Logistics price dynamic pricing method, apparatus and device, and storage medium

Abstract

The invention relates to the field of logistics, and discloses a logistics price dynamic pricing method, device and equipment and a storage medium, and the method is used for carrying out dynamic pricing on waybill prices. The method comprises the following steps: acquiring and preprocessing historical financial cost data, historical transportation operation data and historical market supply and demand data to obtain preprocessed data; performing feature engineering processing on the preprocessed data to obtain a plurality of key indexes related to the freight rate, and performing analysis to obtain association scores among the plurality of key indexes; based on the plurality of key indexes and the association scores among the plurality of key indexes, constructing an intelligent decision model based on a graph neural network; acquiring real-time financial cost data, real-time transportation operation data and real-time market supply and demand data, and performing prediction by using the intelligent decision model to obtain a cost prediction result and a demand prediction result; and generating a dynamic price in combination with the cost prediction result, the demand prediction result and a preset service target.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of logistics, and in particular to a logistics price dynamic pricing method, device, equipment and storage medium. BACKGROUND

[0002] In the rapid development of the logistics industry, the traditional pricing model has been difficult to adapt to market demand. The traditional logistics pricing mostly adopts static strategy, taking fixed cost addition or manual experience as the core basis, which has significant limitations: on the one hand, the logistics cost is dynamically fluctuated by factors such as fuel price, labor cost, warehouse rent, etc., and the static pricing is prone to cost coverage deficiency or overpricing; on the other hand, the market supply and demand relationship changes instantaneously, and the static pricing cannot match the supply and demand in real time, which may cause high empty running rate of transport capacity or insufficient order acceptance capacity. In addition, the traditional method only utilizes data in a single dimension, without mining the correlation value of multi-dimensional data such as finance, operation and supply and demand, resulting in low cost and demand prediction accuracy and lack of scientific support for pricing decision-making.

[0003] Therefore, the prior art still needs to be improved and developed. SUMMARY

[0004] The present application provides a logistics price dynamic pricing method, device, equipment and storage medium for dynamically pricing the shipping price.

[0005] The first aspect of the present application provides a logistics price dynamic pricing method, comprising: obtaining and preprocessing historical financial cost data, historical transportation operation data and historical market supply and demand data to obtain preprocessed data; performing feature engineering processing on the preprocessed data to obtain a plurality of key indicators related to shipping price, and analyzing to obtain correlation scores between the plurality of key indicators; constructing an intelligent decision-making model based on graph neural network based on the plurality of key indicators and the correlation scores between the plurality of key indicators; obtaining real-time financial cost data, real-time transportation operation data and real-time market supply and demand data, and inputting the real-time financial cost data, real-time transportation operation data and real-time market supply and demand data into the intelligent decision-making model to obtain cost prediction results and demand prediction results; combining the cost prediction results, the demand prediction results and a preset business target to generate a dynamic price.

[0006] Optionally, in the first implementation manner of the first aspect of the present application, the obtaining and preprocessing of the historical financial cost data, the historical transportation operation data and the historical market supply and demand data to obtain the preprocessed data comprises: obtaining and integrating the historical financial cost data, the historical transportation operation data and the historical market supply and demand data to obtain integrated data; performing data cleaning on the integrated data to obtain cleaned data; and performing standardization processing on the cleaned data to obtain the preprocessed data.

[0007] Optionally, in the second implementation form of the first aspect of the present application, the feature engineering processing on the preprocessed data to obtain a plurality of key indicators related to the freight rate and analyzing the correlation scores between the plurality of key indicators comprises: extracting effective features from the preprocessed data, and calculating the correlation coefficients between the effective features and the freight rate, and selecting the effective features with the correlation coefficients greater than a preset coefficient threshold as candidate features; performing importance sorting on the candidate features based on the Gini coefficient importance in the random forest algorithm, and selecting the top 10 candidate features as key indicators; and analyzing the correlation scores between the key indicators by the method of calculating the correlation coefficients.

[0008] Optionally, in the third implementation form of the first aspect of the present application, the method of analyzing the correlation scores between the key indicators by calculating the correlation coefficients comprises: identifying the linear correlation between the key indicators by using the Pearson correlation coefficient to obtain a linear correlation coefficient; identifying the monotonic nonlinear correlation between the key indicators by using the Spearman correlation coefficient to obtain a monotonic nonlinear correlation coefficient; and identifying the non-monotonic nonlinear correlation between the key indicators by using the distance correlation coefficient to obtain a non-monotonic nonlinear correlation coefficient; and fusing the linear correlation coefficient, the monotonic nonlinear correlation coefficient and the non-monotonic nonlinear correlation coefficient according to a preset weight to obtain the correlation scores between the key indicators.

[0009] Optionally, in the fourth implementation form of the first aspect of the present application, the method of constructing the intelligent decision-making model based on the graph neural network based on the plurality of key indicators and the correlation scores between the plurality of key indicators comprises: constructing an initial graph structure by taking the key indicators as node features of the graph and taking the correlation scores between the key indicators as edge weights between the nodes; building an initial model based on the graph attention network, the initial model comprising an input layer, a hidden layer and an output layer, the input layer being configured to receive the plurality of node features and the edge weights, the hidden layer being a multi-layer graph neural network structure, the hidden layer being configured to collect neighbor node information through a message passing mechanism according to the plurality of node features and the edge weights, and the output layer being configured to generate a cost prediction result and a demand prediction result according to the neighbor node information output by the hidden layer; generating a historical graph sample based on historical financial cost data, historical transportation operation data and historical market supply and demand data, and training and optimizing the initial model based on the historical graph sample and the initial graph structure to obtain the intelligent decision-making model, and in the training process, introducing an attention mechanism to dynamically adjust the influence of the neighbor nodes.

[0010] Optionally, in a fifth implementation form of the first aspect of the present application, the obtaining real-time financial cost data, real-time transportation operation data and real-time market supply and demand data, and inputting the real-time financial cost data, the real-time transportation operation data and the real-time market supply and demand data into the intelligent decision model to obtain the cost prediction result and the demand prediction result comprises: obtaining the real-time financial cost data, the real-time transportation operation data and the real-time market supply and demand data, and preprocessing and feature engineering the real-time financial cost data, the real-time transportation operation data and the real-time market supply and demand data to obtain to-be-predicted data; combining the to-be-predicted data with a pre-defined correlation graph structure to construct a real-time graph data instance; inputting the real-time graph data instance into the intelligent decision model to obtain the cost prediction result and the demand prediction result.

