An AI-based intelligent logistics pricing method and system

Abstract

The application discloses an AI-based intelligent logistics pricing method and system, and belongs to the field of logistics information technology. Historical logistics order data, real-time market data, cargo attribute data and transportation environment data are collected, multi-dimensional feature engineering is constructed, and a gradient boosting tree algorithm is used for feature importance screening. The screened effective features are input into a hybrid AI prediction model combining gradient boosting tree and deep neural network for basic freight prediction. A dynamic pricing system containing a hierarchical holiday special factor is introduced to correct the basic freight. Isolation forest anomaly detection and data re-verification mechanism are combined to filter abnormal quotes. Finally, personalized and real-time accurate logistics quotes are generated. The application solves the problems of low efficiency, poor accuracy, and inability to respond to market fluctuations and special holiday needs in traditional logistics pricing, significantly improves the pricing efficiency of logistics enterprises and customer satisfaction, and optimizes enterprise cost control.

Description

Technical Field

[0001] This invention relates to the field of logistics information technology, and in particular to an AI-based intelligent logistics pricing method and system. Background Technology

[0002] With the rapid development of e-commerce, the logistics industry has experienced explosive growth. Logistics pricing, as a core component of logistics services, directly impacts a logistics company's market competitiveness and customer satisfaction. Traditional logistics pricing methods rely heavily on manual experience, which has the following drawbacks: (1) Inefficient: Manual quotation requires processing a large amount of information, which takes a long time and cannot meet the customer's need for real-time quotation; (2) Poor accuracy: Manual quotations are greatly affected by personal experience, and different salespersons may give different quotations. It is also difficult to accurately take into account complex factors such as market fluctuations and weather changes. (3) High cost: It requires a large number of professional quoting personnel, which increases the company's labor costs; (4) Untimely response: Unable to respond in real time to changes in market supply and demand and fluctuations in fuel prices, resulting in a disconnect between quoted prices and actual costs.

[0003] In recent years, with the development of artificial intelligence technology, more and more logistics companies have begun to try using AI technology for automated quoting. However, most existing AI quoting systems have the following shortcomings: (1) Using a single prediction model makes it difficult to fully utilize multi-dimensional data information, resulting in limited prediction accuracy; (2) Lack of effective feature selection mechanism, too many input features lead to slow model training speed and high risk of overfitting; (3) The dynamic pricing mechanism is imperfect, especially lacking a tiered pricing system for holidays, which cannot accurately reflect the actual situation of tight transportation capacity and rising labor costs during holidays. (4) Lack of anomaly detection and data re-verification mechanism. When data collection is incorrect, unreasonable quotes will be generated, affecting customer experience. Summary of the Invention

[0004] The purpose of this invention is to propose an AI-based intelligent logistics pricing method and system to overcome the problems of low efficiency, poor accuracy, and inability to respond to market changes in the existing logistics pricing technology.

[0005] To achieve the above objectives, this invention provides an AI-based intelligent logistics pricing method, comprising the following steps: Step S1: Data collection and preprocessing. Collect historical logistics order data, real-time market data, cargo attribute data, and transportation environment data, and clean, normalize, and extract features from the collected raw data. Step S2: Feature importance screening. The gradient boosting tree algorithm is used to calculate the importance score of all extracted features, and the effective features with an importance score greater than the preset threshold are selected. Step S3: Construct a hybrid AI prediction model by inputting the selected effective features into a hybrid model that integrates gradient boosting trees and deep neural networks, and train it to obtain a basic freight prediction model. Step S4: Calculate the dynamic price adjustment factor. Based on the real-time market supply and demand relationship, fuel price fluctuations, the degree of weather impact, transportation route congestion, and special factors such as holidays, calculate the comprehensive dynamic price adjustment coefficient. Step S5: Basic freight forecast and correction. The basic freight is output using the trained basic freight forecast model. The basic freight is then corrected using the comprehensive dynamic price adjustment coefficient to obtain a preliminary quote. Step S6: Anomaly Quotation Detection and Optimization. The Isolation Forest algorithm is used to detect anomalies in the preliminary quotations. After filtering out anomalies, the final logistics quotation is generated and returned to the user.

