Construction method of intelligent pricing model based on freight big data

Abstract

The invention discloses an intelligent pricing model construction method based on freight big data, and relates to the technical field of freight pricing, and the method comprises the following steps: S1, carrying out the multi-channel freight data collection, building an independent data storage server, and storing and managing the collected data; s2, preprocessing and cleaning the collected freight data; s3, carrying out automatic labeling and manual labeling on the cleaned data; s4, establishing and training an intelligent pricing model; s5, carrying out the comprehensive analysis of the freight demand information provided by the user through the intelligent pricing model, and predicting the freight prices under different conditions; according to the method, feature items are analyzed through acquisition and processing of various data in a freight service link, a predictable intelligent pricing model is constructed by using technologies such as big data analysis, data labeling and machine learning, and intelligent prediction and dynamic adjustment of a cargo transportation price are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of freight pricing, and particularly to a method for constructing an intelligent pricing model based on freight big data. Background Art

[0002] Currently, with the rise of freight platforms, the goods transportation industry is undergoing profound changes. Against the backdrop of the rapid development of big data and artificial intelligence technologies, it has become possible to perform automatic intelligent pricing using the massive data of freight platforms. Through data mining and data analysis, it is possible to more accurately predict and formulate goods transportation prices, thereby improving the overall efficiency and competitiveness of the industry.

[0003] However, currently, the logistics industry basically uses manual calculation methods for pricing, relying too much on manual experience, lacking scientificity and accuracy, and it is difficult to meet the actual needs of freight platforms. Therefore, a method for constructing an intelligent pricing model based on freight big data is proposed. Summary of the Invention

[0004] The purpose of the present invention is to solve the problems in the prior art, and a method for constructing an intelligent pricing model based on freight big data is proposed.

[0005] A method for constructing an intelligent pricing model based on freight big data includes the following steps:

[0006] S1. Collect freight data from multiple channels, establish an independent data storage server, and store and manage the collected data;

[0007] S2. Preprocess and clean the collected freight data;

[0008] S3. Automatically annotate and manually annotate the cleaned data;

[0009] S4. Establish and train an intelligent pricing model;

[0010] S5. Use the intelligent pricing model to comprehensively analyze the freight demand information provided by users, and predict the freight prices under different conditions;

[0011] S6. Collect the feedback opinions of users using the intelligent pricing model on the pricing results, and optimize and improve the model according to the feedback results to improve the pricing accuracy and user satisfaction.

[0012] Preferably, in the step S1, the freight data includes waybill data and Internet of Things data. The waybill data includes customer order number, freight bill number, transportation mode, transportation route, cargo type, cargo weight, cargo volume, carrier vehicle type, energy type, transportation mileage, transportation time, shipping address, receiving address, transportation service, and freight composition. The Internet of Things data includes historical oil price, seasonal factors, holiday factors, weather conditions, traffic conditions, and regional transport capacity supply-demand relationship. The waybill data is synchronously stored once every 24 hours, and the Internet of Things data is synchronously stored once every 30 minutes.

[0013] Preferably, in the step S2, the cleaning of the freight data includes the following steps:

[0014] a. The first step is to judge the data integrity: If the number of missing data items exceeds 30%, it is determined as invalid data;

[0015] b. The second step is to judge the data repeatability: Use the unique customer order number or freight bill number to detect data repeatability, and delete or mark duplicate data according to business requirements;

[0016] c. The third step is to judge the data correctness: Set a reasonable range for data items. For data with more than 30% of abnormal data items, it is determined as invalid data.

[0017] Preferably, in the step S3, the automatic annotation judgment rules are as follows:

[0018] Vehicle length: If the weight of the waybill cargo exceeds the vehicle load, mark it as abnormal; if the weight of the waybill cargo is less than 50% of the vehicle load, mark it as abnormal;

[0019] Vehicle type: If the vehicle load exceeds 30 tons and the vehicle type is not a tractor, mark it as abnormal;

[0020] Transportation time: If the transportation duration does not match the transportation mileage, mark it as abnormal;

[0021] Transportation mileage: The transportation of this mileage cannot be completed within the transportation duration;

[0022] Cargo type: When there are a small number of cargo types below 10% for the same customer, mark it as abnormal;

[0023] Cargo weight: If the weight of the waybill cargo exceeds the vehicle load, mark it as abnormal; if the weight of the waybill cargo is less than 50% of the vehicle load, mark it as abnormal; if the cargo weight is 0 or 1, mark it as abnormal; if the unit price per vehicle kilometer is abnormal, mark it as abnormal;

[0024] Transportation costs: If the freight rate exceeds the reasonable range of the unit price per ton-kilometer, it is marked as abnormal; if a person is charged multiple times, it is marked as abnormal; if multiple vehicles are charged under the same account, there may be a situation of inflated freight, so it is marked as abnormal; if the freight income of a single vehicle in a single month exceeds 100,000, there may be a situation of inflated freight, so it is marked as abnormal.