[0011] Optionally, in a sixth implementation form of the first aspect of the present application, the combining the cost prediction result, the demand prediction result and a preset business target to generate a dynamic price comprises: obtaining the preset business target, the preset business target being profit maximization; converting the profit maximization into a business target function and a business condition, and inputting the cost prediction result and the demand prediction result into the business target function, and solving an optimal price meeting the constraint condition using an optimization algorithm; correcting the optimal price according to a preset business rule to generate the dynamic price.

[0012] The second aspect of the present application provides a logistics price dynamic pricing device, comprising: a preprocessing module configured to obtain and preprocess historical financial cost data, historical transportation operation data and historical market supply and demand data to obtain preprocessed data; an analysis module configured to perform feature engineering processing on the preprocessed data to obtain a plurality of key indicators related to a freight rate, and analyze to obtain correlation scores between the plurality of key indicators; a construction module configured to construct an intelligent decision model based on graph neural network based on the plurality of key indicators and the correlation scores between the plurality of key indicators; a prediction module configured to obtain real-time financial cost data, real-time transportation operation data and real-time market supply and demand data, and input the real-time financial cost data, the real-time transportation operation data and the real-time market supply and demand data into the intelligent decision model to obtain a cost prediction result and a demand prediction result; and a generation module configured to combine the cost prediction result, the demand prediction result and a preset business target to generate a dynamic price.

[0013] The third aspect of the present application provides a logistics price dynamic pricing device, comprising: a memory and at least one processor, the memory storing computer readable instructions, and the memory and the at least one processor being interconnected through a circuit; the at least one processor invoking the computer readable instructions in the memory to make the logistics price dynamic pricing device perform each step of the logistics price dynamic pricing method as described above.

[0014] The fourth aspect of the present application provides a computer readable storage medium, wherein computer readable instructions are stored in the computer readable storage medium, and when the computer readable instructions are run on a computer, the computer is caused to perform the steps of the logistics price dynamic pricing method as described above.

[0015] In the technical solution provided by the present application, historical financial costs, transportation operation and market supply and demand data are integrated, cost prediction and demand prediction are performed through an intelligent decision-making model based on a graph neural network, and a dynamic price is generated in combination with the cost prediction result, the demand prediction result and a preset business target, so as to avoid pricing deviation caused by a single data source and ensure the comprehensiveness of the basis for pricing. Moreover, the feature engineering link not only accurately extracts key indicators, but also quantifies the mutual influence between indicators through correlation analysis to provide a clear relationship framework for model construction. The intelligent decision-making model based on the graph neural network innovatively converts the correlation between indicators into a graph structure, which is more adaptable to the complex and dynamic changes of the logistics market than traditional regression models or neural networks, and can significantly improve the accuracy of cost and demand prediction. In addition, real-time data processing and a dynamic price generation mechanism ensure that the pricing strategy can quickly respond to market fluctuations, such as sudden fuel price increases, tight transportation capacity or demand surges, etc. In combination with the preset business target, the balance between enterprise operating targets and market reality is achieved, which not only guarantees the profitability of the enterprise, but also enhances the market competitiveness. The present application is suitable for various logistics scenarios such as e-commerce logistics, trunk transportation and intracity distribution. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The first flowchart of the logistics price dynamic pricing method provided by the embodiment of the present application; Figure 2 The second flowchart of the logistics price dynamic pricing method provided by the embodiment of the present application; Figure 3 The third flowchart of the logistics price dynamic pricing method provided by the embodiment of the present application; Figure 4 The fourth flowchart of the logistics price dynamic pricing method provided by the embodiment of the present application; Figure 5 The fifth flowchart of the logistics price dynamic pricing method provided by the embodiment of the present application; Figure 6 The sixth flowchart of the logistics price dynamic pricing method provided by the embodiment of the present application; Figure 7 The structural schematic diagram of the logistics price dynamic pricing device provided by the embodiment of the present application; Figure 8 The structural schematic diagram of the logistics price dynamic pricing device provided by the embodiment of the present application; DETAILED DESCRIPTION

[0017] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The first embodiment of a dynamic pricing method for logistics prices in this invention includes: S101. Obtain and preprocess historical financial cost data, historical transportation operation data, and historical market supply and demand data to obtain preprocessed data.

[0019] In this embodiment, historical financial cost data includes transportation costs, fuel costs, labor costs, and warehouse rental fees, which are derived from the enterprise's ERP system and financial ledgers; historical transportation operation data includes transportation time, loading rate, route congestion frequency, and vehicle failure rate, which are taken from the logistics management system (TMS) and vehicle GPS records; historical market supply and demand data includes average daily order volume, transportation capacity supply, competitor pricing, and regional logistics demand fluctuations, which are derived from market research platforms and industry databases.

[0020] Preprocessing includes data cleaning, data standardization, and data integration. The three types of data are aligned by time (e.g., daily) and region (e.g., East China, North China) to form a structured dataset, ultimately resulting in clean and uniform preprocessed data.

[0021] S102. Perform feature engineering on the preprocessed data to obtain multiple key indicators related to freight rates, and analyze the correlation scores between these key indicators.

[0022] In this embodiment, feature extraction is performed first to mine derivative features related to freight rates from preprocessed data. For example, the average daily transportation time and the proportion of transportation time during peak hours are extracted from transportation time data; the average daily order volume and the weekly fluctuation coefficient of order volume are extracted from order volume data; and the unit mileage cost and the average warehousing cost per shipment are extracted from cost data. Then, the dimensions are optimized through feature screening. Analysis of variance (ANOVA) is used to remove irrelevant features with too small variance (such as fixed administrative fees). The mutual information method is combined to screen the top 20 features with mutual information values ​​related to freight rates (such as unit mileage cost and supply-demand ratio) to obtain key indicators.

[0023] In this embodiment, the correlation scoring analysis uses two indicators: Pearson correlation coefficient and mutual information value, to calculate the correlation strength between key indicators, such as the Pearson correlation coefficient between unit mileage cost and fuel price, and the mutual information value between supply-demand ratio and order premium rate. The results are quantified into correlation scores in the range of 0-1 to form an indicator correlation matrix.

[0024] S103. Based on multiple key indicators and the correlation scores between multiple key indicators, construct an intelligent decision-making model based on graph neural networks.

