Method for determining transportation scheme, method for training fast loading model, and device

Patent metadata

Field	Value
Publication number	US-2020364664-A1
Application number	US-202016986508-A
Country	US
Kind code	A1
Filing date	Aug 6, 2020
Priority date	Feb 6, 2018
Publication date	Nov 19, 2020
Grant date	—

Abstract

Embodiments of this application provide a method for obtaining a transportation scheme, including: obtaining a plurality of route schemes and a plurality of goods allocation scheme sets corresponding to each route scheme, wherein each route scheme comprises a transportation route for transporting to-be-transported goods, and the each of the goods allocation scheme sets includes at least one goods allocation scheme; obtaining, by using a fast loading model, predicted actual loading rates of each goods allocation scheme, wherein the fast loading model is trained using offline simulation data of a loading scheme that is calculated using a three-dimensional loading algorithm; and evaluating, using the actual loading rates, each route scheme and a corresponding goods allocation scheme, to obtain a target transportation scheme.

First claim

1 - 20 . (canceled)

21 . A method for obtaining a transportation scheme, comprising:
obtaining a plurality of route schemes and a plurality of goods allocation scheme sets corresponding to each of the route schemes, wherein each of the route schemes comprise a transportation route for transporting to-be-transported goods, each of the goods allocation scheme sets comprising at least one goods allocation scheme, and each goods allocation scheme in a goods allocation scheme set corresponding to a route scheme is a scheme for allocating the to-be-transported goods to each transportation route in the corresponding route scheme; obtaining, by using a fast loading model, predicted actual loading rates of each goods allocation scheme in the goods allocation scheme sets, wherein the fast loading model is trained using offline simulation data, the offline simulation data comprises a loading scheme calculated using a three-dimensional loading algorithm, and the predicted actual loading rates are predicted proportions of goods loaded into a container in a goods allocation scheme relative to a limit of the container; and evaluating, using the predicted actual loading rates, each route scheme and each goods allocation scheme in the goods allocation scheme sets, to obtain a target transportation scheme, wherein the target transportation scheme comprises a target route scheme and a target goods allocation scheme corresponding to the target route scheme.

22 . The method according to claim 21 , wherein the obtaining the plurality of route schemes and the plurality of goods allocation scheme sets corresponding to each of the route schemes comprises:
obtaining a target freight bill, wherein the target freight bill comprises transportation node information and to-be-transported goods information, the transportation node information comprises a freight starting point, a freight ending point, and M pickup points, and the to-be-transported goods information comprises information about to-be-transported goods distributed at the M pickup points, wherein M is a positive integer; obtaining the route schemes based on the transportation node information, wherein each transportation route comprises a freight starting point, a freight ending point, and N of the M pickup points, and each route scheme covers the M pickup points, wherein N is a positive integer and N≤M; and allocating the to-be-transported goods for each transportation route in each route scheme, to obtain each goods allocation scheme in the first goods allocation scheme set corresponding to each route scheme.

23 . The method according to claim 22 , wherein the obtaining the route schemes based on the transportation node information comprises:
if an amount of historical route data is greater than a first threshold, initializing transfer hyperparameters of the M pickup points based on the historical route data, to obtain a hyperparameter matrix; obtaining a transfer probability distribution based on the hyperparameter matrix and indicating a probability that a transportation route should be used, wherein the transfer probability distribution comprises a transfer probability of a container in a transportation route between the freight starting point and the M pickup points, between the freight ending point and the M pickup points, or between the M pickup points; and obtaining each transportation route in each of the at least one route scheme based on the transfer probability distribution, to obtain the at least one route scheme.

24 . The method according to claim 23 , wherein the method further comprises:
if the amount of historical route data is not greater than the first threshold, initializing the transfer hyperparameters of the M pickup points by using a heuristic algorithm, to obtain the hyperparameter matrix.

25 . The method according to claim 22 , wherein the allocating the to-be-transported goods for each transportation route in each route scheme, to obtain each goods allocation scheme in the first goods allocation scheme set corresponding to each route scheme comprises:
clustering goods at each of the M pickup points based on a clustering condition, to obtain a clustered set of goods, wherein the clustering condition comprises a length, a width, a height, and a weight of the goods; performing sampling calculation on the clustered set of goods by using a first goods allocation hyperparameter of each of the M pickup points, to obtain a first goods allocation set for each of the M pickup points, wherein the first goods allocation hyperparameter of each of the M pickup points is a hyperparameter for allocating the goods at each of the M pickup points, and each goods allocation in the first goods allocation set of each of the M pickup points is an allocation of goods distributed at a pickup point for a corresponding route scheme; and separately selecting a goods allocation from the first goods allocation set of each of the M pickup points, and combining the goods allocation with a route scheme, to obtain each goods allocation scheme in the first goods allocation scheme set corresponding to each route scheme.

26 . The method according to claim 22 , wherein the obtaining, by using a fast loading model, predicted actual loading rates of each goods allocation scheme in the goods allocation scheme sets comprises:
obtaining a first feature vector of each goods allocation scheme in the first goods allocation scheme set corresponding to each route scheme, wherein the first feature vector is used to indicate a feature value of to-be-transported goods in a goods allocation scheme; and inputting the first feature vector of each goods allocation scheme in the first goods allocation scheme set corresponding to each route scheme into the fast loading model, to obtain the predicted actual loading rate of each goods allocation scheme in the first goods allocation scheme set corresponding to each route scheme, wherein the predicted actual loading rate comprises a volume predicted actual loading rate and a weight predicted actual loading rate, the volume predicted actual loading rate comprises a proportion of a volume of goods allocated in each transportation route relative to a load volume of a container in each route scheme, and the weight predicted actual loading rate comprises a proportion of a weight of goods allocated in each transportation route relative to a load weight of a container in each route scheme.

27 . The method according to claim 26 , wherein the obtaining a first feature vector of each goods allocation scheme in the first goods allocation scheme set corresponding to each route scheme comprises:
obtaining a second feature vector of each piece of the to-be-transported goods, wherein the second feature vector of each piece of the to-be-transported goods comprises a length, a width, a height, and a weight of the corresponding goods; calculating, based on the second feature vector of each piece of the to-be-transported goods, a third feature vector of goods distributed at each of the M pickup points, for each goods allocation scheme in the first goods allocation scheme set corresponding to each route scheme, wherein the third feature vector of each goods allocation scheme in the first goods allocation scheme set corresponding to each route scheme comprises an average value and a covariance of second feature vectors of all pieces of the to-be-transported goods; and performing weighted combination on the third feature vector of each goods allocation scheme in the first goods allocation scheme set corresponding to each route scheme, to obtain the corresponding first feature vector in each goods allocation scheme in the first goods allocation scheme set corresponding to each of the at least one route scheme.

Assignees

• Huawei Tech Co Ltd

Inventors

• Chen Haoyang
• Lee Kai Yan
• Yuan Mingxuan
• Zeng Jia

Classifications

• G06Q10/047 Primary

  Optimisation of routes or paths, e.g. travelling salesman problem · CPC title

• G06F30/27

  using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model · CPC title

• G06Q10/0838

  Historical data · CPC title

• G06Q10/08

  Logistics, e.g. warehousing, loading or distribution; Inventory or stock management · CPC title

• G06Q10/04

  Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" (market predictions or forecasting for commercial activities G06Q30/0202) · CPC title