A decision-making method and system for responding to the timing of cargo dispatch

Abstract

The present invention discloses a decision-making method and system for responding to the timing of cargo sources entering the dispatch, which belongs to the field of intelligent dispatching technology, including: constructing a data set based on the non-real-time data of the cargo owner and the real-time data of the cargo source; based on the LightGBM algorithm, model training is performed through the data set, and whether the historical cargo source is traded is selected as the target variable to establish a classification model; based on the classification model, the historical cargo source is simulated and predicted, and the packing operation is performed, and the transaction probability is divided into several groups with equal intervals, and the best time for the cargo source to enter the dispatch under each layer is determined, and a decision response is made. The present invention is used to achieve the prediction of the transaction time and transaction probability of cargo sources of a larger order of magnitude when the amount of behavioral feature data related to the cargo owner is insufficient, and has a higher coverage rate for the timing of differentiated dispatch intervention.

Description

Technical Field

[0001] The present invention relates to the field of intelligent scheduling technology, and in particular to a decision-making method and system for responding to the timing of cargo source scheduling. Background Art

[0002] The dispatcher is a service staff responsible for matching transactions between drivers and cargo owners on the platform. The dispatcher's matching service is performed within the scope of the cargo candidate pool. Since the time when cargo owners release cargo sources every day varies from peak to valley, there is a manpower gap in the current reserve of dispatchers during the peak shipping period. The present invention accurately predicts the difficulty and duration of cargo source transactions based on the cargo owner's historical behavior data and the multi-dimensional information of cargo owners' release of cargo sources, thereby formulating differentiated strategies for the timing of cargo sources flowing into the candidate pool, solving the capacity pressure of the cargo candidate pool during peak shipping, and the derived scheduling manpower gap and other problems. Specifically, the method involved in the present invention includes portrait statistics based on the cargo owner's historical behavior, real-time cargo source information, etc., and predicts the possible transaction time and transaction rate of the cargo source, thereby deciding the timing of entering the dispatch.

[0003] In order to solve the problem of manpower shortage during peak delivery, there are several methods to solve the problem of manpower shortage during peak delivery and the timing of cargo dispatch:

[0004] Extend the time it takes for goods to enter the pool: Extend the time it takes for all goods to enter the pool. High-quality goods (which can be traded without scheduling and matching) can be taken off the shelves due to transactions before entering the pool, thereby alleviating the capacity pressure of the entire supply pool.

[0005] Historical behavior statistical inference: Through the transaction records of the cargo sources released by the cargo owner in the past, the transaction time of the cargo source released by the cargo owner in the past is counted. For example, the average transaction time of the cargo source of the cargo owner in the past 7 transactions is counted as the transaction time of the cargo source released by the cargo owner in the future; for the cargo source released by the cargo owner, the timing of entering the dispatch can be set within the transaction time.

[0006] However, the prior art has the following defects:

[0007] 1. Extend the time for goods to enter the pool: The disadvantage of this method is that after the time for entering the pool is extended, for cargo owners who are in a hurry to sell goods, the rate of goods being taken off the shelves will increase due to the lack of timely intervention by the dispatcher. At the same time, some high-quality goods will be snatched away by other offline channels.

[0008] 2. Historical behavior statistical inference: The defect of this method is that the inference logic is weak; the historical behavior of the consignor is the result statistics of the consignor over a long period of time, which cannot well reflect the quality of the consignor's current supply. This method does not use real-time data, such as the transaction time of the same-day delivery supply on the platform, and lacks the follow-up and real-time nature of capturing the consignor's behavior. In addition, this method has a low coverage rate for consignors: for consignors whose supply has never been fulfilled in the past / newly registered consignors, there is a lack of historical behavior data. Summary of the invention

[0009] In order to solve the above problems, the purpose of the present invention is to provide a decision-making technology that responds to the timing of cargo scheduling, aiming to achieve higher scheduling and matching efficiency.

[0010] In order to achieve the above technical objectives, the present application provides a decision-making method for responding to the timing of cargo source scheduling, including the following steps:

[0011] Build a data set based on the non-real-time data of cargo owners and the real-time data of cargo sources;

[0012] Based on the LightGBM algorithm, the model is trained through the data set, and whether the historical source of goods has been sold is selected as the target variable to establish a classification model;

[0013] Based on the classification model, the historical supply is simulated and predicted, and the packing operation is carried out. The transaction probability is divided into several groups with equal intervals. The best time for the supply to enter the scheduling under each layer is determined, and the decision response is made.

[0014] Preferably, in the process of constructing the data set, non-real-time data is obtained based on the historical records of the cargo owner publishing the cargo source on the platform and the cargo owner's identity information;

[0015] Based on the bidding mode of the source of goods, the category of goods, the length of the required vehicle and the bidding information of the cargo owner, it is used as real-time data;

[0016] Obtain data sets based on non-real-time data and real-time data.

[0017] Preferably, in the process of acquiring the data set, based on non-real-time data and real-time data, the historical behavior of the shipper, the identity of the shipper, the market conditions and the historical supply information are screened on the principle that the data has high representativeness and avoids introducing irrelevant data as noise to obtain the data set, wherein the missing characteristic values ​​of the data set are filled with a fixed value of -9999.

