to generate freshness time-series coded data;

[0089] The time location coding includes encoding the harvest time as the start point of freshness consumption, the loading time as the start point of cold chain distribution, the estimated arrival time as the fulfillment arrival time, the start time of the acceptance time window as the acceptance start point, and the end time of the acceptance time window as the acceptance end point, and writing them sequentially as 1, 2, 3, 4, 5; forming a time point sequence code according to the above time points in the cold chain distribution process, forming a duration interval code according to the cumulative second difference between adjacent time points, and writing 0 to the corresponding duration interval when a later time point is earlier than an earlier time point to form an invalid time point mark; then, according to the graph connection relationship of the agricultural product node to be delivered, the delivery node, and the acceptance constraint relationship, the five types of time point codes, time point sequence codes, and duration interval codes are merged into freshness time sequence coding data;

[0090] Based on the chronological relationship between harvest time, loading time and expected arrival time, the freshness consumption stage of the agricultural products to be delivered during the cold chain distribution process is coded to generate freshness stage coding data.

[0091] The stage coding includes coding the period from picking time to loading time as the post-harvest waiting stage, and coding the period from loading time to estimated arrival time as the cold chain delivery stage, and writing 1 and 2 respectively; the stage duration in seconds is written according to the cumulative difference in seconds between the start and end times of the stage; if the end time of the stage is earlier than the start time of the stage, the stage duration in seconds is written as 0; and then the stage coding is merged with the time sequence coding to generate freshness stage coding data.

[0092] Based on the loading time, estimated arrival time, and the end time of the acceptance time window, the delivered time and the remaining deliverable time are encoded to generate time constraint coding data.

[0093] The duration encoding includes writing the cumulative difference in seconds between the loading time and the estimated arrival time into the delivered duration, and writing the cumulative difference in seconds between the estimated arrival time and the end of the acceptance time window into the remaining deliverable duration. The duration status is written as 1, 2, and 3 according to normal duration status, abnormal duration status, and over-window duration status, respectively. When the estimated arrival time is earlier than the loading time, the delivered duration is written as 0 and the abnormal duration status is written. When the estimated arrival time is later than the end of the acceptance time window, the remaining deliverable duration is written as 0 and the over-window duration status is written.

[0094] Based on the positional relationship between the expected arrival time and the start and end times of the acceptance time window, the status of the acceptance time window matching is encoded to generate acceptance matching code data.

[0095] The status coding includes classifying the matching status according to the positional relationship between the expected arrival time and the start and end times of the acceptance time window; when the expected arrival time is earlier than the start time of the acceptance time window, it is coded as the pre-window status; when the expected arrival time is between the start and end times of the acceptance time window, it is coded as the in-window status; when the expected arrival time is later than the end time of the acceptance time window, it is coded as the post-window status, and these are written as 1, 2, and 3 respectively. Then, according to the edge relationship vector connection direction of the acceptance constraint relationship, the corresponding matching status codes are merged into the acceptance matching code data.

[0096] The freshness time-series coding data, freshness stage coding data, duration constraint coding data, and acceptance matching coding data are overlaid onto the cold chain distribution map representation data to generate freshness time-series map representation data.

[0097] The overlay process involves aligning the nodes in the cold chain delivery fulfillment diagram with the agricultural product to be delivered, the delivery node, and the acceptance constraint as the alignment objects. This includes incorporating the freshness time-series coding data and freshness stage coding data into the initial graph node representation data of the agricultural product to be delivered node, incorporating the duration constraint coding data into the initial graph node representation data of the delivery node, and incorporating the acceptance matching coding data into the edge relationship vector of the acceptance constraint. When multiple coding data exist for the same fusion object, they are fused from earliest to latest according to the expected arrival time.

[0098] In this embodiment, the generation of freshness remaining life projection data specifically includes:

[0099] The freshness time series graph representation data is input into the multi-layer GPS update layer to extract cold chain environment data, cargo compartment loading data and estimated unloading time, and door opening disturbance input data is generated by aggregating the door opening and closing time and loading location.

[0100] The door opening disturbance input data includes the compartment temperature, compartment humidity, external ambient temperature, door opening time, door closing time, temperature and humidity recovery status, compartment area, loading position, and estimated unloading time;

[0101] The cumulative number of seconds between the door opening time and the door closing time is written into the door opening duration. If the door closing time is earlier than the door opening time, the door opening duration is written to 0 and an abnormal door opening flag is generated.

