A distribution management method and device, a storage medium and an electronic device

Abstract

The present application relates to a kind of distribution management method, device, storage medium and electronic equipment, it is related to distribution management technical field, the present application constructs user receiving probability distribution model using deep learning, and correction is carried out in combination with real-time telephone communication result, effectively solve the problem of too early or no one to receive mail caused by recipient not at home, greatly reduce the secondary delivery rate and loss risk of fresh products, by injecting the scoring system with fresh type and its preservation threshold as core constraint variable, ensure that the order of high time requirement can obtain higher distribution priority, complete delivery before product quality deterioration, improve user experience, while realizing the balanced scheduling of regional transport capacity through dynamic scoring mechanism, effectively alleviate the technical pain points of uneven resource allocation in peak period, avoid the phenomenon that part of terminal is overloaded while another part is idle, based on real-time scoring dynamically generates path, can flexibly respond to road condition anomaly, user temporary change of mind and other emergent conditions.

Description

Technical Field

[0001] This invention relates to the field of distribution management technology, and in particular to a distribution management method, apparatus, storage medium and electronic device. Background Technology

[0002] Compared to traditional commodity delivery, fresh produce delivery is characterized by high timeliness requirements, complex temperature control conditions, high perishability, and diverse product categories. Any delays, temperature control failures, or inappropriate routes during the delivery process can directly affect the quality of fresh produce, potentially leading to customer complaints and financial losses. Therefore, ensuring both delivery efficiency and quality has become a critical technical challenge for fresh produce e-commerce companies.

[0003] Currently, the most common fresh food order delivery management methods in the industry rely primarily on fixed delivery route planning, manual scheduling, and experience-driven delivery decisions. While these traditional methods can meet daily operational needs to some extent, they have significant limitations when facing complex scenarios such as a surge in order volume, uneven regional delivery, and personalized user needs.

[0004] Resource mismatch: During peak hours (such as morning and evening commutes), orders surge, leading to insufficient capacity; during off-peak hours, capacity is idle.

[0005] Delivery time conflict: Lack of accurate awareness of the recipient's schedule leads to the user not being home when the courier arrives, causing fresh products to rot and spoil due to prolonged storage.

[0006] Lack of flexibility: It is impossible to adjust delivery priorities in real time based on road conditions, weather, and the rate of deterioration of fresh produce quality;

[0007] To address the aforementioned technical deficiencies, a solution is proposed. Summary of the Invention

[0008] The purpose of this invention is to effectively solve the problem of premature delivery or no one to pick up the package when the recipient is not at home, greatly reduce the secondary delivery rate and loss risk of fresh products, and ensure that orders with high timeliness requirements can receive higher delivery priority by injecting the types of fresh products and their preservation thresholds as core constraint variables into the scoring system.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: a distribution management method, comprising the following steps:

[0010] Step 1: Collect recipients' historical delivery time preferences, types of fresh produce in orders and their freshness thresholds, and monitor the geographical location information and load status of delivery resources in the region in real time, and integrate them into a delivery source dataset;

[0011] Step 2: Construct a user receiving probability distribution model based on a deep learning network, and then obtain the recipient's historical delivery time preference as training samples to train the user receiving probability distribution model, thus obtaining the optimized user receiving probability distribution model.

[0012] Step 3: Obtain the delivery source dataset and parse it to obtain the types of fresh produce in the order and their preservation thresholds to input into the user receiving probability distribution model, obtain the order receiving probability distribution results, and at the same time obtain the user's telephone communication results to correct the order receiving probability distribution results;

[0013] Step 4: Obtain the delivery source dataset and parse it to obtain the geographical location information and load status of delivery resources in the real-time monitoring area. Combine the corrected order acceptance probability distribution results to calculate the real-time delivery score of the order.

[0014] Step 5: Dynamically generate delivery routes based on real-time delivery scores for orders and send them to delivery terminals.

[0015] Furthermore, the specific process for obtaining training samples is as follows:

[0016] S101. Obtain the delivery source dataset collected in step one and extract features, including the recipient's historical signing time, signing duration, order type, fresh produce category, historical rejections, and rescheduling frequency, to obtain continuous time information.