[0018] Preferably, during the model training process, the LightGBM algorithm is used to discretize continuous floating-point eigenvalues ​​into finite integers, and a histogram with a width of k is constructed. When traversing the data, the statistics are accumulated in the histogram based on the discretized values ​​as indexes. After traversing the data once, the histogram accumulates the required statistics, and then the optimal segmentation point is traversed based on the discrete values ​​of the histogram.

[0019] Preferably, in the process of determining the best time for the supply to enter scheduling in each layer, based on the several groups divided, the transaction time of the supply is statistically inferred according to the average transaction time / median transaction time / upper quartile of the supply, to determine the best time for the supply to enter scheduling in each layer.

[0020] Preferably, in the process of determining the best time for the supply to enter the scheduling under each layer, the period before the median transaction time is used as the best time for scheduling intervention.

[0021] The present invention discloses a decision-making system for responding to the timing of cargo source scheduling, which is used to implement the above-mentioned decision-making method for responding to the timing of cargo source scheduling, including:

[0022] The data collection module is used to construct a data set based on the non-real-time data of the cargo owner and the real-time data of the cargo source;

[0023] The model building module is used to train the model through the data set based on the LightGBM algorithm, select whether the historical source of goods has been sold as the target variable, and establish a classification model;

[0024] The decision response module is used to simulate and predict historical supply sources based on the classification model, and perform packing operations, divide the transaction probabilities into several groups with equal intervals, determine the best time for the supply to enter the scheduling under each layer, and make decision responses.

[0025] Preferably, the data collection module is also used to obtain non-real-time data based on the historical records of the cargo owner releasing the source of goods on the platform and the identity information of the cargo owner; use the bidding mode of the source of goods, the category of goods, the required vehicle length and the information of the cargo owner's bid as real-time data; and obtain a data set based on non-real-time data and real-time data.

[0026] Preferably, the data acquisition module is also used to obtain a data set by screening the non-real-time data and real-time data among the historical behavior of the consignor, the identity of the consignor, the market conditions and the historical supply information on the principle of high representativeness of the data and avoiding the introduction of irrelevant data as noise, wherein the missing characteristic values ​​of the data set are filled with a fixed value of -9999.

[0027] Preferably, the decision response module is also used to determine the best time for the supply to enter scheduling in each layer, based on the several groups divided, and statistically infer the transaction time of the supply according to the average transaction time / median transaction time / upper quartile of the supply, to determine the best time for the supply to enter scheduling in each layer, wherein the period before the median transaction time is used as the best time for scheduling intervention.

[0028] The present invention discloses the following technical effects:

[0029] The present invention can timely capture changes in supply information and adjust the prediction results accordingly. Its follow-up and accuracy are better than the prediction method that only uses historical data statistics.

[0030] The present invention has achieved positive business effects in actual use. In the scenario where dispatchers match drivers and cargoes, without adding additional manpower reserves:

[0031] a) The matching success rate increased by 2 percentage points;

[0032] b) Reduced the delisting of goods due to the inability to intervene in a timely manner: the delisting rate of goods decreased by 1.78 percentage points; improved user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 It is a schematic diagram of the LightGB model structure of the present invention;

[0035] Figure 2 It is a schematic diagram of the overall process of the present invention;

[0036] Figure 3 It is a classification of situations in which a model is used to predict the dispatch of supply of goods in one embodiment of the present invention. DETAILED DESCRIPTION

[0037] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0038] like Figure 1-Figure 2 As shown, the present invention provides a decision-making technology for responding to the timing of cargo entering scheduling. The method involved in the present invention uses data from both real-time and non-real-time sources. By designing Spark programs and Python scripts, real-time multi-dimensional evaluations are performed on the cargo sources released by cargo owners, thereby inferring the timing of entering scheduling; by reasonably arranging the timing of entering scheduling, the scheduling efficiency is maximized. Specifically, the non-real-time data used in the present invention includes the historical records of cargo owners releasing cargo sources on the platform and the cargo owners' identity information; the real-time data includes the cargo source's bidding mode, cargo category, required vehicle length, cargo owner's bid and other information. Specifically, it includes the following process:

[0039] 1. Basic data preparation:

[0040] The present invention uses prediction technology based on machine learning models to conduct multi-dimensional evaluation of the quality of the source of goods, in which the quality and representativeness of the data are crucial to the final effect of the model. Combined with the understanding of industry knowledge and feature selection technology in the field of artificial intelligence: in the historical behavior of the owner / owner identity / market conditions / historical source information, based on the principle that the data has high representativeness and avoids the introduction of irrelevant data as noise, more than 60 feature data and more than 800,000 sample data were finally determined. At the same time, in order to help the algorithm better complete the task, the feature data is preprocessed, such as using a fixed value of -9999 to fill in the missing feature values.