[0025] Deductions and fines: If the transaction price of the waybill is inconsistent with the payment amount, it is marked as abnormal.

[0026] Transportation unit price: Determine the reasonable range of freight prices based on the unit price per ton-kilometer and the unit price per vehicle-kilometer.

[0027] Preferably, in step S3, the manual annotation is divided into an annotation management terminal module, an annotation work terminal module, and an annotation review terminal module;

[0028] The annotation management terminal module is responsible for task allocation, progress monitoring, annotation rule setting, and annotator management;

[0029] The annotation work terminal module is used to receive the tasks assigned by the annotation management terminal module, provide the provided annotation tools to annotate the data, and finally submit the annotation results to the annotation management terminal module and the annotation review terminal module;

[0030] The annotation review terminal module reviews the annotation results submitted by the annotation work terminal module, ensures the accuracy and consistency of the annotation, gives feedback and correction opinions, counts the review results, and provides a basis for the performance of annotators;

[0031] The work process of the manual annotation includes the following steps:

[0032] (1) Determine the annotation requirements, including data types and annotation types;

[0033] (2) Allocate annotation tasks and assign the tasks to suitable annotators;

[0034] (3) The annotator logs in to the annotation work terminal module and receives the tasks;

[0035] (4) Use the annotation tools to annotate the data according to the annotation specifications;

[0036] (5) After the annotation is completed, submit the annotation results;

[0037] (6) The reviewer logs in to the annotation review terminal module and receives the annotation tasks to be reviewed;

[0038] (7) Review the annotation results one by one to ensure the accuracy and consistency of the annotation. If any annotation errors or non-compliant places are found, give feedback and correction opinions;

[0039] (8) After the review is completed, submit the review results to the annotation management terminal module;

[0040] (9) Evaluate and reward the annotators according to the review results.

[0041] Preferably, in the step S4, training the intelligent pricing model includes the following steps:

[0042] (1) Divide the effective data set into a training set, a validation set and a test set;

[0043] (2) Adopt batch processing, incremental processing, and automatic online training methods to train the intelligent pricing model. Data fusion, statistical framework, and hierarchical model technologies are used to train a domain model separately for each transportation business scenario;

[0044] (3) Optimize the model performance by adjusting the parameters of the intelligent pricing model. Based on the data, perform feature importance ranking and recursive feature elimination to screen the features that have the greatest impact on the freight rate, and perform weighted processing on the features to improve the prediction accuracy of the model;

[0045] (4) Set evaluation indicators to measure the performance of the intelligent pricing model;

[0046] (5) Adopt the cross-validation method to evaluate the stability and generalization ability of the model. According to the evaluation results, select the optimal model for deployment.

[0047] Preferably, the data annotation quality evaluation rules are as follows:

[0048] Completion quality score = qualified annotation score (full score 50) + annotation quantity score (full score 50);

[0049] Qualified annotation score = number of qualified randomly checked data / total number of randomly checked data * 50;

[0050] Annotation quantity score = (number of standard fields / number of completed tasks) / 10% * 50.

[0051] Compared with the existing technology, the advantages of the present invention are as follows:

[0052] Through the collection and processing of various types of data in the freight business process, the present invention analyzes the characteristic items, and uses big data analysis, data annotation, machine learning and other technologies to construct a predictable intelligent pricing model, realizing the intelligent prediction and dynamic adjustment of freight transportation prices. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flowchart of the construction method of the intelligent pricing model based on freight big data of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0054] To make the technical means, creative features, achieved purposes and effects of the present invention easily understood, the present invention will be further described below in conjunction with specific embodiments.

[0055] Refer to Figure 1 As shown, a method for constructing an intelligent pricing model based on freight big data includes the following steps:

[0056] S1. Collect freight data from multiple channels, establish an independent data storage server, and store and manage the collected data;

[0057] S2. Preprocess and clean the collected freight data;

[0058] S3. Automatically label and manually label the cleaned data;

[0059] S4. Establish and train an intelligent pricing model;

[0060] S5. Use the intelligent pricing model to comprehensively analyze the freight demand information provided by users, and predict the freight prices under different conditions;

[0061] S6. Collect the feedback opinions of users using the intelligent pricing model on the pricing results, optimize and improve the model according to the feedback results, and improve the pricing accuracy and user satisfaction.