[0017] S102. Using days as the period, the period is divided into several time slices, and then the continuous time information is transformed into discrete probability distribution feature vectors as training samples.

[0018] Furthermore, the specific process of obtaining the optimized user reception probability distribution model is as follows:

[0019] S201. A user reception probability distribution model is constructed using a long short-term memory network. The user reception probability distribution model includes an input layer, multiple hidden layers, and an output layer. The output layer uses the Softmax function to output the probability distribution of successful user acceptance within several future time windows.

[0020] S202. Obtain training samples and use 7:2:1 as the training set. Mark each successfully signed order record as a positive sample, and mark the unanswered or rescheduled order records as negative samples to form a time-stamped supervised learning sample pair.

[0021] S203. The stochastic gradient descent algorithm is adopted. During the training process, the weight parameters and bias terms in the neural network are continuously adjusted through the backpropagation algorithm, and the user reception probability distribution model is tested for inference using reserved validation set data.

[0022] If the overlap between the time period with the highest predicted probability and the actual receipt time period is less than a set threshold, then increase the network depth or adjust the learning rate until the probability distribution result output by the user reception probability distribution model tends to be stable and accurate, thereby obtaining the optimized user reception probability distribution model.

[0023] Furthermore, the specific process for obtaining the corrected order acceptance probability distribution is as follows:

[0024] S301. Obtain the delivery source dataset and use the preset parsing engine to extract two key dimensions of data for the current order to be processed from the delivery source dataset:

[0025] Fresh produce attribute characteristics: used to identify the physical attributes of fresh produce in an order and its corresponding preservation threshold, i.e., the critical time Tlimit from when the product leaves the warehouse to when its quality is damaged.

[0026] Predictive contextual features include the current timestamp of order generation, current weather conditions, and holiday attributes;

[0027] S302. Input the fresh produce attribute features and prediction context features into the optimized user reception probability distribution model obtained in step two to output a discrete probability density function that varies with time. The discrete probability density function represents the historical probability of the recipient signing for the order at different time periods without considering external real-time intervention.

[0028] S303. Set the current time to tnow. If for time point t, t-tnow > Tlimit, then the receiving probability f(t) at that time point is penalized and set to zero.

[0029] S304. Trigger automatic voice interaction to obtain user feedback results. The user feedback results are classified and coded as: immediate acceptance, acceptance at a specified time, and unacceptable acceptance. If the user feedback result is acceptance at a specified time, a corrected distribution is constructed using a Gaussian kernel function K(t) centered at the specified time tuser, and then weighted and fused with the original probability distribution to obtain the corrected order acceptance probability distribution result. :

[0030] ,in 1 is the preset proportional coefficient. 2 represents the preset feedback confidence level;

[0031] If the user reports that the order cannot be signed for, the probability of receiving the order in the current delivery round will be globally set to zero.

[0032] Furthermore, the specific process for calculating the real-time delivery score for an order is as follows:

[0033] S401. Parse the real-time parameters from the delivery source dataset, extract the integral value of the current time point tnow in the corrected probability distribution curve fcorrected(t), and convert the integral value into the receiving probability factor Fp.

[0034] S402. Obtain the actual road network distance L between the delivery person and the order location, introduce the traffic congestion coefficient to calculate the estimated delivery time Test. If the estimated delivery time Test is much smaller than the fresh food preservation threshold, the score will get a positive gain, and the estimated delivery time Test will be converted into the distance time consumption factor Fd.

[0035] S403. Monitor the load status of delivery terminals in the area to calculate the load rate, which is the ratio between the current order volume and the maximum capacity. If the deliveryman's load is higher than the preset load threshold, the corresponding balance factor Fb will output a negative value.

[0036] S404. After normalizing the above factors, calculate the real-time delivery score of the order according to the following formula:

[0037] Where e1, e2 and e3 are preset weighting coefficients.

[0038] Furthermore, the specific process for generating delivery routes is as follows:

[0039] S501. Using the delivery terminal as the node and the order location as the target point, construct the vehicle routing model with time window constraints as follows:

[0040] Objective function: Let xij be the delivery path from order i to order j, and define the optimization objective as:

[0041]

[0042] Where Si is the real-time delivery rating of order i. This is the urgency level coefficient for the order. The time and distance costs between paths;

[0043] Hard constraints include the maximum capacity of the delivery terminal, the remaining working hours of the delivery personnel, and the strong constraint of the freshness threshold for fresh products, that is, the delivery time must be less than the freshness threshold.

