Navigation container freight rate prediction method based on artificial intelligence

Abstract

The invention discloses a shipping container freight rate prediction method based on artificial intelligence, and relates to the technical field of freight rate prediction.The method comprises the steps that a port state prediction model is constructed through a graph neural network GNN, a port congestion index is calculated based on the port throughput, the ship queuing condition, the berth utilization rate and the transportation scheduling condition, and a port network graph is constructed to predict the freight rate of a shipping container; describing a connection relationship between ports by using an adjacent matrix, performing information propagation by using GNN, and calculating updated port features; a port congestion index is adopted to train an LSTM regression model, the future port state is predicted, and therefore the mutual influence between ports is dynamically captured; and based on the predicted port state, a reinforcement learning RL training agent is adopted to construct a cargo circulation prediction model, a Markov decision process MDP is adopted to define a cargo circulation state, and a deep Q network DQN is adopted to optimize a cargo circulation path, so that the agent can adjust a cargo circulation strategy in different port states, and the cargo circulation prediction accuracy is improved. Therefore, market change is adapted and prediction reliability is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of freight rate prediction, and in particular to a shipping container freight rate prediction method based on artificial intelligence. Background Art

[0002] The prediction of shipping container freight rates is crucial for supply chain management. Currently, the main methods for freight rate prediction include time series analysis, machine learning models, and economic models based on supply and demand relationships. Time series methods (such as ARIMA, SARIMA) rely on historical data for trend speculation, but the market is greatly affected by unexpected events, and simply relying on historical patterns often cannot effectively predict price changes. Machine learning methods (such as LSTM, XGBoost) can improve the prediction accuracy by learning a large number of data features, but they highly depend on data quality and feature selection, and it is difficult to deal with complex unstructured influencing factors. Economic models based on supply and demand relationships mainly model freight rate fluctuations from a macro perspective, considering factors such as trade volume, fuel price, and ship capacity, but their ability to capture short-term market dynamics is limited, and it is difficult to reflect sudden market changes in a timely manner.

[0003] The fluctuation of port capacity is one of the core factors affecting freight rates. Traditional models often take port throughput as a key variable, but this indicator itself has a lag, making it difficult to predict the actual congestion situation of the port in advance; Freight rate fluctuations are not only related to throughput but also affected by various factors such as ship queuing, berth utilization rate, and truck capacity, and these factors are often not fully considered in the existing prediction framework; Especially during unexpected events, the port throughput capacity may drop significantly in a short period of time, leading to drastic fluctuations in freight rates. However, traditional methods are difficult to obtain and integrate these sudden factors in a timely manner, resulting in prediction lags or misjudgments.

[0004] Although some models attempt to introduce new data sources, such as shipping company operation data, market supply and demand adjustment strategies, etc., the availability and real-time nature of these data are still limited. At the same time, there are also methods that use a rule-driven approach to adjust the prediction results when the market fluctuation exceeds a specific threshold, but this method depends on fixed rules and has poor adaptability, and the effect is not ideal in a rapidly changing market environment; Therefore, there is an urgent need for a shipping container freight rate prediction solution based on artificial intelligence to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides a shipping container freight rate prediction method based on artificial intelligence to solve the problems that traditional prediction solutions rely on lagged data, do not fully consider port congestion and unexpected events, and are difficult to adjust in a timely manner, resulting in lagged or inaccurate freight rate predictions.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: An embodiment of the present invention provides a shipping container freight rate prediction method based on artificial intelligence, which includes: Step S1: Collect shipping data and construct a standardized data set; Step S2: Based on the standardized data set, construct a port status prediction model to predict the port status; Step S3: According to the prediction result of the port status, construct a cargo flow prediction model to predict the cargo flow status; Step S4: Based on the cargo flow status prediction result of Step S3, construct a freight rate prediction model to predict the freight rate; Step S5: On the basis of the freight rate prediction in Step S4, combine market supply and demand data and fuel price data, and use the Bayesian optimization method to dynamically adjust the parameters of the freight rate prediction model to determine the final freight rate prediction result.

[0008] As a preferred solution of the shipping container freight rate prediction method based on artificial intelligence of the present invention, wherein: the shipping data includes ship trajectory data, port berth data, transportation scheduling data, historical freight rate data and real-time weather information; In Step S1, the shipping data is cleaned, de-duplicated and normalized to construct a standardized data set.

[0009] As a preferred solution of the shipping container freight rate prediction method based on artificial intelligence of the present invention, wherein: the ship trajectory data includes the position, speed, course and destination port information of the ships on the route, which is obtained by using the Automatic Identification System (AIS), and the ship trajectory data is calibrated by combining the data of the Vessel Traffic Service (VTS); The port berth data includes the real-time berth occupancy, estimated berthing time and operation completion time, which are collected by using the Port Management System (PMS) and the Port Operation Scheduling System (TOS); The transportation scheduling data includes route adjustment, liner delay and capacity change.