[0025] In this embodiment, firstly, key indicators are used as nodes in a graph (such as unit mileage cost, supply-demand ratio, and order volume), and the correlation scores between indicators are used as edge weights (the higher the correlation score, the greater the edge weight), constructing an undirected weighted graph to intuitively represent the correlation between indicators. Next, a Graph Attention Network (GAT) is used as the basic architecture of the intelligent decision-making model, which can strengthen the influence of highly correlated indicators on the prediction results through an attention mechanism, adapting to multi-indicator correlation scenarios. Finally, the constructed graph structure and key indicator data are divided into a 7:2:1 training set, validation set, and test set. Mean squared error (MSE) is used as the joint loss function for cost prediction and demand prediction, and the Adam optimizer (learning rate set to 0.001) is selected. Iterative training is performed until the validation set loss converges (e.g., the loss decreases by less than 0.0001 for 10 consecutive rounds).

[0026] In this embodiment, the model output layer is designed with dual output channels, which output cost predictions and demand predictions respectively, ensuring that both types of prediction needs are taken into account.

[0027] S104. Obtain real-time financial cost data, real-time transportation operation data, and real-time market supply and demand data, and input the real-time financial cost data, real-time transportation operation data, and real-time market supply and demand data into the intelligent decision-making model to obtain cost forecast results and demand forecast results.

[0028] In this embodiment, data is collected through IoT devices (vehicle GPS to collect real-time operational data such as transportation location and driving speed), the company's real-time financial system (such as real-time fuel purchase price and temporary labor costs), and the market real-time monitoring platform (such as real-time order submission volume and online transportation capacity). The delay is controlled within 5 minutes to meet the real-time requirements of dynamic pricing.

[0029] In this embodiment, real-time data preprocessing must be consistent with historical data. A sliding window method (window size set to 30 minutes) is used to detect real-time outliers (such as sudden spikes in transport speed). Short-term missing data is supplemented through linear interpolation, followed by standardization to ensure the data format matches the model input requirements. The preprocessed real-time data is input into the trained model, which performs forward propagation calculations and outputs structured prediction results: cost prediction results are presented as a range of unit transport costs for the next 24 hours combined with confidence levels (e.g., 12-15 yuan / km, 92% confidence level). Demand prediction results are presented as predicted order volume plus a probability distribution of fluctuations (e.g., 800 orders in the next 24 hours, with a 75% probability of fluctuation of ±50 orders).

[0030] S105. Combine cost forecast results, demand forecast results, and preset business objectives to generate dynamic prices.

[0031] In this embodiment, firstly, predefined business objectives are established, distinguishing between short-term and long-term goals. For example, the short-term objective is to improve the off-season order conversion rate (≥30%), while the long-term objective is to ensure gross profit margin (≥15%). These objectives are quantified as pricing constraints. Next, a basic pricing logic is established, using cost forecast results as the bottom line. For example, the upper limit of the cost forecast range (e.g., 15 yuan / km) is taken as the base price, plus 5% (15.75 yuan / km). Then, dynamic adjustments are made based on demand forecast results. If the demand forecast is higher than the capacity supply (supply-demand ratio > 1.2), the base price is increased by 3%-8% (the higher the supply-demand ratio, the larger the increase, e.g., 8% increase when the supply-demand ratio is 1.5), increasing revenue. If the demand forecast is lower than the capacity supply (supply-demand ratio < 0.8), the price is decreased by 2%-5% (the lower the supply-demand ratio, the larger the decrease, e.g., 5% decrease when the supply-demand ratio is 0.6), attracting orders. Finally, a business objective verification is introduced. If the adjusted price does not meet the gross profit margin requirement (e.g., the gross profit margin is only 12% after a 5% discount during the off-season), the floating ratio is re-optimized (e.g., a 3% discount) until all constraints are met, and finally a dynamic price is generated.

[0032] It is understood that the executing entity of this invention can be a dynamic pricing device for logistics prices, a terminal, or a server; no specific limitation is made here. This embodiment of the invention will be described using a server as an example.

[0033] This embodiment provides a dynamic pricing method for logistics. It integrates historical financial cost, transportation operation, and market supply and demand data, and uses a graph neural network-based intelligent decision-making model to predict costs and demand. Combining the cost and demand prediction results with preset business objectives, it generates dynamic prices, avoiding pricing biases caused by a single data source and ensuring the comprehensiveness of the pricing basis. Furthermore, the feature engineering process not only accurately extracts key indicators but also quantifies the mutual influence between indicators through correlation analysis, providing a clear relational framework for model construction. The graph neural network-based intelligent decision-making model innovatively transforms the correlations between indicators into a graph structure, which is more adaptable to the complex dynamic changes in the logistics market than traditional regression models or neural networks, significantly improving the accuracy of cost and demand predictions. In addition, real-time data processing and dynamic price generation mechanisms ensure that the pricing strategy can quickly respond to market fluctuations, such as sudden increases in fuel prices, capacity shortages, or surges in demand. Simultaneously, combined with preset business objectives, it achieves a balance between enterprise operating goals and market realities, ensuring both enterprise profitability and enhanced market competitiveness. It is applicable to various logistics scenarios such as e-commerce logistics, trunk transportation, and same-city delivery.

[0034] Please see Figure 2 The second embodiment of the dynamic pricing method for logistics prices in this invention includes: S201. Obtain and integrate historical financial cost data, historical transportation operation data, and historical market supply and demand data to obtain integrated data; In this embodiment, financial cost data is extracted daily from the enterprise's SAP system, including more than 20 fields such as fuel cost (RMB / liter), labor cost (RMB / hour), and vehicle depreciation (RMB / day); transportation operation data is synchronized in real time from the TMS system, including more than 30 fields such as transportation distance (km), travel time (hours), and loading rate (%); market supply and demand data is crawled hourly from logistics platforms (such as Lalamove and Yunmanman) using web crawling tools (such as Scrapy), including multiple fields such as regional order volume, competitor quotations, and customer reviews.

[0035] In this embodiment, Apache Flink is used to build a real-time data pipeline, converting structured data (such as SQL tables) and unstructured data (such as log files) into Parquet format, and linking them using the shipping order ID + timestamp as a unique key to form an integrated dataset.

[0036] S202. Perform data cleaning on the integrated data to obtain cleaned data.

[0037] In this embodiment, for duplicate entries of the same order, the record with the latest timestamp is retained, and duplicate data is cleared.

[0038] In this embodiment, when handling missing values, for numerical fields (such as fuel costs), if the missing rate is <5%, the field is filled with the rolling average of the previous 3 days; if the missing rate is 5%-20%, the KNN nearest neighbor algorithm is used to fill the missing values ​​(taking the average of the 3 most similar orders). For categorical fields (such as weather), the mode is used to fill the missing values, and missing records are marked for subsequent analysis.