[0102] The door opening disturbance branch is used to transmit the door opening disturbance input data according to the loading relationship, the car area adjacency relationship and the loading position adjacency relationship, and generate the door opening disturbance local update data;

[0103] Specifically: Based on acceptance constraints and loading relationships, determine the agricultural product nodes where door opening and unloading will occur. Configure the area of ​​the carriage containing this agricultural product node as the disturbance initiation area, and the loading position of this agricultural product node as the disturbance initiation position. Based on the disturbance initiation area and disturbance initiation position, classify the agricultural product nodes within the carriage according to disturbance propagation priority. Agricultural product nodes in the same disturbance initiation area are classified into the direct disturbance level, with a disturbance impact coefficient of 1. Agricultural product nodes not classified into the direct disturbance level but adjacent to the disturbance initiation position are classified into the location-adjacent disturbance level, with a disturbance impact coefficient of 0.7. Agricultural product nodes not classified into either the direct or location-adjacent disturbance levels but adjacent to the disturbance initiation area within the carriage are classified into the area-adjacent disturbance level, with a disturbance impact coefficient of 0.5. Add the door opening duration to the expected unloading duration to obtain the door opening disturbance duration. The absolute value of the difference between the compartment temperature and the external ambient temperature is taken and divided by 20 degrees Celsius. If the result is greater than 1, it is recorded as 1 to obtain the temperature disturbance. The absolute value of the change in humidity in the compartment before and after the door is opened is taken and divided by 30 percentage points. If the result is greater than 1, it is recorded as 1 to obtain the humidity disturbance. The temperature and humidity recovery state code value is divided by 3 to obtain the recovery state value. The average of the temperature disturbance, humidity disturbance, and recovery state value is calculated to obtain the door opening disturbance intensity. The disturbance initiation area and disturbance initiation position are then coded. The code is used as the source of the door opening disturbance. The duration, intensity, and source of the door opening disturbance are encoded as 4-dimensional, 4-dimensional, and 8-dimensional, respectively, and then concatenated in sequence to generate a 16-dimensional door opening disturbance message. The duration and intensity of the door opening disturbance in the message are multiplied by the disturbance influence coefficient, while the source of the door opening disturbance remains unchanged. The attenuated door opening disturbance message is then incorporated into the freshness time series graph representation data of the corresponding agricultural product node to be delivered to generate local update data of the door opening disturbance.

[0104] By utilizing the fulfillment chain attention branches along the fulfillment sequence, the order of fulfillment impact between delivery nodes is organized, generating full-path update data for the fulfillment chain;

[0105] Specifically, the process involves: arranging delivery nodes into a delivery node fulfillment sequence based on their order of fulfillment; and, according to the acceptance constraints between the agricultural product nodes and the delivery nodes, a link aggregation is performed on the delivery nodes associated with the same agricultural product in the fulfillment sequence to generate an agricultural product fulfillment chain. If the first delivery node in the agricultural product fulfillment chain has no preceding delivery node, the corresponding fulfillment chain update value is written to 0, and the preceding fulfillment impact value is calculated starting from the next delivery node. For each subsequent delivery node in the agricultural product fulfillment chain, the preceding delivery nodes are extracted. All preceding delivery nodes are identified, and their estimated arrival time, node waiting status, estimated unloading time, freshness acceptance requirements, and freshness time-series representation data are extracted. The estimated arrival time and estimated unloading time of each preceding delivery node are added together to obtain the preceding fulfillment completion time. The estimated arrival time of subsequent delivery nodes is subtracted from the preceding fulfillment completion time to obtain the fulfillment connection interval. A fulfillment connection interval less than 0 is written as 4; an interval not less than 0 and not greater than 600 seconds is written as 3; and an interval greater than 600 seconds and not greater than 1800 seconds is written as 2. Write 1 when the estimated unloading time is greater than 1800 seconds; write 1 when the estimated unloading time is no more than 300 seconds, write 2 when the estimated unloading time is greater than 300 seconds but no more than 900 seconds, and write 3 when the estimated unloading time is greater than 900 seconds; write the freshness acceptance requirement code value according to the minimum remaining life corresponding to the freshness acceptance requirement, write 1 when the minimum remaining life is no more than 7200 seconds, and write 2 when the minimum remaining life is greater than 7200 seconds; combine the node waiting status code value, the estimated unloading time code value, and the freshness acceptance requirement code value. The preceding fulfillment impact value is obtained by adding the coded value and the fulfillment connection interval coded value. The preceding fulfillment impact value of a single preceding delivery node is divided by the sum of all preceding fulfillment impact values ​​associated with the same subsequent delivery node to obtain the fulfillment attention coefficient of that preceding delivery node. When the sum of all preceding fulfillment impact values ​​is 0, 1 is divided by the number of preceding delivery nodes as the fulfillment attention coefficient. The freshness time series representation data of the preceding delivery node is weighted according to the fulfillment attention coefficient and written along the agricultural product fulfillment chain to the subsequent delivery nodes and their associated agricultural product nodes to be delivered, generating full path update data of the fulfillment chain.