[0044] S502. Sort orders in descending order according to real-time delivery rating S, and set the order with the highest rating as the priority node in the path sequence. After determining the high-rated node, search for low-rated nodes around the high-rated node. If the Costij added by inserting a low-rated node does not affect the timeliness of the high-rated node, then merge it into the current path.

[0045] S503. By monitoring the load of multiple delivery resources in the area in real time, if the paths of two delivery terminals overlap, the order is assigned to the delivery terminal that can generate a higher real-time score by comparing their respective score gains for the same order.

[0046] S504. When the telephone communication result in step three changes, causing the order score S to surge instantly, immediately interrupt the current route calculation and re-insert the order into the next nearest node in the current driving route of the delivery terminal.

[0047] S505. After the delivery person completes each order, retrieve the latest delivery source dataset, update the real-time delivery rating of all orders to be delivered, and regenerate the delivery routes for the remaining orders.

[0048] The present invention also provides a delivery management device, comprising a data collection unit, a model building unit, a probability estimation unit, an order scoring calculation unit, and a route planning unit, wherein:

[0049] The data collection unit is used to collect recipients' historical delivery time preferences, the types of fresh produce in the order and their freshness thresholds, and at the same time monitor the geographical location information and load status of delivery resources in the area in real time, and integrate them into a delivery source dataset.

[0050] The model building unit is used to build a user receiving probability distribution model based on a deep learning network, and then obtain the recipient's historical delivery time preference as a training sample to train the user receiving probability distribution model, thus obtaining the optimized user receiving probability distribution model.

[0051] The probability estimation unit is used to acquire the delivery source dataset, parse the types of fresh produce in the order and their freshness thresholds to input into the user receiving probability distribution model, obtain the order receiving probability distribution result, and acquire the user's telephone communication results to correct the order receiving probability distribution result.

[0052] The order rating calculation unit is used to acquire the delivery source dataset, parse the geographic location information and load status of delivery resources in the real-time monitoring area, and combine the corrected order acceptance probability distribution results to calculate the real-time delivery rating of the order.

[0053] The route planning unit is used to dynamically generate delivery routes based on real-time delivery scores of orders and send them to delivery terminals.

[0054] The present invention also provides a storage medium having a computer program stored thereon, which, when executed by a processor of an electronic device, causes the electronic device to perform the method described in any one of claims 1 to 6.

[0055] The present invention also provides an electronic device, comprising: a memory storing a computer program; and a processor for reading the computer program stored in the memory to execute the method according to any one of claims 1 to 6.

[0056] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0057] This delivery management method, device, storage medium, and electronic equipment utilize deep learning to construct a user reception probability distribution model and combine it with real-time telephone communication results for correction. This effectively solves the problem of premature delivery or no one to pick up the package when the recipient is not at home, greatly reducing the secondary delivery rate and loss risk of fresh products. By injecting the types of fresh products and their freshness thresholds as core constraint variables into the scoring system, it ensures that orders with high timeliness requirements receive higher delivery priority and are delivered before product quality deteriorates, thus improving the user experience. At the same time, the dynamic scoring mechanism achieves balanced scheduling of transportation capacity within the region, effectively alleviating the technical pain point of uneven resource allocation during peak periods and avoiding the phenomenon of some terminals being overloaded while others are idle. Based on real-time scoring, the dynamic route generation can flexibly respond to emergencies such as abnormal road conditions and users changing their minds at the last minute. Attached Figure Description

[0058] Figure 1 A schematic diagram of the overall method of the present invention is shown;

[0059] Figure 2 A schematic diagram of the overall structure of the present invention is shown. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Example 1:

[0062] like Figure 1 As shown, a delivery management method is characterized by comprising the following steps:

[0063] Step 1: Collect recipients' historical delivery time preferences, types of fresh produce in orders and their freshness thresholds, and monitor the geographical location information and load status of delivery resources in the region in real time, and integrate them into a delivery source dataset;

[0064] Step 2: Construct a user receiving probability distribution model based on a deep learning network, and then obtain the recipient's historical delivery time preference as training samples to train the user receiving probability distribution model, thus obtaining the optimized user receiving probability distribution model.