[0062] In this embodiment, in step S1, the freight data includes waybill data and Internet of Things data. The waybill data includes customer order number, freight bill number, transportation mode, transportation route, cargo type, cargo weight, cargo volume, carrier vehicle type, energy type, transportation mileage, transportation time, shipping address, receiving address, transportation service and freight composition. The Internet of Things data includes historical oil prices, seasonal factors, holiday factors, weather conditions, traffic conditions, regional transport capacity supply and demand relationship; the waybill data is synchronously stored once every 24 hours, and the Internet of Things data is synchronously stored once every 30 minutes.

[0063] In this embodiment, in step S2, the cleaning of the freight data includes the following steps:

[0064] a. The first step is to judge data integrity: if the number of missing data items exceeds 30%, it is determined as invalid data;

[0065] b. The second step is to judge data repeatability: use the unique customer order number or freight bill number to detect data repeatability, and delete or mark duplicate data according to business requirements;

[0066] c. The third step is to judge data correctness: set a reasonable range for data items, and for data with abnormal data items exceeding 30%, it is determined as invalid data.

[0067] In this embodiment, in step S3, the automatic annotation judgment rules are as follows:

[0068] Conductor: If the weight of the goods in the waybill exceeds the vehicle's load capacity, mark it as abnormal; if the weight of the goods in the waybill is less than 50% of the vehicle's load capacity, mark it as abnormal;

[0069] Vehicle type: If the vehicle's load capacity exceeds 30 tons and the vehicle type is not a tractor, mark it as abnormal;

[0070] Transportation time: If the transportation duration does not match the transportation mileage, mark it as abnormal;

[0071] Transportation mileage: The transportation of this mileage cannot be completed within the transportation duration;

[0072] Goods type: For the same customer, when there is a small amount of goods type less than 10%, mark it as abnormal;

[0073] Goods weight: If the weight of the goods in the waybill exceeds the vehicle's load capacity, mark it as abnormal; if the weight of the goods in the waybill is less than 50% of the vehicle's load capacity, mark it as abnormal; if the goods weight is 0 or 1, mark it as abnormal; if the vehicle-kilometer unit price is abnormal, mark it as abnormal;

[0074] Transportation fee: If the freight rate exceeds the reasonable range of the ton-kilometer unit price, mark it as abnormal; when one person collects money multiple times, mark it as abnormal; when multiple vehicles collect money with the same account, there may be a situation of inflated freight, mark it as abnormal; when the monthly freight income of a single vehicle exceeds 100,000, there may be a situation of inflated freight, mark it as abnormal;

[0075] Deductions and fines: If the transaction price of the waybill is inconsistent with the payment amount, mark it as abnormal;

[0076] Transportation unit price: Judge the reasonable range of the freight price according to the ton-kilometer unit price and the vehicle-kilometer unit price.

[0077] In this embodiment, in step S3, the manual annotation is divided into an annotation management terminal module, an annotation work terminal module, and an annotation review terminal module;

[0078] The annotation management terminal module is for task assignment, progress monitoring, annotation rule setting, and annotator management;

[0079] The annotation work terminal module is used to receive the tasks assigned by the annotation management terminal module, provide the provided annotation tools to annotate the data, and finally submit the annotation results to the annotation management terminal module and the annotation review terminal module;

[0080] The annotation review terminal module reviews the annotation results submitted by the annotation work terminal module, ensures the accuracy and consistency of the annotation, gives feedback and correction opinions, counts the review results, and provides a basis for the annotator's performance;

[0081] The workflow of the manual annotation includes the following steps:

[0082] (1) Determine the annotation requirements, including data types and annotation types;

[0083] (2) Allocate annotation tasks and assign them to appropriate annotators;

[0084] (3) The annotator logs in to the annotation workbench module to receive tasks;

[0085] (4) Use the annotation tool to annotate the data according to the annotation specifications;

[0086] (5) After the annotation is completed, submit the annotation results;

[0087] (6) The reviewer logs in to the annotation review module to receive the annotation tasks to be reviewed;

[0088] (7) Review the annotation results one by one to ensure the accuracy and consistency of the annotation. If any annotation errors or non-compliant places are found, give feedback and correction opinions;

[0089] (8) After the review is completed, submit the review results to the annotation management module;

[0090] (9) Evaluate the quality and give rewards to the annotators according to the review results.