[0106] The local update data of the door opening disturbance and the full path update data of the performance chain are input into the feedforward update unit to perform layer-by-layer fusion and update of the freshness time series graph representation data, and generate freshness update graph representation data.

[0107] Specifically, according to the node connection relationship in the cold chain delivery fulfillment diagram, the partial update data of door opening disturbance is aligned to the corresponding agricultural product node to be delivered, and the full path update data of the fulfillment chain is aligned to the corresponding delivery node and its associated agricultural product node to be delivered; when the same agricultural product node to be delivered receives both the partial update data of door opening disturbance and the full path update data of the fulfillment chain, the node joint update data is formed in the order of writing the impact of door opening disturbance first and the impact of fulfillment chain last; the feedforward update unit dimensionally aligns the node joint update data with the original freshness time series diagram representation data, padding the end with 0s if the dimension is insufficient, and padding the end with 0s if the dimension is excessive. The data is mapped to a fixed length according to the field order; then, the node joint update data after dimension alignment is subjected to a first feedforward mapping, activation processing and a second feedforward mapping. The first feedforward mapping outputs 256-dimensional intermediate update data, the activation processing writes values ​​less than 0 to 0, and the second feedforward mapping outputs 128-dimensional node candidate update data. The node candidate update data is merged into the freshness time series graph representation data of the corresponding agricultural product nodes to be delivered and the delivery nodes, and the output of the current GPS update layer is used as the input of the next GPS update layer. The multi-layer GPS update layer is set to 3 layers, and freshness update graph representation data is generated after updating layer by layer.

[0108] Based on the freshness update map representation data, the remaining freshness lifespan of the agricultural products to be delivered at each delivery node is extrapolated to generate node freshness lifespan data.

[0109] Specifically: Based on the acceptance constraints between the agricultural product nodes to be delivered and the delivery nodes, the agricultural product nodes to be delivered, the delivery nodes, and the estimated arrival time are arranged as nodes for lifetime estimation; according to the correlation between agricultural product categories and freshness acceptance requirements, the longest deliverable time for the corresponding category is written into the category baseline lifetime; if the longest deliverable time is not obtained, the cumulative number of seconds between the harvest time and the end of the acceptance time window is written into the category baseline lifetime; if the freshness acceptance requirements do not include a minimum remaining lifetime, 20% of the category baseline lifetime is written into the minimum remaining lifetime; the category baseline lifetime is corrected according to the initial freshness state; if the initial freshness state is sufficient, the category baseline lifetime is written into the initial freshness lifetime; if the initial freshness state is critical, the category baseline lifetime is reduced by the product of the minimum remaining lifetime corresponding to the freshness acceptance requirements and the state code value, and then written into the initial freshness lifetime; if the initial freshness state is insufficient, the category baseline lifetime is reduced by... The minimum remaining lifespan corresponding to the freshness acceptance requirement is multiplied by the status code value and written into the initial freshness lifespan. When the initial freshness status is missing or unknown, the minimum remaining lifespan corresponding to the freshness acceptance requirement is written into the initial freshness lifespan. The delivered time is converted into the basic consumption amount, which is equal to the delivered time. The door opening disturbance duration, door opening disturbance intensity, and disturbance impact coefficient in the door opening disturbance local update data are multiplied to generate the disturbance consumption amount. The preceding performance impact value, performance attention coefficient, and the expected unloading time of the corresponding preceding delivery node in the fulfillment chain full path update data are multiplied to generate the performance consumption amount. The basic consumption amount, disturbance consumption amount, and performance consumption amount are deducted from the initial freshness lifespan to obtain the remaining freshness lifespan of the agricultural product to be delivered when it is expected to arrive at the delivery node. When the remaining freshness lifespan is less than 0, it is written as 0. The results are arranged according to the agricultural product node, delivery node, and expected arrival time to generate node freshness lifespan data.