[0006] Preferably, in step S1, the various types of data collected specifically include: Historical logistics order data: order number, origin, destination, cargo weight, cargo volume, mode of transport, transport distance, transport time, actual freight cost, and order completion status; Real-time market data: recent transaction prices on the same route, competitor quotes, supply and demand ratio of transport capacity, and real-time fuel prices; Cargo attribute data: cargo type, fragility, hazard class, insulation requirements, loading and unloading difficulty level; Transportation environment data: real-time weather conditions, road congestion index, traffic control information, and holiday information.

[0007] Preferably, in step S1, the feature extraction process specifically includes: Categorical features are transformed using a combination of one-hot encoding and label encoding, while numerical features are standardized using the following formula: ; in, These are the standardized feature values. These are the original eigenvalues. This is the mean of the feature in the training set. The standard deviation of this feature in the training set; Simultaneously, cross-features are constructed, including cross-features of transportation distance and cargo weight, and cross-features of transportation mode and cargo type.

[0008] Preferably, in step S2, the formula for calculating the feature importance score is: ; in, Features Importance score To increase the number of gradient boosting trees, For the first The set of all nodes of the tree For nodes Splitting gain, For indicator functions, when node Use features The value is 1 when splitting, otherwise it is 0; Preset threshold The value range is [0.005, 0.01], and the importance score is selected. The features are used as valid features.

[0009] Preferably, in step S3, the constructed hybrid AI prediction model specifically includes: Step S31: The gradient boosting tree module is used to extract the nonlinear features of the structured data and output the first prediction result; Step S32: The deep neural network module is used to extract the implicit relationships of high-dimensional sparse features and output the second prediction result; Step S33: The prediction results of the two modules are merged using a weighted fusion method to obtain the final basic freight forecast value. The fusion formula is as follows: ; in, This is the base freight forecast after integration. The prediction results of the gradient boosting tree module, The prediction results are from the deep neural network module. The weighting coefficient has a value range of [0,1] and the optimal value is determined by grid search.

[0010] Preferably, in step S4, the formula for calculating the dynamic price adjustment coefficient is: ; in, To incorporate the dynamic price adjustment coefficient, This is the supply and demand adjustment coefficient. This is the fuel price adjustment coefficient. This is the weather impact coefficient. The traffic congestion coefficient. Special price adjustment coefficients for holidays.

[0011] The preferred method for calculating the coefficients of each component is as follows: Supply and demand adjustment coefficient ,in The available capacity for the current route. This represents the current order demand for the route. This is the supply and demand sensitivity coefficient; Fuel price adjustment coefficient ,in Based on current fuel prices, As the benchmark fuel price, Fuel sensitivity coefficient; Weather Influence Coefficient According to the weather rating system, sunny is 1.0, light rain is 1.05, moderate rain is 1.1, heavy rain is 1.2, and blizzard is 1.5. Traffic congestion coefficient According to the congestion index, smooth traffic is 1.0, slow traffic is 1.05, congestion is 1.15, and severe congestion is 1.3. Special price adjustment coefficient for holidays The rates are set according to the type of holiday and the number of days in advance for transportation: 1.2 for statutory long holidays, 1.4 for 3-7 days in advance, and 1.6 for less than 3 days in advance; 1.1 for ordinary holidays, 1.25 for 3-7 days in advance, and 1.4 for less than 3 days in advance; and 1.0 for non-holidays and weekends.

[0012] Preferably, in step S5, the formula for calculating the preliminary quote is: ; in, This is a preliminary quote. Based on the basic freight forecast, To incorporate the dynamic price adjustment coefficient, Additional fees include loading and unloading fees, insurance fees, upstairs delivery fees, and express delivery fees.

[0013] Preferably, in step S6, the specific steps for anomaly detection using the isolated forest algorithm are as follows: Step S61: Construct containing A forest of isolated trees, where each tree randomly selects features and split points to divide the data; Step S62: Calculate the average path length of each sample point across all isolated trees. ; Step S63: Calculate the anomaly score based on the average path length. The calculation formula is: ; in, For the sample Average path length across all isolated trees For a given sample size The expected average path length at that time; Step S64: When abnormal scores If the price is deemed abnormal, it will be filtered and recalculated.