[0039] In this embodiment, when handling outliers, a box plot is drawn for continuous variables (such as transportation distance), and IQR is calculated as IQR = Q3 - Q1. Values ​​exceeding [Q1 - 1.5IQR, Q3 + 1.5IQR] are considered outliers and replaced with Q3 + 1.5IQR (upper limit) or Q1 - 1.5IQR (lower limit). For discrete variables (such as order status), outlier records that do not conform to preset enumeration values ​​(such as completed, in transit) are deleted. The cleaned data must pass validation: missing rate <1%, duplication rate <0.5%, and outlier handling ratio <3%.

[0040] S203. Standardize the cleaned data to obtain preprocessed data.

[0041] In this embodiment, continuous variables are standardized using Z-score. Variables with a fixed value range (such as load factor 0-100%) are standardized using Min-Max, mapped to the [0,1] interval, preserving the relative magnitude of the data. Categorical variables (such as season and mode of transport) are encoded using one-hot encoding, converting the season field into four binary fields (0 or 1) for spring, summer, autumn, and winter.

[0042] In this embodiment, financial, operational, and market data are integrated to comprehensively reflect the influencing factors of logistics pricing, avoiding the bias of data from a single department. Moreover, the data cleaning process specifically addresses the common noise problem in logistics data, identifying and processing outliers through scientific methods to ensure the rationality of data distribution. In addition, standardization eliminates the differences in the dimensions of different indicators, enabling each feature to participate fairly in model training and avoiding model bias caused by differences in numerical magnitude. At the same time, the completeness, consistency, and accuracy of the preprocessed dataset are significantly improved, directly reducing the difficulty of subsequent feature engineering and the number of iterations in model training, providing a fundamental guarantee for the accuracy of the final pricing. This approach is particularly suitable for the logistics industry, where data sources are complex and data quality varies greatly.

[0043] Please see Figure 3 A third embodiment of a dynamic pricing method for logistics prices in this invention includes: S301. Extract effective features from the preprocessed data, calculate the correlation coefficient between the effective features and the freight rate, and select effective features with a correlation coefficient greater than the preset coefficient threshold as candidate features.

[0044] In this embodiment, basic and derived features are extracted from the preprocessed data. Basic features include original fields (such as fuel price and transportation distance). Derivative features are calculated through business logic, such as unit weight cost = total cost / cargo weight, daily order density = regional order volume / regional area, and timeliness deviation rate = (actual timeliness - standard timeliness) / standard timeliness. The Pearson correlation coefficient between each feature and historical freight rates is calculated, and a threshold of 0.4 (absolute value) is set to select 40 candidate features, such as fuel price (r=0.72), transportation distance (r=0.68), and order volume (r=-0.53, a negative correlation indicates that more orders result in lower prices), while weakly correlated features such as vehicle age (r=0.12) are excluded.

[0045] S302. Based on the importance of the Gini coefficient in the random forest algorithm, the candidate features are ranked by importance, and the top 10 candidate features are selected as key indicators.

[0046] In this embodiment, 40 candidate features are used as input variables, and historical freight rates are used as the target variable to train a random forest model: using the RandomForestRegressor from the Scikit-learn library, with nestimators=200 and maxdepth=10, the model performance is evaluated using 5-fold cross-validation (R²>0.85). After model training, the featureimportances attribute is extracted to obtain the Gini importance score of each feature (summed to 1). A higher score indicates a greater impact of the feature on freight rates. The top 10 features are selected as key indicators, sorted in descending order of score, such as: fuel price (0.18), transport distance (0.15), order volume (0.12), load factor (0.10), average competitor price (0.09), peak / off-peak season coefficient (0.08), labor cost (0.07), route complexity (0.06), weather impact (0.05), and customer level (0.05). Key indicators must meet the coverage requirement: cumulative importance score > 0.8, ensuring coverage of most freight rate influencing factors.

[0047] S303. The correlation scores between key indicators are obtained by calculating the correlation coefficient.

[0048] In this embodiment, the correlation score between key indicators is obtained by calculating the correlation coefficient. Specifically, this includes: using the Pearson correlation coefficient to identify the linear correlation between key indicators and obtaining the linear correlation coefficient; using the Spearman correlation coefficient to identify the monotonic nonlinear correlation between key indicators and obtaining the monotonic nonlinear correlation coefficient; using the distance correlation coefficient to identify the non-monotonic nonlinear correlation between indicators and obtaining the non-monotonic nonlinear correlation coefficient; and fusing the linear correlation coefficient, the monotonic nonlinear correlation coefficient, and the non-monotonic nonlinear correlation coefficient according to preset weights to obtain the correlation score between key indicators.

[0049] In this embodiment, the Pearson correlation coefficient is used to calculate the strength of the linear association for any two key indicators (e.g., A is the fuel price per unit and B is the unit transportation cost). Specifically, sample data (e.g., 1000 historical records) for the two indicators are obtained, and their respective means are calculated. The covariance and standard deviation are calculated based on the means of the two key indicators. The Pearson coefficient for any two key indicators is calculated based on the covariance and standard deviation, with a value range of [-1, 1]. The larger the absolute value, the stronger the linear association. For example, the r=0.82 for fuel price per unit and unit cost indicates a strong positive linear association; the r=-0.65 for order volume and price indicates a strong negative linear association. After calculation, r is standardized and converted to a linear association coefficient in the range [0, 1] (r for positive association and |r| for negative association).

[0050] In this embodiment, Spearman's rank correlation coefficient is used to identify the monotonic trends (whether linear or not) between indicators. Specifically, the sample data of two indicators A and B are sorted and assigned rank values ​​(e.g., the smallest is 1, the second smallest is 2, and the average rank is taken for the same value); the rank difference is calculated; finally, based on the rank difference, the Spearman's rank correlation coefficient is calculated. The Spearman's rank correlation coefficient ranges from [-1, 1], and the larger the absolute value, the stronger the monotonic correlation. For example, ρ = 0.79 for order volume and loading rate indicates that the loading rate monotonically increases with the increase of order volume (non-linear but consistent trend); ρ = 0.85 for transportation distance and timeliness indicates that the timeliness monotonically increases with the longer the distance. Similarly, ρ is standardized and converted into a monotonic non-linear correlation coefficient in the range of [0, 1] for subsequent fusion.