[0110] Based on the node freshness lifetime data and freshness acceptance requirements, the acceptance satisfaction is determined, and the determination results are converted into freshness risk status and freshness deliverable status, and then the remaining freshness lifetime projection data is generated. Freshness risk status includes freshness normal status, freshness critical delivery status and freshness non-compliance status; freshness deliverable status includes deliverable status, critical deliverable status and non-deliverable status.

[0111] Specifically, based on the node lifespan projection object, the difference between the remaining freshness lifespan in the node freshness lifespan data and the minimum remaining freshness lifespan corresponding to the freshness acceptance requirements is processed; when the difference is greater than 0, the freshness risk status is written into the freshness normal status, and the freshness deliverable status is written into the deliverable status; when the difference is equal to 0, the freshness risk status is written into the freshness critical delivery status, and the freshness deliverable status is written into the critical deliverable status; when the difference is less than 0, the freshness risk status is written into the freshness substandard status, and the freshness deliverable status is written into the undeliverable status; then, according to the agricultural product node to be delivered, the delivery node, and the expected arrival time, the results of the remaining freshness lifespan, freshness risk status, and freshness deliverable status are arranged to generate freshness remaining lifespan projection data.

[0112] In this embodiment, the generation of logistics route planning results specifically includes:

[0113] Based on the freshness remaining life projection data and acceptance constraints, the nodes of agricultural products to be delivered that are expected to not meet the freshness acceptance requirements upon arrival, as well as their associated delivery nodes, are screened to generate freshness risk node data.

[0114] Freshness risk node data includes the agricultural product node to be delivered, associated delivery node, estimated arrival time, acceptance time window, remaining freshness life and minimum remaining freshness life; when the remaining freshness life is less than the minimum remaining freshness life corresponding to the freshness acceptance requirements, or when the freshness risk status is the freshness non-compliance status, the agricultural product node to be delivered, associated delivery node, estimated arrival time, acceptance time window, remaining freshness life and minimum remaining freshness life are written into the freshness risk node data.

[0115] The life gap value of freshness risk nodes is calculated based on the freshness risk node data, and the adjustment priority of freshness risk nodes is arranged in combination with the acceptance time window to generate risk node adjustment order data.

[0116] The lifetime gap value equals the minimum remaining lifetime minus the remaining lifetime for freshness; the larger the lifetime gap value, the higher the adjustment priority; when lifetime gap values ​​are the same, the earlier the end time of the acceptance window, the higher the adjustment priority; when the end time of the acceptance window is the same, the later the expected arrival time, the higher the adjustment priority.

[0117] Based on the risk node adjustment order data, the forward position is retrieved between delivery nodes, and the estimated arrival time after the forward movement is verified in combination with road traffic conditions and estimated travel time to generate the forward movement relationship of delivery nodes;

[0118] The forward position retrieval includes selecting candidate insertion positions one by one along the order of delivery, with the associated delivery node as the adjustment object; re-accumulating the estimated travel time between the candidate insertion position and the associated delivery node, and between the associated delivery node and the subsequent delivery node, to obtain the estimated arrival time after forward movement; when the estimated arrival time after forward movement is within the acceptance time window, and the road condition of the associated road segment is not impassable, the candidate insertion position is written into the forward movement relationship of the delivery node.

[0119] Based on the forward movement relationship of delivery nodes, the associated delivery nodes are inserted into the candidate forward movement positions, and the crossed delivery nodes are extended in the original relative order. The passage relationship and freshness delivery status are re-verified, and the logistics route planning results are generated. The logistics route planning results include the adjusted delivery node order, associated road segments, adjusted estimated arrival time, and adjusted freshness delivery status.