[0014] An AI-based intelligent logistics pricing system includes: The data acquisition module is used to collect historical logistics order data, real-time market data, cargo attribute data, and transportation environment data. The data preprocessing module, connected to the data acquisition module, is used to clean, normalize, and extract features from the raw data. The feature selection module, connected to the data preprocessing module, is used to calculate feature importance and select effective features using the gradient boosting tree algorithm. The model training and prediction module, connected to the feature selection module, is used to build and train a hybrid AI prediction model that integrates gradient boosting trees and deep neural networks, and outputs basic freight prediction values. The dynamic pricing module, connected to the model training and prediction module, is used to calculate the comprehensive dynamic pricing coefficient, which includes special factors related to holidays, and to adjust the basic freight cost to obtain a preliminary quote. The anomaly detection module, connected to the dynamic pricing module, is used to perform anomaly detection on the preliminary price using the isolated forest algorithm and trigger a data re-verification mechanism. The quotation generation and display module, connected to the anomaly detection module, is used to generate the final logistics quotation and display it to the user; The closed-loop optimization module is connected to the quotation generation and display module and the model training and prediction module, respectively. It is used to collect user feedback information and periodically use new order data to incrementally train the model.

[0015] Therefore, the AI-based intelligent logistics pricing method and system described above have the following advantages: (1) The present invention adopts a hybrid AI prediction model that integrates gradient boosting tree and deep neural network, which can make full use of the feature information of structured data and high-dimensional sparse data, and significantly improve the accuracy of basic freight prediction. (2) The present invention adds a feature importance screening step based on gradient boosting tree before training the hybrid AI prediction model, which can effectively remove redundant features, shorten the model training time, and reduce the average absolute error of prediction. (3) This invention introduces a tiered special price adjustment coefficient for holidays, which is dynamically adjusted according to the impact of different holidays and the advance booking time, thereby improving the accuracy of quotations during holidays; (4) This invention uses the isolated forest algorithm to detect abnormal quotes and adds a data re-verification mechanism, which can effectively filter unreasonable quotes caused by data errors and improve the reliability of quotes; (5) This invention realizes the full automation of logistics quotation, greatly improves quotation efficiency, shortens the response time from hours to seconds, and reduces the labor costs of enterprises; (6) The present invention has a closed-loop optimization module, which can continuously optimize the model performance and enable the quotation system to adapt to the ever-changing market environment.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an AI-based intelligent logistics pricing method according to an embodiment of the present invention. Figure 2 This is a structural block diagram of an AI-based intelligent logistics quotation system according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0019] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0020] Example 1 like Figure 1-2 As shown in the figure, this embodiment proposes an AI-based intelligent logistics pricing method, and the specific implementation steps are as follows: Step S1: Data collection and preprocessing. Collect historical logistics order data, real-time market data, cargo attribute data, and transportation environment data, and clean, normalize, and extract features from the collected raw data.

[0021] The various types of data collected specifically include: Historical logistics order data: order number, origin, destination, cargo weight, cargo volume, mode of transport, transport distance, transport time, actual freight cost, and order completion status; Real-time market data: recent transaction prices on the same route, competitor quotes, supply and demand ratio of transport capacity, and real-time fuel prices; Cargo attribute data: cargo type, fragility, hazard class, insulation requirements, loading and unloading difficulty level; Transportation environment data: real-time weather conditions, road congestion index, traffic control information, and holiday information.

[0022] Feature extraction processing specifically includes: Categorical features are transformed using a combination of one-hot encoding and label encoding, while numerical features are standardized using the following formula: ; in, These are the standardized feature values. These are the original eigenvalues. This is the mean of the feature in the training set. The standard deviation of this feature in the training set; Simultaneously, cross-features are constructed, including cross-features of transportation distance and cargo weight, and cross-features of transportation mode and cargo type.