[0065] The specific process for obtaining training samples is as follows:

[0066] S101. Obtain the delivery source dataset collected in step one and extract features, including the recipient's historical signing time, signing duration, order type, fresh produce category, historical rejections, and rescheduling frequency, to obtain continuous time information.

[0067] S102. Using days as the period, the period is divided into several time slices, and then the continuous time information is transformed into discrete probability distribution feature vectors as training samples.

[0068] The specific process of obtaining the optimized user reception probability distribution model is as follows:

[0069] S201. A user reception probability distribution model is constructed using a long short-term memory network. The user reception probability distribution model includes an input layer, multiple hidden layers, and an output layer. The output layer uses the Softmax function to output the probability distribution of successful user acceptance within several future time windows.

[0070] S202. Obtain training samples and use 7:2:1 as the training set. Mark each successfully signed order record as a positive sample, and mark the unanswered or rescheduled order records as negative samples to form a time-stamped supervised learning sample pair.

[0071] S203. The stochastic gradient descent algorithm is adopted. During the training process, the weight parameters and bias terms in the neural network are continuously adjusted through the backpropagation algorithm, and the user reception probability distribution model is tested for inference using reserved validation set data.

[0072] If the overlap between the time period with the highest predicted probability and the actual receipt time period is less than a set threshold, then increase the network depth or adjust the learning rate until the probability distribution result output by the user reception probability distribution model tends to be stable and accurate, thereby obtaining the optimized user reception probability distribution model.

[0073] Step 3: Obtain the delivery source dataset and parse it to obtain the types of fresh produce in the order and their preservation thresholds to input into the user receiving probability distribution model, obtain the order receiving probability distribution results, and at the same time obtain the user's telephone communication results to correct the order receiving probability distribution results;

[0074] The specific process for obtaining the corrected order acceptance probability distribution is as follows:

[0075] S301. Obtain the delivery source dataset and use the preset parsing engine to extract two key dimensions of data for the current order to be processed from the delivery source dataset:

[0076] Fresh produce attribute characteristics: used to identify the physical attributes of fresh produce in an order and its corresponding preservation threshold, i.e., the critical time Tlimit from when the product leaves the warehouse to when its quality is damaged.

[0077] Predictive contextual features include the current timestamp of order generation, current weather conditions, and holiday attributes;

[0078] S302. Input the fresh produce attribute features and prediction context features into the optimized user reception probability distribution model obtained in step two to output a discrete probability density function that varies with time. The discrete probability density function represents the historical probability of the recipient signing for the order at different time periods without considering external real-time intervention.

[0079] S303. Set the current time to tnow. If for time point t, t-tnow > Tlimit, then the receiving probability f(t) at that time point is penalized and set to zero.

[0080] S304. Trigger automatic voice interaction to obtain user feedback results. The user feedback results are classified and coded as: immediate acceptance, acceptance at a specified time, and unacceptable acceptance. If the user feedback result is acceptance at a specified time, a corrected distribution is constructed using a Gaussian kernel function K(t) centered at the specified time tuser, and then weighted and fused with the original probability distribution to obtain the corrected order acceptance probability distribution result. :

[0081] ,in 1 is the preset proportional coefficient. 2 represents the preset feedback confidence level;

[0082] If the user reports that the order cannot be signed for, the probability of receiving the order in the current delivery round will be globally set to zero.

[0083] Step 4: Obtain the delivery source dataset and parse it to obtain the geographical location information and load status of delivery resources in the real-time monitoring area. Combine the corrected order acceptance probability distribution results to calculate the real-time delivery score of the order.

[0084] The specific process for calculating the real-time delivery rating for an order is as follows:

[0085] S401. Parse the real-time parameters from the delivery source dataset, extract the integral value of the current time point tnow in the corrected probability distribution curve fcorrected(t), and convert the integral value into the receiving probability factor Fp.