[0120] Specifically: The delivery order corresponding to the candidate forward position is written into the adjusted delivery order of the associated delivery node; the adjusted delivery order of delivery nodes whose original delivery order is between the candidate forward position and the original delivery order of the associated delivery node is increased by 1; the passage relationship between adjacent delivery nodes is reconstructed according to the adjusted delivery order, and the road segment, road traffic status and estimated travel time are re-associated, and the adjusted estimated arrival time is generated segment by segment; when the road traffic status is impassable, the corresponding adjustment result is deleted; the remaining freshness life is updated according to the adjusted estimated arrival time, and the difference with the minimum remaining life is processed to generate the adjusted freshness deliverable status; when the adjusted estimated arrival time meets the acceptance time window, and the adjusted freshness deliverable status is a deliverable status or a critical deliverable status, the adjusted delivery node order, associated road segment, adjusted estimated arrival time and adjusted freshness deliverable status are arranged into the logistics route planning result.

[0121] A logistics route planning system based on big data processing includes:

[0122] The dataset generation module is used to acquire data on agricultural products to be delivered, cold chain environment data, vehicle loading data, delivery node data, road traffic data, and customer acceptance data, and to perform standardized processing and object association to generate agricultural product cold chain delivery datasets.

[0123] The fulfillment graph construction module is used to construct a cold chain delivery fulfillment graph based on the agricultural product cold chain delivery dataset.

[0124] The graph representation encoding module is used to input the cold chain delivery fulfillment graph into the improved GraphGPS model, and use the input encoding layer to vectorize the graph node attributes and edge relationship attributes to generate cold chain delivery graph representation data.

[0125] The freshness temporal enhancement module is used to perform freshness temporal enhancement processing on the cold chain distribution map representation data using the freshness temporal coding layer, and generate freshness temporal map representation data;

[0126] The freshness lifetime projection module is used to perform graph update processing on the freshness time series graph representation data using multiple GPS update layers, and to integrate the effects of door opening disturbance and fulfillment chain to generate freshness remaining lifetime projection data.

[0127] The route planning generation module is used to identify freshness risk nodes based on freshness remaining life projection data, and determine the forward and backward movement relationship of delivery nodes based on freshness risk nodes and road traffic data, thereby generating logistics route planning results.

[0128] Example 1: To verify the feasibility of this invention in practice, it was applied to the morning delivery scenario of an agricultural product cold chain distribution company. The vehicle was loaded with 18 boxes of leafy greens, 12 boxes of strawberries, and 10 boxes of tomatoes. The leafy greens were picked at 06:10, the strawberries at 05:50, and the tomatoes at 05:40, all loaded at 07:05. The temperature inside the vehicle was 4.2 degrees Celsius, the humidity was 88%, and the ambient temperature was 31.6 degrees Celsius. The delivery task included preceding delivery nodes, intermediate delivery nodes, and subsequent delivery nodes, with estimated arrival times of 08:30, 09:45, and 11:10, and estimated unloading times of 360 seconds, 540 seconds, and 480 seconds, respectively. The acceptance time window for the intermediate delivery nodes was from 09:20 to 10:00, and the freshness acceptance requirement for the strawberries was a remaining shelf life of no less than 7200 seconds upon arrival.

[0129] During implementation, the picking time, loading time, door opening time, door closing time, estimated arrival time, and acceptance time window are all converted into cumulative seconds. The agricultural products to be delivered, the truck area, loading location, delivery nodes, and road sections are then linked to construct a cold chain delivery fulfillment map. In this scenario, strawberries are loaded near the truck door. At the preceding delivery node, the door opens at 08:32 and closes at 08:39, with temperature and humidity in recovery status. Simultaneously, the actual unloading time at the preceding delivery node is extended to 600 seconds due to customer waiting, the waiting time at the intermediate delivery node is 240 seconds, and the road leading to the subsequent delivery node is congested, increasing the estimated travel time by 900 seconds.

[0130] When applying this invention, the cold chain delivery fulfillment map is input into an improved GraphGPS model. A freshness time-series encoding layer expresses the freshness consumption relationship between harvesting, loading, expected arrival, and acceptance time windows. A door-opening disturbance branch handles the local impacts of door opening, external high temperatures, and proximity of loading locations on agricultural products inside the truck. A fulfillment chain attention branch handles the continuous impacts of delayed unloading, node waiting, and road congestion on subsequent fulfillment nodes. Based on this, the model extrapolates the remaining freshness life of each agricultural product at its expected arrival and moves or postpones delivery nodes that may not meet freshness acceptance requirements, generating logistics route planning results oriented towards on-site execution.