[0023] Step S2: Feature importance screening. The gradient boosting tree algorithm is used to calculate the importance score of all extracted features, and the effective features with an importance score greater than the preset threshold are selected. The formula for calculating feature importance score is: ; in, Features Importance score To increase the number of gradient boosting trees, For the first The set of all nodes of the tree For nodes Splitting gain, For indicator functions, when node Use features The value is 1 when splitting, otherwise it is 0; Preset threshold The value range is [0.005, 0.01], and the importance score is selected. The features are used as valid features.

[0024] Step S3: Construct a hybrid AI prediction model by inputting the processed multi-dimensional features into a hybrid model that integrates gradient boosting trees and deep neural networks, and train it to obtain a basic freight prediction model.

[0025] The constructed hybrid AI prediction model specifically includes: Step S31: The gradient boosting tree module is used to extract the nonlinear features of the structured data and output the first prediction result; Step S32: The deep neural network module is used to extract the implicit relationships of high-dimensional sparse features and output the second prediction result; Step S33: The prediction results of the two modules are merged using a weighted fusion method to obtain the final basic freight forecast value. The fusion formula is as follows: ; in, This is the base freight forecast after integration. The prediction results of the gradient boosting tree module, The prediction results are from the deep neural network module. The weighting coefficient has a value range of [0,1] and the optimal value is determined by grid search.

[0026] Step S4: Calculate the dynamic price adjustment factor. Based on the real-time market supply and demand relationship, fuel price fluctuations, the degree of weather impact, transportation route congestion, and special factors such as holidays, calculate the comprehensive dynamic price adjustment coefficient. The formula for calculating the dynamic price adjustment coefficient is: ; in, To incorporate the dynamic price adjustment coefficient, This is the supply and demand adjustment coefficient. This is the fuel price adjustment coefficient. This is the weather impact coefficient. The traffic congestion coefficient. Special price adjustment coefficients for holidays.

[0027] The calculation method for each partial factor is as follows: Supply and demand adjustment coefficient ,in The available capacity for the current route. This represents the current order demand for the route. This is the supply and demand sensitivity coefficient; Fuel price adjustment coefficient ,in Based on current fuel prices, As the benchmark fuel price, Fuel sensitivity coefficient; Weather Influence Coefficient According to the weather rating system, sunny is 1.0, light rain is 1.05, moderate rain is 1.1, heavy rain is 1.2, and blizzard is 1.5. Traffic congestion coefficient According to the congestion index, smooth traffic is 1.0, slow traffic is 1.05, congestion is 1.15, and severe congestion is 1.3. Special price adjustment coefficient for holidays The rates are tiered based on the type of holiday and the number of days in advance for transportation: For statutory long holidays (Spring Festival, National Day), the rate is 1.2 for more than 7 days in advance, 1.4 for 3-7 days in advance, and 1.6 for less than 3 days in advance; For ordinary holidays (New Year's Day, Qingming Festival, May Day, Dragon Boat Festival, Mid-Autumn Festival), the rate is 1.1 for more than 7 days in advance, 1.25 for 3-7 days in advance, and 1.4 for less than 3 days in advance; and 1.0 for non-holidays and weekends.

[0028] Step S5: Basic freight forecasting and correction. The basic freight is output using the trained basic freight forecasting model. The basic freight is then corrected using the comprehensive dynamic price adjustment coefficient to obtain a preliminary quote.

[0029] The formula for calculating the preliminary quote is: ; in, This is a preliminary quote. Based on the basic freight forecast, To incorporate the dynamic price adjustment coefficient, Additional fees include loading and unloading fees, insurance fees, upstairs delivery fees, and express delivery fees.

[0030] Step S6: Anomaly Quotation Detection and Optimization. The Isolation Forest algorithm is used to detect anomalies in the preliminary quotations. After filtering out anomalies, the final logistics quotation is generated and returned to the user.