[0086] S402. Obtain the actual road network distance L between the delivery person and the order location, introduce the traffic congestion coefficient to calculate the estimated delivery time Test. If the estimated delivery time Test is much smaller than the fresh food preservation threshold, the score will get a positive gain, and the estimated delivery time Test will be converted into the distance time consumption factor Fd.

[0087] S403. Monitor the load status of delivery terminals in the area to calculate the load rate, which is the ratio between the current order volume and the maximum capacity. If the deliveryman's load is higher than the preset load threshold, the corresponding balance factor Fb will output a negative value.

[0088] S404. After normalizing the above factors, calculate the real-time delivery score of the order according to the following formula:

[0089] Where e1, e2 and e3 are preset weighting coefficients.

[0090] Step 5: Dynamically generate delivery routes based on real-time delivery scores for orders and send them to delivery terminals.

[0091] The specific process for generating delivery routes is as follows:

[0092] S501. Using the delivery terminal as the node and the order location as the target point, construct the vehicle routing model with time window constraints as follows:

[0093] Objective function: Let xij be the delivery path from order i to order j, and define the optimization objective as:

[0094]

[0095] Where Si is the real-time delivery rating of order i. This is the urgency level coefficient for the order. The time and distance costs between paths;

[0096] Hard constraints include the maximum capacity of the delivery terminal, the remaining working hours of the delivery personnel, and the strong constraint of the freshness threshold for fresh products, that is, the delivery time must be less than the freshness threshold.

[0097] S502. Sort orders in descending order according to real-time delivery rating S, and set the order with the highest rating as the priority node in the path sequence. After determining the high-rated node, search for low-rated nodes around the high-rated node. If the Costij added by inserting a low-rated node does not affect the timeliness of the high-rated node, then merge it into the current path.

[0098] S503. By monitoring the load of multiple delivery resources in the area in real time, if the paths of two delivery terminals overlap, the order is assigned to the delivery terminal that can generate a higher real-time score by comparing their respective score gains for the same order.

[0099] S504. When the telephone communication result in step three changes, causing the order score S to surge instantly, immediately interrupt the current route calculation and re-insert the order into the next nearest node in the current driving route of the delivery terminal.

[0100] S505. After the delivery person completes each order, retrieve the latest delivery source dataset, update the real-time delivery rating of all orders to be delivered, and regenerate the delivery routes for the remaining orders.

[0101] This invention utilizes deep learning to construct a user reception probability distribution model and combines it with real-time telephone communication results for correction. This effectively solves the problem of premature delivery or no one to pick up the package due to the recipient not being home, greatly reducing the secondary delivery rate and loss risk of fresh produce. By injecting the types of fresh produce and their preservation thresholds as core constraint variables into the scoring system, it ensures that orders with high timeliness requirements receive higher delivery priority and are delivered before product quality deteriorates, thus improving the user experience. At the same time, the dynamic scoring mechanism achieves balanced scheduling of transportation capacity within the region, effectively alleviating the technical pain point of uneven resource allocation during peak periods and avoiding the phenomenon of some terminals being overloaded while others are idle. Based on real-time scoring, the dynamic path generation can flexibly respond to emergencies such as abnormal road conditions and users changing their minds.

[0102] Example 2:

[0103] like Figure 2 As shown, the present invention also provides a delivery management device, including a data collection unit, a model building unit, a probability estimation unit, an order scoring calculation unit, and a route planning unit, wherein:

[0104] The data collection unit is used to collect recipients' historical delivery time preferences, the types of fresh produce in the order and their freshness thresholds, and at the same time monitor the geographical location information and load status of delivery resources in the area in real time, and integrate them into a delivery source dataset.

[0105] The model building unit is used to build a user receiving probability distribution model based on a deep learning network, and then obtain the recipient's historical delivery time preference as a training sample to train the user receiving probability distribution model, thus obtaining the optimized user receiving probability distribution model.

[0106] The probability estimation unit is used to acquire the delivery source dataset, parse the types of fresh produce in the order and their freshness thresholds to input into the user receiving probability distribution model, obtain the order receiving probability distribution result, and acquire the user's telephone communication results to correct the order receiving probability distribution result.

[0107] The order rating calculation unit is used to acquire the delivery source dataset, parse the geographic location information and load status of delivery resources in the real-time monitoring area, and combine the corrected order acceptance probability distribution results to calculate the real-time delivery rating of the order.