[0051] In this embodiment, distance correlation coefficient is used to capture non-monotonic and non-linear associations (such as inverted U-shaped and S-shaped relationships). Specifically, the distance matrix (Euclidean distance between samples) of two indicators A and B is calculated; the distance matrix is ​​centered to obtain a centered distance matrix; the distance covariance and distance variance are calculated; and the distance correlation coefficient between the two indicators A and B is calculated based on the distance covariance and distance variance, with a value range of [0,1]. The larger the value, the stronger the non-monotonic association. For example, the dCor of route complexity (0-10 points) and customer complaint rate is 0.68, indicating that the complaint rate is lowest at medium complexity (non-monotonic relationship). dCor is directly used as the non-monotonic non-linear association coefficient and retained in the [0,1] interval.

[0052] In this embodiment, weak correlation information is filtered out by using correlation coefficient thresholds to reduce feature redundancy, lower model complexity, and improve training efficiency. At the same time, the ranking of key indicators based on Gini importance ensures that the core factors with the greatest impact on freight rates are retained, enabling the model to focus on key driving factors and improve interpretability and prediction accuracy. In addition, quantifying the mutual influence between indicators provides a node relationship basis for graph neural networks, enabling the model to capture the synergistic effect of multiple factors, which is more realistic than traditional single-factor analysis.

[0053] Please see Figure 4 A fourth embodiment of a dynamic pricing method for logistics prices in this invention includes: S401. Construct an initial graph structure using key indicators as the node features of the graph and the correlation scores between key indicators as the edge weights between nodes.

[0054] In this embodiment, 10 selected key indicators are used as graph nodes, each containing 5 features: the mean of the past 30 days, standard deviation, daily change rate, correlation coefficient with freight rates, and historical maximum impact weight. Node types are divided into three categories: cost-related (fuel price per unit, labor cost), operational-related (transportation distance, load factor, route complexity), and market-related (order volume, competitor average price, peak / off-peak season coefficient, weather impact, customer level), facilitating subsequent hierarchical learning. A weighted directed graph is constructed using the association scores obtained in claim 4 as edge weights: for example, the edge weight pointing from fuel price to unit cost is 0.75 (strong association), the edge weight pointing from order volume to competitor average price is 0.42 (medium association), and the edge weight pointing from weather impact to transportation distance is 0.18 (weak association). The graph structure is stored as an adjacency matrix (10×10) and a node feature matrix (10×5), visualized using the NetworkX library to verify the rationality of node connections (e.g., dense connections between cost-related nodes).

[0055] S402. Build an initial model based on graph attention network. The initial model includes an input layer, a hidden layer, and an output layer. The input layer is used to receive multiple node features and edge weights. The hidden layer is a multi-layer graph neural network structure. The hidden layer is used to collect neighbor node information through a message passing mechanism based on multiple node features and edge weights. The output layer is used to generate cost prediction results and demand prediction results based on the neighbor node information output by the hidden layer.

[0056] In this embodiment, a Graph Attention Network (GAT) architecture is adopted, with the following specific structure: The input layer receives a node feature matrix (10×5) and an adjacency matrix (10×10); the hidden layer contains two GAT layers. The first layer sets up eight attention heads (multi-head attention mechanism), each attention head maps the node features to an 8-dimensional space through a linear transformation (weight matrix), calculates the attention coefficients, and normalizes them using the softmax function to obtain the attention weights; it aggregates neighbor features and concatenates the outputs of the eight attention heads (dimension 8×8=64). The second GAT layer sets up one attention head, maps the 64-dimensional features to 32 dimensions, and outputs the high-order feature representation of the nodes. The output layer is divided into two branches: the cost prediction branch outputs the unit cost prediction value (MSE loss) through a fully connected layer (32→1); the demand prediction branch outputs the order quantity prediction value (MAE loss) through a fully connected layer (32→1). The total loss is the weighted sum of the two (weight 0.6:0.4, because cost is more critical to pricing).

[0057] S403. Based on historical financial cost data, historical transportation operation data, and historical market supply and demand data, generate historical graph samples, and combine the historical graph samples with the initial graph structure to train and optimize the initial model to obtain an intelligent decision-making model. During the training process, an attention mechanism is introduced to dynamically adjust the influence of neighboring nodes.

[0058] In this embodiment, based on historical financial cost data, historical transportation operation data, and historical market supply and demand data, time windows are divided according to a fixed period (such as weekly), and one graph data sample is generated for each window.

[0059] Each historical graph sample contains three parts: Node features: Ten key indicators are used as nodes (such as fuel price and transportation distance). Each node contains five features (mean, standard deviation, maximum / minimum value, and correlation coefficient with freight rate) to form a node feature matrix (10×5).

[0060] Edge weights: Based on the correlation scoring method of claim 4, calculate the dynamic correlation scores between indicators within each window (e.g., the correlation score between order volume and price is higher in the peak season window) to form an adjacency matrix (10×10).

[0061] Label data: The actual cost (e.g., average weekly unit cost) and actual demand (e.g., average weekly order volume) corresponding to each window serve as supervision labels for model training.

[0062] In this embodiment, the initial graph structure provides a priori association framework for the model. The node features of the initial graph structure adopt historical global statistical values ​​(such as the average and fluctuation characteristics of fuel unit prices over the past 3 years), and the edge weights adopt the average association score over the entire period (such as the average score of fuel unit price-cost being 0.72), which are then transformed into the initial parameters of the model. The initial features of the 10 nodes are mapped to 64-dimensional vectors, and the average association score is used as the initial weight of the edges, stored as trainable parameters.

[0063] In this embodiment, the historical graph samples are divided into training and validation sets in a 7:3 ratio. The optimizer is Adam, with an initial learning rate of 0.001, decaying by 10% every 20 rounds; the batch size is set to 8, and the iteration count is 100 rounds. During training, an attention mechanism dynamically adjusts the edge weights: for example, when the training samples contain peak season data, the model automatically increases the attention weights of order volume and price; when fuel price increase data is included, the weights of fuel unit price and cost are increased. Early stopping is used to prevent overfitting. The final model achieves a cost prediction MAE of 3.2 yuan / ton and a demand prediction MAE of 12 orders / day on the validation set, meeting business requirements.

[0064] In this embodiment, the graph structure better reflects the complex relationships within a logistics system. By using indicators as nodes and correlation scores as edge weights, it intuitively presents the mutual influence between multiple factors, overcoming the limitations of traditional models that treat features as independent variables. Simultaneously, the graph attention mechanism dynamically adjusts the influence weights of neighboring nodes in the graph structure, enabling the model to automatically focus on key correlations and improve adaptability to dynamic scenarios. Furthermore, the multi-layer message passing mechanism captures higher-order correlations, and by aggregating neighboring node information, it more comprehensively learns the transmission paths between indicators, thereby improving prediction accuracy.