[0091] The rules for evaluating the quality of data annotation are as follows:

[0092] Completion quality score = qualified annotation score (full score 50) + annotation quantity score (full score 50);

[0093] Qualified annotation score = number of qualified sampled data / total number of sampled data * 50;

[0094] Annotation quantity score = (number of standard fields / number of completed tasks) / 10% * 50.

[0095] Overall quality: Grade A: Completion quality score > 80; Grade B: Completion quality score 60 - 80; Grade C: Completion quality score < 60. Reward the annotators according to the completion quality score.

[0096] In this embodiment, in step S4, training the intelligent pricing model includes the following steps:

[0097] (1) Divide the valid data set into a training set, a validation set, and a test set;

[0098] (2) Use batch processing, incremental processing, and automatic online training methods to train the intelligent pricing model. The data fusion, statistical framework, and hierarchical model technologies are used, and a domain model is trained separately for each transportation business scenario;

[0099] (3) Optimize the model performance by adjusting the parameters of the intelligent pricing model. Based on the data, conduct feature importance ranking and recursive feature elimination to screen the features with the greatest impact on freight rates, and perform weighted processing on the features to improve the prediction accuracy of the model.

[0100] (4) Set evaluation metrics to measure the performance of the intelligent pricing model.

[0101] (5) Adopt the method of cross-validation to evaluate the stability and generalization ability of the model. According to the evaluation results, select the optimal model for deployment.

[0102] The intelligent pricing model can provide the following prediction services:

[0103] Logistics transportation service pricing: Through comprehensive analysis of factors such as historical freight data, transportation distance, cargo type, transportation time, and market demand, the intelligent pricing model can predict the transportation service prices under different conditions. This service mainly provides a more reasonable and transparent pricing plan for logistics enterprises, while optimizing their own profit structure.

[0104] Dynamic price adjustment: During the logistics transportation process, various factors such as market supply and demand relationships, traffic conditions, and weather conditions will affect transportation prices. The intelligent pricing model dynamically adjusts the transportation prices based on real-time data and accordingly adjusts the prices to reduce the impact on the enterprise's profits.

[0105] Inventory management optimization: In the field of inventory, through the analysis of factors such as historical sales data, market demand, and inventory costs, the intelligent pricing model can predict the inventory demand in the future for a period of time and help enterprises formulate reasonable inventory strategies, reduce inventory backlogs and out-of-stock risks, and improve inventory turnover and capital utilization rate.

[0106] Supply chain collaborative management: In supply chain management, the intelligent pricing model promotes collaborative cooperation between upstream and downstream enterprises. Through comprehensive analysis of the data in each link of the supply chain, the model can predict the cost differences under different suppliers and different transportation modes and provide the optimal supply chain solution for enterprises, reduce the overall supply chain cost, and improve the stability and competitiveness of the supply chain.

[0107] As is known by common technical knowledge, the present invention can be implemented by other embodiments that do not depart from its spiritual essence or essential features. Therefore, the above-disclosed embodiments are illustrative in all aspects and are not the only ones. All changes within the scope of the present invention or within the scope equivalent to the present invention are encompassed by the present invention.

Claims

1. A method for constructing an intelligent pricing model based on freight big data, characterized by: The following steps are involved: S1. Collect freight data from multiple channels and establish an independent data storage server to store and manage the collected data; S2, preprocessing and cleaning the collected freight data; S3, automatically and manually label the cleaned data; S4. Build and train intelligent pricing models; S5. Use the intelligent pricing model to comprehensively analyze the freight demand information provided by users and predict the freight prices under different conditions; S6. Collect feedback from users of the intelligent pricing model on pricing results, optimize and improve the model based on the feedback results, and improve pricing accuracy and user satisfaction.

2. The method for constructing an intelligent pricing model based on freight big data according to claim 1, characterized in that: In step S1, the freight data includes waybill data and IoT data. The waybill data includes customer order number, freight bill number, mode of transport, transport route, cargo type, cargo weight, cargo volume, carrier vehicle type, energy type, transport mileage, transport time, shipping address, receiving address, transportation service and freight composition. The IoT data includes historical oil prices, seasonal factors, holiday factors, weather conditions, traffic conditions, and regional transportation capacity supply and demand. The waybill data is synchronously stored once every 24 hours, and the IoT data is synchronously stored once every 30 minutes.