[0108] The route planning unit is used to dynamically generate delivery routes based on real-time delivery scores of orders and send them to delivery terminals.

[0109] The size of the interval and threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value.

[0110] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0111] The above description is only a preferred embodiment 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 delivery management method, characterized in that, Includes the following steps: Step 1: Collect recipients' historical delivery time preferences, types of fresh produce in orders and their freshness thresholds, and monitor the geographical location information and load status of delivery resources in the region in real time, and integrate them into a delivery source dataset; Step 2: Construct a user receiving probability distribution model based on a deep learning network, and then obtain the recipient's historical delivery time preference as training samples to train the user receiving probability distribution model, thus obtaining the optimized user receiving probability distribution model. Step 3: Obtain the delivery source dataset and parse it to obtain the types of fresh produce in the order and their preservation thresholds to input into the user receiving probability distribution model, obtain the order receiving probability distribution results, and at the same time obtain the user's telephone communication results to correct the order receiving probability distribution results; Step 4: Obtain the delivery source dataset and parse it to obtain the geographical location information and load status of delivery resources in the real-time monitoring area. Combine the corrected order acceptance probability distribution results to calculate the real-time delivery score of the order. Step 5: Dynamically generate delivery routes based on real-time delivery scores for orders and send them to delivery terminals.

2. The delivery management method according to claim 1, characterized in that, The specific process for obtaining training samples is as follows: S101. Obtain the delivery source dataset collected in step one and extract features, including the recipient's historical signing time, signing duration, order type, fresh produce category, historical rejections, and rescheduling frequency, to obtain continuous time information. S102. Using days as the period, the period is divided into several time slices, and then the continuous time information is transformed into discrete probability distribution feature vectors as training samples.

3. The delivery management method according to claim 1, characterized in that, The specific process of obtaining the optimized user reception probability distribution model is as follows: S201. A user reception probability distribution model is constructed using a long short-term memory network. The user reception probability distribution model includes an input layer, multiple hidden layers, and an output layer. The output layer uses the Softmax function to output the probability distribution of successful user acceptance within several future time windows. S202. Obtain training samples and use 7:2:1 as the training set. Mark each successfully signed order record as a positive sample, and mark the unanswered or rescheduled order records as negative samples to form a time-stamped supervised learning sample pair. S203. The stochastic gradient descent algorithm is adopted. During the training process, the weight parameters and bias terms in the neural network are continuously adjusted through the backpropagation algorithm, and the user reception probability distribution model is tested for inference using reserved validation set data. If the overlap between the time period with the highest predicted probability and the actual receipt time period is less than a set threshold, then increase the network depth or adjust the learning rate until the probability distribution result output by the user reception probability distribution model tends to be stable and accurate, thereby obtaining the optimized user reception probability distribution model.

4. The delivery management method according to claim 1, characterized in that, The specific process for obtaining the corrected order acceptance probability distribution is as follows: S301. Obtain the delivery source dataset and use the preset parsing engine to extract two key dimensions of data for the current order to be processed from the delivery source dataset: Fresh produce attribute characteristics: used to identify the physical attributes of fresh produce in an order and its corresponding preservation threshold, i.e., the critical time Tlimit from when the product leaves the warehouse to when its quality is damaged. Predictive contextual features include the current timestamp of order generation, current weather conditions, and holiday attributes; S302. Input the fresh produce attribute features and prediction context features into the optimized user reception probability distribution model obtained in step two to output a discrete probability density function that varies with time. The discrete probability density function represents the historical probability of the recipient signing for the order at different time periods without considering external real-time intervention. S303. Set the current time to tnow. If for time point t, t-tnow > Tlimit, then the receiving probability f(t) at that time point is penalized and set to zero. S304. Trigger automatic voice interaction to obtain user feedback results. The user feedback results are classified and coded as: immediate acceptance, acceptance at a specified time, and unacceptable acceptance. If the user feedback result is acceptance at a specified time, a corrected distribution is constructed using a Gaussian kernel function K(t) centered at the specified time tuser, and then weighted and fused with the original probability distribution to obtain the corrected order acceptance probability distribution result. : ,in 1 is the preset proportional coefficient. 2 represents the preset feedback confidence level; If the user reports that the order cannot be signed for, the probability of receiving the order in the current delivery round will be globally set to zero.