[0065] Please see Figure 5 The fifth embodiment of a dynamic pricing method for logistics prices in this invention includes: S501. Obtain real-time financial cost data, real-time transportation operation data, and real-time market supply and demand data, and perform preprocessing and feature engineering on the real-time financial cost data, real-time transportation operation data, and real-time market supply and demand data to obtain the data to be predicted.

[0066] In this embodiment, real-time data is collected through multiple channels. Financial cost data is pulled from the enterprise's financial system API every 5 minutes (e.g., current fuel price of 8.5 yuan / liter, highway toll standard); transportation operation data is uploaded in real time through vehicle terminals (GPS combined with sensors) (e.g., 20 vehicles on the road, average loading rate of 75%); market supply and demand data is updated every 10 minutes through web crawler (e.g., 80 new orders in the region in the past hour, current average price of competitors of 120 yuan / ton).

[0067] In this embodiment, the preprocessing stage is completed in the stream processing engine: cleaning (filtering duplicate orders and correcting abnormal oil prices), missing value imputation (interpolating with data from the previous 5 minutes), and standardization (converting to Z-score based on historical mean and standard deviation). Feature engineering extracts 10 key indicators consistent with historical data and calculates real-time feature values ​​(e.g., current peak / off-peak season coefficient = 1.2, indicating peak season), ensuring matching with the model input dimensions (10 indicators, each containing two features: real-time value and 5-minute change rate).

[0068] S502. Combine the data to be predicted with a predefined association graph structure to construct a real-time graph data instance.

[0069] In this embodiment, based on a predefined graph structure template (10 nodes, edge weights equal to historical average association scores), real-time feature values ​​are updated to the corresponding nodes. For example, the fuel price node feature is updated to the current value of 8.5 yuan / liter and a 2% increase over 5 minutes; the order volume node feature is updated to 80 orders and a 15% increase compared to the previous period. For temporary strong associations appearing in the real-time data (such as sudden heavy rain, where the weather impact increases from 0.2 to 0.8), the corresponding edge weights are dynamically adjusted: the edge weight between weather impact and transportation timeliness is temporarily increased from 0.3 to 0.7, and the edge weight between weather impact and cost is increased from 0.2 to 0.5, reflecting the strong impact of the current weather on operations. After the real-time graph data instance is constructed, it is converted into the model input format.

[0070] S503. Input the real-time graph data instance into the intelligent decision-making model to obtain cost prediction results and demand prediction results.

[0071] In this embodiment, real-time graph data instances are input into the intelligent decision-making model, which quickly calculates node attention weights using a graph attention mechanism. For example, when oil prices rise, the attention weight of the fuel price node increases from 0.15 to 0.25, prioritizing its impact on the cost node. The intelligent decision-making model outputs two predictions: a predicted unit transportation cost for the next 24 hours (e.g., 105 yuan / ton) and a predicted demand (e.g., an estimated order volume of 300 orders). The predictions include confidence scores (e.g., 92% confidence for cost prediction and 88% confidence for demand prediction).

[0072] In this embodiment, real-time indicator values ​​are combined with predefined association graphs, which can retain stable associations verified in history and incorporate temporary strong associations in a timely manner, making the graph data instances more relevant to the current situation. In addition, the high-resolution graph neural network can quickly focus on key indicators through an optimized attention mechanism, achieving a second-level prediction response, which is more adaptable to the high-frequency fluctuations of the logistics market than the traditional batch prediction mode.

[0073] Please see Figure 6 The sixth embodiment of a dynamic pricing method for logistics prices in this invention includes: S601. Obtain the preset business objective, which is to maximize profits.

[0074] In this embodiment, the predetermined business objective is explicitly set as profit maximization, while also considering market share and customer satisfaction. Specific definitions: Daily total profit = Σ (unit price per route - predicted unit cost) × predicted demand for that route, where the predicted unit cost comes from the cost forecast result of claim 6, and the predicted demand comes from the demand forecast result. Supporting objectives include: a repurchase rate of ≥90% for core customers (annual order volume > 1000 orders), maintaining price competitiveness (compared to the average price of competitors) within the range of [-5%, +10%], ensuring a balance between long-term operations and short-term profits. Business objectives require management approval and must be written into the pricing system's configuration file, supporting quarterly adjustments.

[0075] S602. Convert profit maximization into a business objective function and business conditions, and input the cost forecast results and demand forecast results into the business objective function. Use an optimization algorithm to solve for the optimal price that satisfies the constraints.

[0076] In this embodiment, the objective function is maxP=Σ(p i -c i )×d i , where p i Let c be the unit price of the i-th route. i d is the predicted unit cost. i d is the demand forecast value. i With p i Negative correlation, based on historical elasticity coefficients k i It is a constant. This refers to demand elasticity.

[0077] Constraints: p i ≥c i ×1.1 (gross profit margin ≥ 10%); p i ≤avgcompetitori×1.1 (not exceeding 10% of the average price of competitors); p i≥avgcompetitori×0.95 (not less than 5% of the average price of competitors, maintaining competitiveness); for core customers, p i ≤Standard price × 0.95 (additional discount). Solve using a genetic algorithm: Initialize 50 price schemes (p i The combination of these factors, with the fitness function being the total profit P, yields the theoretically optimal price scheme p through 30 iterations of selection (retaining the top 20% of high-fitness schemes), crossover (randomly swapping the prices of two lines), and mutation (randomly adjusting the prices of 10% of lines). opt .

[0078] S603. Adjust the optimal price according to the preset business rules to generate a dynamic price.

[0079] In this embodiment, p is modified based on preset business rules. opt Bulk order discount (single order weight > 1 ton, p = p) opt ×0.95); Surcharge for urgent orders (time requirement <6 hours, p=p opt ×1.2); First order discount for new customers (p=p opt ×0.9); Regional policy adjustments (e.g., in remote areas where costs are high, p=p opt ×1.1). The revised price needs to pass risk control verification: check for price discrimination (price difference ≤5% for customers on the same route and under the same conditions) and whether a low-price warning has been triggered (manual review is required when the gross profit margin is <8%). After the final price is generated, it is synchronized to the customer's ordering platform and the business personnel's CRM system via API interface, and the effective time is set (e.g., valid for 1 hour, automatically recalculated after expiration) to achieve true dynamic pricing.