[0131] To verify the actual effect of route adjustment in the embodiments, based on the same delivery vehicle, the same batch of agricultural products, and the same road conditions, the traditional method of planning only according to delivery order and travel time was used as a control. The statistical results are shown in the table below:

[0132] Table 1. Performance Comparison of the Invention and Traditional Logistics Route Planning Methods

[0133] Freshness acceptance rate (%) 91.2 94.8 Error in estimating remaining shelf life (in minutes) 18.4 14.7 Arrival time window matching rate (%) 92.6 95.1 Door opening disturbance recognition coverage (%) 84.5 89.7 Delivery delay identification rate (%) 82.8 88.1 Average delivery delay (min) 11.6 9.8 Path adjustment efficiency (%) 87.4 91.0

[0134] As can be clearly seen from Table 1, the method of the present invention is superior to the traditional method in many indicators.

[0135] The freshness acceptance satisfaction rate of the method of this invention is increased from 91.2% of the traditional route planning method to 94.8%, an increase of 3.6 percentage points, indicating that the route planning results have a stronger ability to guarantee the freshness status of agricultural products upon arrival. This improvement is not simply achieved by shortening the travel time, but because this invention introduces the estimation of remaining freshness before route generation, incorporating harvesting time, loading time, estimated arrival time, and acceptance time window into the same delivery fulfillment relationship for judgment. Figure 2 This further demonstrates that when strawberries are associated with intermediate nodes, the remaining freshness lifespan under the traditional method is 113.8 min, which is lower than the acceptance requirement of 120.0 min; the method of the present invention improves it to 124.4 min after adjustment, which can meet the acceptance requirement, indicating that the present invention can effectively adjust for freshness risk nodes.

[0136] The remaining shelf life estimation error decreased from 18.4 min to 14.7 min, a reduction of 3.7 min, indicating that the present invention's estimation of freshness status upon arrival is closer to the actual acceptance situation. This is because the freshness time-series encoding can express the time consumption relationship between harvesting, loading, expected arrival, and acceptance time windows, so that the model no longer judges freshness changes solely based on transportation time. Figure 2When the preceding node is associated with leafy vegetables, the remaining freshness lifespan increases from 132.6 min to 136.2 min; when the following node is associated with tomatoes, the remaining freshness lifespan increases from 96.4 min to 103.1 min. This also shows that the present invention forms a continuous deduction for multiple delivery nodes, rather than treating a single risk node in isolation.

[0137] The door opening disturbance recognition coverage increased from 84.5% to 89.7%, a 5.2 percentage point improvement. This is because the invention addresses the impact of door opening and closing, external temperature, humidity changes in the cargo compartment, and proximity of loading locations through a door opening disturbance branch, allowing local cold chain disturbances generated during unloading at preceding nodes to be included in the freshness lifespan estimation process. The fulfillment delay transmission recognition rate increased from 82.8% to 88.1%, a 5.3 percentage point improvement. This is because the fulfillment chain attention branch continuously transmits node waiting, unloading delays, and the connection relationships between preceding and following nodes to subsequent delivery nodes, avoiding the isolated calculation of each delivery node in traditional methods.

[0138] The arrival time window matching rate increased from 92.6% to 95.1%, a 2.5 percentage point improvement; the average delivery delay time decreased from 11.6 minutes to 9.8 minutes, a reduction of 1.8 minutes; and the route adjustment effectiveness increased from 87.4% to 91.0%, a 3.6 percentage point improvement. This demonstrates that after identifying freshness risk nodes, the present invention can adjust the delivery nodes by moving them forward or backward based on road traffic conditions. This adjustment revolves around freshness acceptance requirements and delivery fulfillment relationships, thus improving the deliverability of cold chain agricultural products upon arrival while maintaining delivery timeliness.