5. The distribution management method according to claim 1, characterized in that, The specific process for calculating the real-time delivery rating for an order is as follows: S401. Parse the real-time parameters from the delivery source dataset, extract the integral value of the current time point tnow in the corrected probability distribution curve fcorrected(t), and convert the integral value into the receiving probability factor Fp. S402. Obtain the actual road network distance L between the delivery person and the order location, introduce the traffic congestion coefficient to calculate the estimated delivery time Test. If the estimated delivery time Test is much smaller than the fresh food preservation threshold, the score will get a positive gain, and the estimated delivery time Test will be converted into the distance time consumption factor Fd. S403. Monitor the load status of delivery terminals in the area to calculate the load rate, which is the ratio between the current order volume and the maximum capacity. If the deliveryman's load is higher than the preset load threshold, the corresponding balance factor Fb will output a negative value. S404. After normalizing the above factors, calculate the real-time delivery score of the order according to the following formula: Where e1, e2 and e3 are preset weighting coefficients.

6. The delivery management method according to claim 1, characterized in that, The specific process for generating delivery routes is as follows: S501. Using the delivery terminal as the node and the order location as the target point, construct the vehicle routing model with time window constraints as follows: Objective function: Let xij be the delivery path from order i to order j, and define the optimization objective as: Where Si is the real-time delivery rating of order i. This is the urgency level coefficient for the order. The time and distance costs between paths; Hard constraints include the maximum capacity of the delivery terminal, the remaining working hours of the delivery personnel, and the strong constraint of the freshness threshold for fresh products, that is, the delivery time must be less than the freshness threshold. S502. Sort orders in descending order according to real-time delivery rating S, and set the order with the highest rating as the priority node in the path sequence. After determining the high-rated node, search for low-rated nodes around the high-rated node. If the Costij added by inserting a low-rated node does not affect the timeliness of the high-rated node, then merge it into the current path. S503. By monitoring the load of multiple delivery resources in the area in real time, if the paths of two delivery terminals overlap, the order is assigned to the delivery terminal that can generate a higher real-time score by comparing their respective score gains for the same order. S504. When the telephone communication result in step three changes, causing the order score S to surge instantly, immediately interrupt the current route calculation and re-insert the order into the next nearest node in the current driving route of the delivery terminal. S505. After the delivery person completes each order, retrieve the latest delivery source dataset, update the real-time delivery rating of all orders to be delivered, and regenerate the delivery routes for the remaining orders.

7. A distribution management device according to claim 1, characterized in that, It includes a data collection unit, a model building unit, a probability estimation unit, an order scoring calculation unit, and a route planning unit, among which: The data collection unit is used to collect recipients' historical delivery time preferences, the types of fresh produce in the order and their freshness thresholds, and at the same time monitor the geographical location information and load status of delivery resources in the area in real time, and integrate them into a delivery source dataset. The model building unit is used to build a user receiving probability distribution model based on a deep learning network, and then obtain the recipient's historical delivery time preference as a training sample to train the user receiving probability distribution model, thus obtaining the optimized user receiving probability distribution model. The probability estimation unit is used to acquire the delivery source dataset, parse the types of fresh produce in the order and their freshness thresholds to input into the user receiving probability distribution model, obtain the order receiving probability distribution result, and acquire the user's telephone communication results to correct the order receiving probability distribution result. The order rating calculation unit is used to acquire the delivery source dataset, parse the geographic location information and load status of delivery resources in the real-time monitoring area, and combine the corrected order acceptance probability distribution results to calculate the real-time delivery rating of the order. The route planning unit is used to dynamically generate delivery routes based on real-time delivery scores of orders and send them to delivery terminals.

8. A storage medium, characterized in that, It stores a computer program that, when executed by the processor of the electronic device, causes the electronic device to perform the method described in any one of claims 1 to 6.

9. An electronic device, characterized in that, include: Memory, which stores computer programs; A processor reads a computer program stored in memory to perform the method described in any one of claims 1 to 6.

Images