[0031] The specific steps for anomaly detection using the Isolation Forest algorithm are as follows: Step S61: Construct containing A forest of isolated trees, where each tree randomly selects features and split points to divide the data; Step S62: Calculate the average path length of each sample point across all isolated trees. ; Step S63: Calculate the anomaly score based on the average path length. The calculation formula is: ; in, For the sample Average path length across all isolated trees For a given sample size The expected average path length at that time; Step S64: When abnormal scores If the price is deemed abnormal, it will be filtered and recalculated.

[0032] An AI-based intelligent logistics pricing system includes: The data acquisition module is used to collect historical logistics order data, real-time market data, cargo attribute data, and transportation environment data. The data preprocessing module, connected to the data acquisition module, is used to clean, normalize, and extract features from the raw data. The feature selection module, connected to the data preprocessing module, is used to calculate feature importance and select effective features using the gradient boosting tree algorithm. The model training and prediction module, connected to the feature selection module, is used to build and train a hybrid AI prediction model that integrates gradient boosting trees and deep neural networks, and outputs basic freight prediction values. The dynamic pricing module, connected to the model training and prediction module, is used to calculate the comprehensive dynamic pricing coefficient, which includes special factors related to holidays, and to adjust the basic freight cost to obtain a preliminary quote. The anomaly detection module, connected to the dynamic pricing module, is used to perform anomaly detection on the preliminary price using the isolated forest algorithm and trigger a data re-verification mechanism. The quotation generation and display module, connected to the anomaly detection module, is used to generate the final logistics quotation and display it to the user; The closed-loop optimization module is connected to the quotation generation and display module and the model training and prediction module, respectively. It is used to collect user feedback information and periodically use new order data to incrementally train the model.

[0033] One specific implementation method is as follows: This embodiment provides an AI-based intelligent logistics pricing method, with the following specific steps: S1: Data Acquisition and Preprocessing We collected 1 million historical logistics order records from the past three years from the internal databases of logistics companies; we collected real-time market data through third-party data interfaces, including recent transaction prices on the same routes, competitor quotations, supply and demand ratios of transport capacity, and real-time fuel prices; we obtained cargo attribute data through user input, including cargo type, fragility, hazard level, insulation requirements, and loading and unloading difficulty coefficient; and we collected transportation environment data through meteorological APIs and traffic APIs, including real-time weather conditions, road congestion index, traffic control information, and holiday information.

[0034] The collected raw data is cleaned, records with more than 30% missing values ​​are deleted, and records with fewer missing values ​​are filled with the mean or median; numerical features are standardized, and categorical features are coded; cross features are constructed, such as the product of transportation distance and cargo weight, and the combination of transportation mode and cargo type.

[0035] S2: Feature Importance Screening The gradient boosting tree module is used to calculate the importance scores of all extracted features, and a preset threshold is set. Features with an importance score greater than 0.008 were selected as the final input features, and redundant features were removed. Among them, the top 5 features in terms of importance score were transportation distance, cargo weight, transportation mode, supply-demand ratio, and fuel price, which is consistent with actual business experience.

[0036] S3: Building a Hybrid AI Prediction Model The processed dataset was divided into training and testing sets in an 8:2 ratio. A gradient boosting tree module was constructed, with 100 trees and a maximum depth of 6. A deep neural network module was also constructed, containing three hidden layers with 128, 64, and 32 neurons in each layer, respectively. The ReLU activation function was used, and the output layer used a linear activation function.

[0037] A weighted fusion method is used to combine the prediction results of the two modules, and the optimal weight coefficients are determined by a grid search method. The mean absolute error (MAE) was 0.6. The hybrid model was trained using the training set and its performance was evaluated using the test set. The model's mean absolute error (MAE) was 12.5 and its root mean square error (RMSE) was 18.3, which is better than both the single gradient boosting tree model and the deep neural network model.

[0038] S4: Calculation of dynamic price adjustment factor: Calculate the price adjustment coefficients for each item based on real-time data: Supply and demand adjustment coefficient: The current available capacity S for the route is 50 vehicles, and the order demand D is 40 orders. Supply and demand sensitivity coefficient. If it is 0.5, then ; Fuel price adjustment coefficient: Current fuel price The benchmark fuel price is 7.5 yuan per liter. The price is 7.0 yuan / liter, and the fuel sensitivity coefficient is... If it is 0.3, then ; Weather impact coefficient: If the weather is moderate rain, then... ; Traffic congestion coefficient: If the road congestion index is congested, then... ; Special price adjustment factor for holidays: For non-holidays, then... ; Comprehensive dynamic price adjustment coefficient .