[0080] In this embodiment, profit maximization is the core objective. Mathematical modeling transforms the pricing problem into a quantifiable optimization problem, avoiding the empirical biases inherent in traditional pricing and ensuring objectivity in decision-making. Simultaneously, while pursuing profit, constraints such as cost coverage and market competition are incorporated to prevent customer loss due to excessively high pricing or losses due to excessively low pricing, achieving a balance of interests among multiple parties. Furthermore, a correction process translates the company's business strategy into actionable price adjustments, making the theoretically optimal price more aligned with real-world scenarios, thereby enhancing customer acceptance and market competitiveness. This process leverages the accuracy of the data model while retaining the flexibility of business experience. The generated dynamic price can be adjusted in real-time according to costs, demand, and the competitive environment, creating greater profit margins for logistics companies.

[0081] The above describes the dynamic pricing method for logistics in the embodiments of the present invention. The following describes the apparatus in the embodiments of the present invention. Please refer to [link / reference]. Figure 7 The implementation methods of the dynamic pricing device for logistics prices in this invention include: The preprocessing module 701 is used to acquire and preprocess historical financial cost data, historical transportation operation data, and historical market supply and demand data to obtain preprocessed data. Analysis module 702 is used to perform feature engineering on the preprocessed data to obtain multiple key indicators related to freight rates, and to analyze the correlation scores between the multiple key indicators. Module 703 is used to build an intelligent decision-making model based on graph neural networks based on multiple key indicators and the correlation scores between multiple key indicators. The prediction module 704 is used to acquire real-time financial cost data, real-time transportation operation data and real-time market supply and demand data, and input the real-time financial cost data, real-time transportation operation data and real-time market supply and demand data into the intelligent decision-making model to obtain cost prediction results and demand prediction results. The generation module 705 is used to generate a dynamic price by combining the cost forecast results, the demand forecast results, and the preset business objectives.

[0082] In this embodiment, the preprocessing module 701 includes: an integration unit 7011, used to acquire and integrate historical financial cost data, historical transportation operation data, and historical market supply and demand data to obtain integrated data; a cleaning unit 7012, used to clean the integrated data to obtain cleaned data; and a standardization unit 7013, used to standardize the cleaned data to obtain preprocessed data.

[0083] In this embodiment, the analysis module 702 includes: an extraction unit 7021, used to extract effective features from the preprocessed data, calculate the correlation coefficient between the effective features and the freight rate, and select effective features with a correlation coefficient greater than a preset coefficient threshold as candidate features; a ranking unit 702, used to rank the candidate features based on the importance of the Gini coefficient in the random forest algorithm, and select the top 10 candidate features as key indicators; and an analysis unit 7023, used to analyze and obtain the correlation score between key indicators by calculating the correlation coefficient.

[0084] In this embodiment, the construction module 703 includes: a first construction unit 7031, used to construct an initial graph structure using key indicators as node features of the graph and the correlation scores between key indicators as edge weights between nodes; a building unit 7032, used to build an initial model based on a graph attention network, the initial model including an input layer, a hidden layer and an output layer, the input layer being used to receive multiple node features and edge weights, the hidden layer being a multi-layer graph neural network structure, the hidden layer being used to collect neighbor node information through a message passing mechanism based on multiple node features and edge weights, the output layer being used to generate cost prediction results and demand prediction results based on the neighbor node information output by the hidden layer; and a training unit 7033, used to generate historical graph samples based on historical financial cost data, historical transportation operation data and historical market supply and demand data, and to train and optimize the initial model by combining the historical graph samples and the initial graph structure to obtain an intelligent decision-making model, and during the training process, an attention mechanism is introduced to dynamically adjust the influence of neighbor nodes.

[0085] In this embodiment, the prediction module 704 includes: a processing unit 7041, used to acquire real-time financial cost data, real-time transportation operation data, and real-time market supply and demand data, and to preprocess and perform feature engineering on the real-time financial cost data, the real-time transportation operation data, and the real-time market supply and demand data to obtain data to be predicted; a second construction unit 7042, used to combine the data to be predicted with a predefined association graph structure to construct a real-time graph data instance; and a prediction unit 7043, used to input the real-time graph data instance into an intelligent decision-making model to obtain cost prediction results and demand prediction results.

[0086] In this embodiment, the generation module 705 includes: an acquisition unit 7051, used to acquire a preset business objective, wherein the preset business objective is profit maximization; a conversion unit 7052, used to convert profit maximization into a business objective function and business conditions, and input cost prediction results and demand prediction results into the business objective function, and use an optimization algorithm to solve for the optimal price that satisfies the constraints; and a generation unit 7053, used to modify the optimal price according to preset business rules to generate a dynamic price.

[0087] In this embodiment, historical financial cost, transportation operation, and market supply and demand data are integrated, and cost and demand forecasts are performed using an intelligent decision-making model based on graph neural networks. The dynamic price is generated by combining the cost forecast results, demand forecast results, and preset business objectives, avoiding pricing deviations caused by a single data source and ensuring the comprehensiveness of the pricing basis. Furthermore, the feature engineering process not only accurately extracts key indicators but also quantifies the mutual influence between indicators through correlation analysis, providing a clear relational framework for model construction. The intelligent decision-making model based on graph neural networks innovatively transforms the correlations between indicators into a graph structure, which is more adaptable to the complex dynamic changes in the logistics market than traditional regression models or neural networks, significantly improving the accuracy of cost and demand forecasts. In addition, the real-time data processing and dynamic price generation mechanism ensure that the pricing strategy can quickly respond to market fluctuations, such as sudden increases in fuel prices, capacity shortages, or surges in demand. Simultaneously, combined with preset business objectives, it achieves a balance between corporate operating goals and market realities, ensuring corporate profitability and enhancing market competitiveness. This model is applicable to various logistics scenarios such as e-commerce logistics, trunk transportation, and same-city delivery.

[0088] Figure 7 The structure of the logistics price dynamic pricing device shown does not constitute a limitation on the logistics price dynamic pricing device, and can implement the steps of the logistics price dynamic pricing method provided in the above method embodiments.

[0089] above Figure 7 The dynamic pricing device for logistics prices in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The dynamic pricing device for logistics prices in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0090] Figure 8 This is a schematic diagram of the structure of a dynamic pricing device for logistics provided in an embodiment of the present invention. The device 800 can vary significantly due to different configurations or performance, and may include one or more central processing units (CPUs) 810 (e.g., one or more processors) and a memory 820, and one or more storage media 830 (e.g., one or more mass storage devices) for storing application programs 833 or data 832. The memory 820 and storage media 830 can be temporary or persistent storage. The program stored in the storage media 830 may include one or more modules (not shown), each module including a series of instruction operations on the device 800. Furthermore, the processor 810 may be configured to communicate with the storage media 830 and execute the series of instruction operations in the storage media on the device 800.