[0139] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A logistics route planning method based on big data processing, characterized in that, Includes the following steps: Acquire data on agricultural products to be delivered, cold chain environment data, vehicle loading data, delivery node data, road traffic data, and customer acceptance data, and perform standardized processing and object association to generate an agricultural product cold chain delivery dataset; A cold chain delivery fulfillment map was constructed based on the agricultural product cold chain delivery dataset; The cold chain delivery fulfillment map is input into the improved GraphGPS model. The graph node attributes and edge relationship attributes are vectorized and encoded using the input encoding layer to generate cold chain delivery graph representation data. The improved GraphGPS model includes an input encoding layer, a freshness time-series encoding layer, a multi-layer GPS update layer, and an output layer. The freshness temporal coding layer is used to perform freshness temporal enhancement processing on the cold chain distribution map representation data to generate freshness temporal map representation data; The freshness time series graph representation data is updated by using a multi-layer GPS update layer, and the effects of opening disturbance and fulfillment chain are integrated to generate freshness remaining life projection data. Based on the remaining freshness life projection data, identify freshness risk nodes that are expected to fail to meet freshness acceptance requirements upon arrival. Then, determine the forward and backward movement relationships of delivery nodes based on freshness risk nodes and road traffic data, and generate logistics route planning results.

2. The logistics route planning method based on big data processing according to claim 1, characterized in that, The standardization process includes time unification, unit unification, status coding, missing data completion, and anomaly marking; the object association includes identifier matching, location mapping, and node association.

3. The logistics route planning method based on big data processing according to claim 1, characterized in that, The construction of the cold chain delivery fulfillment map specifically includes: Based on the agricultural products to be delivered and their loading locations in the cold chain agricultural product distribution dataset, establish nodes for agricultural products to be delivered, nodes for the area of ​​the vehicle, and the loading relationship between the two. Based on the delivery order, estimated unloading time, and node waiting status in the delivery node data, establish delivery nodes and establish a fulfillment priority relationship between adjacent delivery nodes; Based on the road segments, road traffic conditions, and estimated travel time between delivery nodes, road segment nodes are established, and road segment nodes are connected to associated delivery nodes to form traffic relationships; Based on the acceptance time window and freshness acceptance requirements of the agricultural products to be delivered at the delivery nodes, an acceptance constraint relationship is established between the agricultural product nodes to be delivered and the delivery nodes. A graph structure is then arranged for the agricultural product nodes to be delivered, the vehicle area nodes, the delivery nodes, the road segment nodes, the loading relationship, the order of performance, the passage relationship, and the acceptance constraint relationship to generate a cold chain delivery performance diagram.

4. The logistics route planning method based on big data processing according to claim 1, characterized in that, The generation of the cold chain delivery map representation data specifically includes: The cold chain delivery fulfillment map is input into the improved GraphGPS model, and the node attributes of each graph are extracted by the input encoding layer according to the node type and a node attribute encoding queue is established. The improvements of the improved GraphGPS model compared to the existing GraphGPS model are as follows: the original location structure encoding used to represent the topological location of nodes is replaced with freshness time sequence encoding oriented towards picking time, loading time, expected arrival time and acceptance time window; the original local message passing used for information transmission between adjacent nodes is replaced with door opening disturbance branch oriented towards the influence of door opening and closing, temperature and humidity changes and proximity of loading location; and the original global attention used for modeling the association of nodes in the whole graph is replaced with fulfillment chain attention branch oriented towards the influence of delivery sequence, node waiting and unloading connection. The category field, time field, status field, and numerical field in the node attribute encoding queue are vectorized and mapped separately to generate the initial representation data of the graph nodes; Extract the start and end nodes, relationship types, relationship connection order, and attribute fields of loading relationships, performance sequence relationships, passage relationships, and acceptance constraint relationships. Convert them into edge relationship vectors and align them with the initial representation data of the graph nodes to generate cold chain distribution graph representation data.

5. The logistics route planning method based on big data processing according to claim 1, characterized in that, The generation of the freshness time series graph representation data specifically includes: Input the cold chain distribution map representation data into the freshness time-series coding layer, extract the picking time, loading time, estimated arrival time and acceptance time window, and encode the time position of each time field to generate freshness time-series coded data; Based on the chronological relationship between harvest time, loading time and expected arrival time, the freshness consumption stage of the agricultural products to be delivered during the cold chain distribution process is coded to generate freshness stage coding data. Based on the loading time, estimated arrival time, and the end time of the acceptance time window, the delivered time and the remaining deliverable time are encoded to generate time constraint coding data. Based on the positional relationship between the expected arrival time and the start and end times of the acceptance time window, the status of the acceptance time window matching is encoded to generate acceptance matching code data. Freshness time-series coding data, freshness stage coding data, duration constraint coding data, and acceptance matching coding data are overlaid onto the cold chain distribution map representation data to generate freshness time-series map representation data.