[0039] S5: Basic Freight Forecast and Revision Output basic freight forecast values ​​using a trained hybrid AI prediction model It costs 200 yuan, plus additional fees. The initial quote is 20 yuan (including 10 yuan for loading and unloading and 10 yuan for insurance). Yuan.

[0040] S5: Abnormal Quotation Detection and Optimization: The Isolation Forest algorithm was used to detect anomalies in the initial quote. The anomaly score of the quote was calculated to be 0.32, which is less than 0.6. Therefore, the quote was judged to be a normal quote, and a final logistics quote of 311 yuan was generated and returned to the user.

[0041] Example 2 This embodiment demonstrates the process of using the method of the present invention to quote logistics prices during the May Day holiday, highlighting the application effects of feature importance screening and special holiday price adjustments.

[0042] S1: Data Acquisition and Preprocessing We collected 1.5 million historical logistics order data entries from the past four years from the internal databases of logistics companies, including 300,000 orders placed during holidays; we also collected real-time market data through third-party data interfaces, including recent transaction prices for the same routes, competitor quotes, supply-demand ratios for transport capacity, and real-time fuel prices; we obtained cargo attribute data through user input: cargo type is general cargo, weight is 500kg, volume is 2 cubic meters, and there are no special transport requirements; and we collected transport environment data through meteorological and traffic APIs: the weather is sunny, and the road congestion index is slow.

[0043] The collected raw data was cleaned, records with more than 30% missing values ​​were deleted, and records with fewer missing values ​​were filled with the mean or median. Numerical features were standardized and categorical features were coded. Cross features were constructed, such as the product of transportation distance and cargo weight, and the combination of transportation mode and cargo type, resulting in a total of 62 initial features.

[0044] S2: Feature Importance Screening The importance scores of all 62 features are calculated using the gradient boosting tree module, and a preset threshold is set. We selected 38 features with an importance score greater than 0.008 as the final input features and removed 24 redundant features.

[0045] S3: Hybrid Model Training The selected features were divided into training and testing sets in an 8:2 ratio. A gradient boosting tree module was constructed, with 120 trees and a maximum depth of 7. A deep neural network module was also constructed, containing three hidden layers with 128, 64, and 32 neurons in each layer, respectively. The ReLU activation function was used, and the output layer used a linear activation function.

[0046] A weighted fusion method is used to combine the prediction results of the two modules, and the optimal weight coefficients are determined by a grid search method. The mean absolute error (MAE) was 0.65. The hybrid model was trained using the training set and its performance was evaluated using the test set. The model's mean absolute error (MAE) was 10.2 and its root mean square error (RMSE) was 14.8, which is better than the hybrid model without feature selection (MAE=12.1, RMSE=17.6). At the same time, the model training time was reduced by 36%.

[0047] S4: Calculation of dynamic price adjustment factor: The current date is April 28th, and the customer requests delivery on May 1st, which falls within 3 days before the May Day holiday. Therefore, a special holiday price adjustment factor applies. .

[0048] Other itemized price adjustment coefficients are calculated as follows: Supply and demand adjustment coefficient: The current available capacity S for the route is 45 vehicles, and the order demand D is 60 orders. Supply and demand sensitivity coefficient. If it is 0.5, then ; Fuel price adjustment coefficient: Current fuel price The benchmark fuel price is 7.8 yuan per liter. The price is 7.0 yuan / liter, and the fuel sensitivity coefficient is... If it is 0.3, then ; Weather impact coefficient: If the weather is sunny, then ; Traffic congestion index: If the road congestion index is slow, then... ; Comprehensive dynamic price adjustment coefficient .