[0091] Device 800 may also include one or more power supplies 840, one or more wired or wireless network interfaces 850, one or more input / output interfaces 860, and / or one or more operating systems 831, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc.

[0092] This invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of a dynamic pricing method for logistics prices.

[0093] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0094] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0095] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A logistics price dynamic pricing method, characterized by, The logistics price dynamic pricing method comprises: Obtain and preprocess historical financial cost data, historical transportation operation data and historical market supply and demand data to obtain preprocessed data; Perform feature engineering processing on the preprocessed data to obtain a plurality of key indicators related to the freight rate, and analyze the correlation scores between the plurality of key indicators; Based on the plurality of key indicators and the correlation scores between the plurality of key indicators, an intelligent decision-making model based on a graph neural network is constructed; Obtain real-time financial cost data, real-time transportation operation data and real-time market supply and demand data, and input the real-time financial cost data, real-time transportation operation data and real-time market supply and demand data into the intelligent decision-making model to obtain cost prediction results and demand prediction results; Combine the cost prediction results, the demand prediction results and a preset business target to generate a dynamic price.

2. The logistics price dynamic pricing method according to claim 1, characterized in that, The obtaining and preprocessing of historical financial cost data, historical transportation operation data and historical market supply and demand data to obtain preprocessed data comprises: Obtain and integrate historical financial cost data, historical transportation operation data and historical market supply and demand data to obtain integrated data; Perform data cleaning on the integrated data to obtain cleaned data; Perform standardization processing on the cleaned data to obtain preprocessed data.

3. The method of claim 1, wherein, The feature engineering processing on the preprocessed data to obtain a plurality of key indicators related to the freight rate, and analyzing the correlation scores between the plurality of key indicators comprises: Extract effective features from the preprocessed data, calculate the correlation coefficients between the effective features and the freight rate, and select effective features with correlation coefficients greater than a preset coefficient threshold as candidate features; Sort the candidate features based on the Gini coefficient importance in the random forest algorithm, and select the top 10 candidate features as key indicators; Analyze the correlation scores between the key indicators by calculating the correlation coefficients.

4. The method of claim 3, wherein, The analysis of the correlation scores between the key indicators by calculating the correlation coefficients comprises: Identify the linear correlation between the key indicators using the Pearson correlation coefficient to obtain a linear correlation coefficient; Identify the monotonic nonlinear correlation between the key indicators using the Spearman correlation coefficient to obtain a monotonic nonlinear correlation coefficient; Identify the non-monotonic nonlinear correlation between the key indicators using the distance correlation coefficient to obtain a non-monotonic nonlinear correlation coefficient; Fuse the linear correlation coefficient, the monotonic nonlinear correlation coefficient and the non-monotonic nonlinear correlation coefficient according to a preset weight to obtain the correlation scores between the key indicators.

5. The method of claim 1, wherein, The construction of an intelligent decision-making model based on a graph neural network based on the plurality of key indicators and the correlation scores between the plurality of key indicators comprises: Construct an initial graph structure by taking the key indicators as node features of the graph and taking the correlation scores between the key indicators as edge weights between the nodes. The initial model based on the graph attention network is built, and the initial model includes an input layer, a hidden layer and an output layer. The input layer is used to receive a plurality of node features and edge weights. The hidden layer is a multi-layer graph neural network structure, and is used to collect neighbor node information through a message passing mechanism set according to the plurality of node features and the edge weights. The output layer is used to generate a cost prediction result and a demand prediction result according to the neighbor node information output by the hidden layer. The historical graph samples are generated based on historical financial cost data, historical transportation operation data and historical market supply and demand data, and the initial model is trained and optimized in combination with the historical graph samples and the initial graph structure to obtain the intelligent decision model. In the training process, the attention mechanism is introduced to dynamically adjust the influence of neighbor nodes.

6. The method of claim 1, wherein, The real-time financial cost data, the real-time transportation operation data and the real-time market supply and demand data are acquired, and the real-time financial cost data, the real-time transportation operation data and the real-time market supply and demand data are input into the intelligent decision model to obtain the cost prediction result and the demand prediction result, including: The real-time financial cost data, the real-time transportation operation data and the real-time market supply and demand data are acquired, and the real-time financial cost data, the real-time transportation operation data and the real-time market supply and demand data are preprocessed and feature engineered to obtain to-be-predicted data; The to-be-predicted data is combined with a predefined correlation graph structure to construct a real-time graph data instance; The real-time graph data instance is input into the intelligent decision model to obtain the cost prediction result and the demand prediction result.

7. The method of claim 1, wherein, The cost prediction result, the demand prediction result and a preset business target are combined to generate a dynamic price, including: The preset business target is profit maximization; The profit maximization is converted into a business objective function and a business condition, and the cost prediction result and the demand prediction result are input into the business objective function. An optimization algorithm is used to solve an optimal price that meets the constraint condition; The optimal price is corrected according to a preset business rule to generate the dynamic price.

8. A logistics price dynamic pricing apparatus, characterized by, It includes: A preprocessing module is configured to acquire and preprocess historical financial cost data, historical transportation operation data and historical market supply and demand data to obtain preprocessed data; An analysis module is configured to perform feature engineering processing on the preprocessed data to obtain a plurality of key indicators related to the freight rate, and analyze the correlation scores between the plurality of key indicators; A construction module is configured to construct an intelligent decision model based on graph neural network based on the plurality of key indicators and the correlation scores between the plurality of key indicators; A prediction module is configured to acquire real-time financial cost data, real-time transportation operation data and real-time market supply and demand data, and input the real-time financial cost data, the real-time transportation operation data and the real-time market supply and demand data into the intelligent decision model to obtain a cost prediction result and a demand prediction result; A generation module is configured to combine the cost prediction result, the demand prediction result and a preset business target to generate a dynamic price.

9. A logistics price dynamic pricing apparatus characterized by, It includes a memory and at least one processor, and the memory stores computer readable instructions. The at least one processor invokes the computer readable instructions in the memory to perform the individual steps of the method for dynamic pricing of logistics prices as claimed in any one of claims 1-7.

10. A computer-readable storage medium having stored thereon computer-readable instructions, wherein, The computer readable instructions, when executed by a processor, implement the individual steps of the method for dynamic pricing of logistics prices as claimed in any one of claims 1-7.

Images