6. The logistics route planning method based on big data processing according to claim 1, characterized in that, The generation of the freshness remaining life projection data specifically includes: The freshness time series graph representation data is input into the multi-layer GPS update layer to extract cold chain environment data, cargo compartment loading data and estimated unloading time, and door opening disturbance input data is generated by aggregating the door opening and closing time and loading location. The door opening disturbance branch is used to transmit the door opening disturbance input data according to the loading relationship, the car area adjacency relationship and the loading position adjacency relationship, and generate the door opening disturbance local update data; By utilizing the fulfillment chain attention branches along the fulfillment sequence, the order of fulfillment impact between delivery nodes is organized, generating full-path update data for the fulfillment chain; The local update data of the door opening disturbance and the full path update data of the performance chain are input into the feedforward update unit to perform layer-by-layer fusion and update of the freshness time series graph representation data, and generate freshness update graph representation data. Based on the freshness update map representation data, the remaining freshness lifespan of the agricultural products to be delivered at each delivery node is extrapolated to generate node freshness lifespan data. Based on the node freshness lifetime data and freshness acceptance requirements, acceptance satisfaction is determined, and the determination results are converted into freshness risk status and freshness deliverable status, and then freshness remaining lifetime projection data is generated. The freshness risk status includes freshness normal status, freshness critical delivery status and freshness non-compliance status; the freshness deliverable status includes deliverable status, critical deliverable status and non-deliverable status.

7. The logistics route planning method based on big data processing according to claim 1, characterized in that, The generation of the logistics route planning results specifically includes: Based on the freshness remaining life projection data and acceptance constraints, the nodes of agricultural products to be delivered that are expected to not meet the freshness acceptance requirements upon arrival, as well as their associated delivery nodes, are screened to generate freshness risk node data. The life gap value of freshness risk nodes is calculated based on the freshness risk node data, and the adjustment priority of freshness risk nodes is arranged in combination with the acceptance time window to generate risk node adjustment order data. Based on the risk node adjustment order data, the forward position is retrieved between delivery nodes, and the estimated arrival time after the forward movement is verified in combination with road traffic conditions and estimated travel time to generate the forward movement relationship of delivery nodes; Based on the forward movement relationship of delivery nodes, the associated delivery nodes are inserted into the candidate forward movement positions, and the crossed delivery nodes are extended in the original relative order. The passage relationship and freshness delivery status are re-verified, and the logistics route planning result is generated. The logistics route planning result includes the adjusted delivery node order, associated road segments, adjusted estimated arrival time, and adjusted freshness delivery status.

8. A logistics route planning system based on big data processing, executing the logistics route planning method based on big data processing as described in any one of claims 1 to 7, characterized in that, include: The dataset generation module is used to acquire data on agricultural products to be delivered, cold chain environment data, vehicle loading data, delivery node data, road traffic data, and customer acceptance data, and to perform standardized processing and object association to generate agricultural product cold chain delivery datasets. The fulfillment graph construction module is used to construct a cold chain delivery fulfillment graph based on the agricultural product cold chain delivery dataset. The graph representation encoding module is used to input the cold chain delivery fulfillment graph into the improved GraphGPS model, and use the input encoding layer to vectorize the graph node attributes and edge relationship attributes to generate cold chain delivery graph representation data. The freshness temporal enhancement module is used to perform freshness temporal enhancement processing on the cold chain distribution map representation data using the freshness temporal coding layer, and generate freshness temporal map representation data; The freshness lifetime projection module is used to perform graph update processing on the freshness time series graph representation data using multiple GPS update layers, and to integrate the effects of door opening disturbance and fulfillment chain to generate freshness remaining lifetime projection data. The route planning generation module is used to identify freshness risk nodes based on freshness remaining life projection data, and determine the forward and backward movement relationship of delivery nodes based on freshness risk nodes and road traffic data, thereby generating logistics route planning results.

Images