[0041] 2. Supply quality assessment - establishing a supply transaction probability model

[0042] Based on comprehensive considerations such as prediction efficiency and accuracy, the LightGB algorithm is selected. LightGB discretizes continuous floating-point feature values ​​(Float data) into finite integers and constructs a histogram (Bin data) with a width of k. When traversing the data, the statistics are accumulated in the histogram based on the discretized values ​​as indexes (Construst histogram). After traversing the data once, the histogram accumulates the required statistics, and then the optimal split point (Decision tree) is traversed based on the discrete values ​​of the histogram. The model structure diagram is shown in the figure below. Figure 1 shown.

[0043] 2.1. The process of building the model includes:

[0044] a) Generate training set and test set: Divide the sample data into training set and test set to make final preparations for model training. The training set is used for model training and optimization, while the test set is used to evaluate the performance of the model and ensure the generalization ability of the model.

[0045] b) Model training: Select whether the historical supply has been sold as the target variable and establish a classification model. Through continuous training, verification and tuning, a balance between model fitting ability and generalization ability is achieved.

[0046] c) After the model training is completed, the historical supply sources are simulated and predicted. The output of the model is a discrete value between [0,1]. In order to better understand the transaction time pattern of the supply source, the model prediction results are boxed and the transaction probability is divided into 9 groups with equal intervals. For example, the transaction probability of supply A predicted by the model is 0.12, so it is merged into the predicted transaction rate bucket of [10%, 20%).

[0047] Under the 9 prediction buckets, the median transaction time under each bucket is counted as the inferred value of the future transaction time under the bucket, so as to determine the best time for the supply to enter the scheduling under each layer (scheduling intervention is carried out 10 minutes in advance under the median transaction time), as follows Figure 3 As shown in the table.

[0048] d) Model evaluation and optimization: Use the test set to evaluate the performance of the model and optimize the model parameters based on the evaluation results to improve the accuracy and generalization ability of the model.

[0049] e) Online deployment: The process of deploying the trained machine learning model to the online environment with the help of internal tools so that it can accept external input and generate prediction results. Figure 2 shown.

[0050] The method involved in the present invention attempts to combine the portrait statistics of the cargo owner's historical behavior, the real-time information of the source of goods, etc., to comprehensively predict the possible transaction time and transaction probability of the source of goods, so as to decide the timing of entering the scheduling. Using the present invention, the cargo matching platform can formulate prediction and intervention strategies without relying on the global extension of the source of goods entering the pool: accurately define the source of goods portrait, introduce differentiated scheduling intervention opportunities for thousands of goods, reduce the delisting rate caused by the global extension of intervention, and achieve higher scheduling and matching efficiency.

[0051] At the same time, compared with relying solely on the historical behavior data of the cargo owner, the present invention combines the real-time information of the cargo source, such as the cargo source's bidding mode, cargo category, required vehicle length, cargo owner's bid, etc. The historical behavior data involved in the present invention includes the cargo owner's historical shipping behavior and the cargo owner's historical order sequence. When the amount of behavioral feature data related to the cargo owner is insufficient, the present invention can be used to predict the transaction duration and transaction probability of a larger order of magnitude, and has a higher coverage rate for the timing of differentiated scheduling intervention.

[0052] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0053] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0054] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A decision-making method for responding to the timing of cargo dispatch, characterized in that: The following steps are involved: Build a data set based on the non-real-time data of cargo owners and the real-time data of cargo sources; Based on the LightGBM algorithm, the model is trained through the data set, whether the historical source of goods is sold is selected as the target variable, and a classification model is established; Based on the classification model, simulate and predict the historical supply, and perform packing operations, divide the transaction probabilities into several groups with equal intervals, determine the best time for the supply to enter the dispatch under each layer, and make decision responses; In the process of determining the best time for the supply to enter the dispatch under each layer, based on the divided groups, the supply transaction time is inferred according to the median transaction time of the supply, and the best time for the supply to enter the dispatch under each layer is determined; In the process of determining the best time for the supply to enter the dispatch under each layer, 10 minutes in advance of the median transaction time is the best time for dispatch intervention; During the model training process, the LightGBM algorithm is used to discretize continuous floating-point eigenvalues ​​into finite integers, and a histogram with a width of k is constructed. When traversing the data, the statistics are accumulated in the histogram based on the discretized values ​​as indexes. After traversing the data once, the histogram accumulates the required statistics, and then the optimal segmentation point is traversed based on the discrete values ​​of the histogram.

2. A decision-making method for responding to the timing of cargo dispatch according to claim 1, characterized in that: In the process of constructing the data set, the non-real-time data is obtained based on the historical records of the cargo owner publishing the cargo source on the platform and the cargo owner's identity information; The real-time data is based on the bidding mode of the source of goods, the category of goods, the length of the required vehicle and the bidding information of the owner of the goods; The data set is acquired according to the non-real-time data and the real-time data.

3. A decision-making method for responding to the timing of cargo dispatch according to claim 2, characterized in that: In the process of acquiring the data set, based on the non-real-time data and the real-time data, the historical behavior of the consignor, the identity of the consignor, the market conditions and the historical supply information are screened on the principle that the data has high representativeness and avoids introducing irrelevant data as noise to obtain the data set, wherein the missing characteristic values ​​of the data set are filled with a fixed value of -9999.

Images