[0049] S5: Basic Freight Forecast and Revision The trained hybrid AI prediction model outputs a base freight cost prediction F of 280 yuan, plus additional fees. The initial quote is 30 yuan (including 15 yuan for loading and unloading and 15 yuan for insurance). Yuan.

[0050] In comparison, without considering special price adjustment factors for holidays, the initial price is... The quoted price of 403 yuan was 110 yuan lower than the actual market average price of 405 yuan for the same route at the time, representing an error of 27%. However, the quoted price of 403 yuan after taking into account the special price adjustment factor for holidays was only 2 yuan lower than the market average price, with an error of less than 0.5%, significantly improving the accuracy.

[0051] S6: Abnormal Quotation Detection and Optimization: The Isolation Forest algorithm was used to detect anomalies in the initial quote. The anomaly score of the quote was calculated to be 0.22, which is less than 0.6. Therefore, the quote was judged to be a normal quote, and a final logistics quote of 403 yuan was generated and returned to the user.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An AI-based intelligent logistics pricing method, characterized in that, Includes the following steps: Step S1: Data collection and preprocessing. Collect historical logistics order data, real-time market data, cargo attribute data, and transportation environment data, and clean, normalize, and extract features from the collected raw data. Step S2: Feature importance screening. The gradient boosting tree algorithm is used to calculate the importance score of all extracted features and screen out the valid features whose importance score is greater than the preset threshold. Step S3: Construct a hybrid AI prediction model by inputting the selected effective features into a hybrid model that integrates gradient boosting trees and deep neural networks, and train it to obtain a basic freight prediction model. Step S4: Calculate the dynamic price adjustment factor. Based on the real-time market supply and demand relationship, fuel price fluctuations, the degree of weather impact, transportation route congestion, and special factors such as holidays, calculate the comprehensive dynamic price adjustment coefficient. Step S5: Basic freight forecast and correction. The basic freight is output using the trained basic freight forecast model. The basic freight is then corrected using the comprehensive dynamic price adjustment coefficient to obtain a preliminary quote. Step S6: Anomaly Quotation Detection and Optimization. The Isolation Forest algorithm is used to detect anomalies in the preliminary quotations. After filtering out anomalies, the final logistics quotation is generated and returned to the user.

2. The AI-based intelligent logistics pricing method according to claim 1, characterized in that: In step S1, the various types of data collected specifically include: Historical logistics order data: order number, origin, destination, cargo weight, cargo volume, mode of transport, transport distance, transport time, actual freight cost, and order completion status; Real-time market data: recent transaction prices on the same route, competitor quotes, supply and demand ratio of transport capacity, and real-time fuel prices; Cargo attribute data: cargo type, fragility, hazard class, insulation requirements, loading and unloading difficulty level; Transportation environment data: real-time weather conditions, road congestion index, traffic control information, and holiday information.

3. The AI-based intelligent logistics pricing method according to claim 1, characterized in that: In step S1, the feature extraction process specifically includes: Categorical features are transformed using a combination of one-hot encoding and label encoding, while numerical features are standardized using the following formula: ； in, These are the standardized feature values. These are the original eigenvalues. This is the mean of the feature in the training set. The standard deviation of this feature in the training set; Simultaneously, cross-features are constructed, including cross-features of transportation distance and cargo weight, and cross-features of transportation mode and cargo type.

4. The AI-based intelligent logistics pricing method according to claim 1, characterized in that: In step S2, the formula for calculating the feature importance score is: ； in, Features Importance score To increase the number of gradient boosting trees, For the first The set of all nodes of the tree For nodes Splitting gain, For indicator functions, when node Use features The value is 1 when splitting, otherwise it is 0; Preset threshold The value range is [0.005, 0.01], and the importance score is selected. The features are used as valid features.

5. The AI-based intelligent logistics pricing method according to claim 1, characterized in that: In step S3, the constructed hybrid AI prediction model specifically includes: Step S31: The gradient boosting tree module is used to extract the nonlinear features of the structured data and output the first prediction result; Step S32: The deep neural network module is used to extract the implicit relationships of high-dimensional sparse features and output the second prediction result; Step S33: The prediction results of the two modules are merged using a weighted fusion method to obtain the final basic freight forecast value. The fusion formula is as follows: ； in, This is the base freight forecast after integration. The prediction results of the gradient boosting tree module, The prediction results are from the deep neural network module. The weighting coefficient has a value range of [0,1] and the optimal value is determined by grid search.