3. The method for constructing an intelligent pricing model based on freight big data according to claim 1 is characterized in that: In step S2, the cleaning of freight data includes the following steps: a. The first step is to judge the data integrity: if the number of missing data items exceeds 30%, it is judged as invalid data; b. The second step is to determine the data duplication: Use the unique customer order number or shipping number to detect data duplication, and delete or mark duplicate data according to business needs; c. The third step is to judge the correctness of the data: set a reasonable range for the data items, and determine that the data with more than 30% abnormal data items is invalid.

4. The method for constructing an intelligent pricing model based on freight big data according to claim 1 is characterized in that: In step S3, the automatic labeling judgment rules are as follows: Vehicle length: If the weight of the cargo on the waybill exceeds the vehicle load, it will be marked as abnormal; if the weight of the cargo on the waybill is less than 50% of the vehicle load, it will be marked as abnormal; Vehicle type: If the vehicle load exceeds 30 tons and the vehicle type is not a tractor, the marking is abnormal; Transportation time: If the transportation time does not match the transportation mileage, it will be marked as abnormal; Transport mileage: The mileage cannot be completed within the transport time; Goods type: If a small amount of less than 10% of a certain type of goods appears for the same customer, it will be marked as abnormal; Cargo weight: If the weight of the cargo on the waybill exceeds the vehicle load, it will be marked as abnormal; if the weight of the cargo on the waybill is less than 50% of the vehicle load, it will be marked as abnormal; if the cargo weight is 0 or 1, it will be marked as abnormal; if the vehicle kilometer unit price is abnormal, it will be marked as abnormal; Transportation costs: If the freight rate exceeds the reasonable range of the ton-kilometer unit price, it will be marked as abnormal; if one person collects money multiple times, it will be marked as abnormal; If multiple vehicles receive payment from the same account, the freight may be inflated, which will be marked as abnormal; if the monthly freight income of a single vehicle exceeds 100,000, the freight may be inflated, which will be marked as abnormal; Penalty deduction: The waybill transaction price is inconsistent with the payment amount, and the mark is abnormal; Transport unit price: Determine the reasonable range of freight prices based on the ton-kilometer unit price and vehicle-kilometer unit price.

5. The method for constructing an intelligent pricing model based on freight big data according to claim 1 is characterized in that: In step S3, manual annotation is divided into an annotation management end module, an annotation work end module and an annotation review end module; The annotation management module includes task allocation, progress monitoring, annotation rule setting and annotator management; The annotation work end module is used to accept the tasks assigned by the annotation management end module, use the provided annotation tools to annotate the data, and finally submit the annotation results to the annotation management end module and the annotation review end module; The annotation review module reviews the annotation results submitted by the annotation work module to ensure the accuracy and consistency of the annotations, give feedback and corrections, and compile statistics on the review results to provide a basis for the performance of the annotators; The manual annotation workflow includes the following steps: (1) Determine the annotation requirements, including data type and annotation type; (2) Assign annotation tasks to appropriate annotators; (3) The annotator logs in to the annotation work module and receives the task; (4) Use annotation tools to annotate data according to annotation specifications; (5) After the annotation is completed, submit the annotation results; (6) The auditor logs in to the annotation audit module and receives the annotation tasks to be reviewed; (7) Review the annotation results one by one to ensure the accuracy and consistency of the annotations. If any annotation errors or non-compliance with the specifications are found, provide feedback and correction suggestions; (8) After the review is completed, the review results are submitted to the annotation management module; (9) Based on the audit results, the labelers will be evaluated for quality and rewarded.

6. The method for constructing an intelligent pricing model based on freight big data according to claim 1, characterized in that: In step S4, training the intelligent pricing model includes the following steps: (1) Divide the valid data set into training set, validation set and test set; (2) The intelligent pricing model is trained by batch processing, incremental processing, and automatic online training. Data fusion, statistical framework, and hierarchical model technology are used to train a domain model for each transportation business scenario. (3) Optimize model performance by adjusting the parameters of the intelligent pricing model, sort the features by importance based on the data, perform recursive feature elimination, select the features with the greatest impact on freight rates, and perform weighted processing on the features to improve the prediction accuracy of the model; (4) Set evaluation metrics to measure the performance of the smart pricing model; (5) Use the cross-validation method to evaluate the stability and generalization ability of the model, and select the optimal model for deployment based on the evaluation results.

7. The method for constructing an intelligent pricing model based on freight big data according to claim 5 is characterized by: The data annotation quality assessment rules are as follows: Completion quality score = marking qualified score (maximum score 50) + marking quantity score (maximum score 50); Marked qualified score = number of sampled qualified data / total number of sampled data*50; Number of annotation points = (number of standard fields / number of completed tasks) / 10%*50.

Images