6. The AI-based intelligent logistics pricing method according to claim 1, characterized in that: In step S4, the formula for calculating the dynamic price adjustment coefficient is: ； in, To incorporate the dynamic price adjustment coefficient, This is the supply and demand adjustment coefficient. This is the fuel price adjustment coefficient. This is the weather impact coefficient. The traffic congestion coefficient. Special price adjustment coefficients for holidays.

7. The AI-based intelligent logistics pricing method according to claim 1, characterized in that: The calculation method for each partial factor is as follows: Supply and demand adjustment coefficient ,in The available capacity for the current route. This represents the current order demand for the route. This is the supply and demand sensitivity coefficient; Fuel price adjustment coefficient ,in Based on current fuel prices, As the benchmark fuel price, Fuel sensitivity coefficient; Weather Influence Coefficient According to the weather rating system, sunny is 1.0, light rain is 1.05, moderate rain is 1.1, heavy rain is 1.2, and blizzard is 1.

5. Traffic congestion coefficient According to the congestion index, smooth traffic is 1.0, slow traffic is 1.05, congestion is 1.15, and severe congestion is 1.

3. Special price adjustment coefficient for holidays The rates are tiered based on the type of holiday and the number of days in advance for transportation: 1.2 for statutory long holidays (more than 7 days in advance), 1.4 for 3-7 days in advance, and 1.6 for less than 3 days in advance; 1.1 for ordinary holidays (more than 7 days in advance), 1.25 for 3-7 days in advance, and 1.4 for less than 3 days in advance; and 1.0 for non-holidays and weekends.

8. The AI-based intelligent logistics pricing method according to claim 1, characterized in that: In step S5, the formula for calculating the preliminary quote is: ； in, This is a preliminary quote. Based on the basic freight forecast, To incorporate the dynamic price adjustment coefficient, Additional fees include loading and unloading fees, insurance fees, upstairs delivery fees, and express delivery fees.

9. The AI-based intelligent logistics pricing method according to claim 1, characterized in that: In step S6, the specific steps for anomaly detection using the isolated forest algorithm are as follows: Step S61: Construct containing A forest of isolated trees, where each tree randomly selects features and split points to divide the data; Step S62: Calculate the average path length of each sample point across all isolated trees. ; Step S63: Calculate the anomaly score based on the average path length. The calculation formula is: ； in, For the sample Average path length across all isolated trees For a given sample size The expected average path length at that time; Step S64: When abnormal scores If the price is deemed abnormal, it will be filtered and recalculated.

10. An AI-based intelligent logistics pricing system, used to implement the AI-based intelligent logistics pricing method according to any one of claims 1-9, characterized in that, include: The data acquisition module is used to collect historical logistics order data, real-time market data, cargo attribute data, and transportation environment data. The data preprocessing module, connected to the data acquisition module, is used to clean, normalize, and extract features from the raw data. The feature selection module, connected to the data preprocessing module, is used to calculate feature importance and select effective features using the gradient boosting tree algorithm. The model training and prediction module, connected to the feature selection module, is used to build and train a hybrid AI prediction model that integrates gradient boosting trees and deep neural networks, and outputs basic freight prediction values. The dynamic pricing module, connected to the model training and prediction module, is used to calculate the comprehensive dynamic pricing coefficient, which includes special factors related to holidays, and to adjust the basic freight cost to obtain a preliminary quote. The anomaly detection module, connected to the dynamic pricing module, is used to perform anomaly detection on the preliminary price using the isolated forest algorithm and trigger a data re-verification mechanism. The quotation generation and display module, connected to the anomaly detection module, is used to generate the final logistics quotation and display it to the user; The closed-loop optimization module is connected to the quotation generation and display module and the model training and prediction module, respectively. It is used to collect user feedback information and periodically use new order data to incrementally train the model.

Images