[0010] As a preferred solution of the shipping container freight rate prediction method based on artificial intelligence of the present invention, wherein: in Step S2, the Graph Neural Network (GNN) is used to analyze the port throughput, ship queuing situation, berth utilization rate and transportation scheduling situation, calculate the port congestion index, and predict the port status based on the port congestion index.

[0011] As a preferred solution of the method for predicting the freight rate of shipping containers based on artificial intelligence according to the present invention, wherein: in step S2, the steps of analyzing the port throughput, ship queuing situation, berth utilization rate and transportation scheduling situation by using the graph neural network GNN, calculating the port congestion index, and predicting the port state based on the port congestion index are as follows: Construct a port network graph, with ports as nodes and shipping routes as edges, and use the graph neural network GNN for feature extraction and prediction. Define the port network graph as : , wherein, represents the set of port nodes, represents the th port, is the total number of ports, represents the set of edges, represents the shipping connection between port and port . represents the port feature matrix, represents the feature vector of port . Extract port features, and define the feature vector of port as : , wherein, represents the port throughput, represents the number of queuing ships, represents the berth utilization rate, represents the amount of transportation scheduling adjustment, Define the adjacency matrix : If there is a direct shipping route between port and port , otherwise . wherein, is the -dimensional adjacency matrix, represents the shipping route relationship between port and . Use GNN to propagate port status information and calculate the updated port features: , wherein, represents the The port of the layer Feature Indicating the port of the layer Feature Indicating the port set of neighbors, For the port is the degree, For the port is the degree, is the trainable weight matrix of the GNN, is the activation function, using ReLU for non-linear mapping, Calculating the port congestion index, defining the port congestion index as : , Wherein, Indicating the port congestion index, is the corresponding weight parameter, The weight parameter is optimized by least squares regression, and the formula is: , Wherein, is the true congestion index calculated from historical data, Predicting the port status based on the port congestion index, using the port congestion index as the input, and training a regression model to predict the future port status: , Wherein, is the predicted status of the port , is the LSTM regression model, is the final output of the GNN.

[0012] As a preferred solution of the method for predicting the shipping container freight rate based on artificial intelligence according to the present invention, wherein: in step S3, the cargo flow prediction model uses reinforcement learning RL to train an intelligent agent, learns the cargo flow patterns under different port states, and combines historical freight rate data to predict the cargo flow status.

[0013] As a preferred solution of the method for predicting the shipping container freight rate based on artificial intelligence according to the present invention, wherein: in step S3, the steps of constructing the cargo flow prediction model are The cargo flow status is modeled by a Markov decision process MDP, and the state space is defined as : , Among them, is the cargo flow state at time moment, is the congestion index vector of all ports, is the cargo flow path information, is the historical freight rate data, Define the action space of the agent's decision-making as : , Among them, is the allocation ratio of goods on each shipping route, is the goods reallocation strategy, Define the revenue function of cargo flow as : , Among them, is the reward at time moment, is the revenue of port for handling goods, is the congestion cost of port , is the weight parameter, and the weight parameter is optimized by gradient descent: , Among them, is the historical revenue data, represents the optimization variable of the cargo flow revenue weight parameter, represents the optimization variable of the port congestion cost weight parameter; Use the deep Q-network DQN for training, and the Q-value update formula: , Among them, is the Q-value of the current state-action pair, is the learning rate, is the discount factor, is the maximum Q-value of the next state, The agent optimizes the cargo flow path according to the port state and freight rate data: , Among them, is the predicted cargo flow path.

[0014] As a preferred embodiment of the artificial intelligence-based shipping container freight rate prediction method of the present invention, in step S4, the freight rate prediction model uses the attention mechanism to assign dynamic weights to the port status in step S2, the freight flow prediction result in step S3, and the route capacity data.

[0015] As a preferred embodiment of the artificial intelligence-based shipping container freight rate prediction method of the present invention, in step S4, the steps of constructing a freight rate prediction model for freight rate prediction are as follows: Define the input features of the freight rate prediction : , where represents the feature vector at time , define as the predicted port status value calculated in step S2, and define as the predicted freight flow value calculated in step S3, as the route capacity data, Use the attention mechanism for feature weighting and calculate the weights: , where represents the dynamic weight of feature at time , is the learnable parameter of the attention mechanism, is the value of feature at time , and the subscript represents the dimension of the parameter matrix of the attention mechanism. After feature weighting, a new feature vector is obtained: , where represents the weighted feature vector; Use the long short-term memory network LSTM for freight rate prediction and define the hidden state update: , , where represents the LSTM hidden state at time , is the weight parameter of the LSTM network, represents the predicted freight rate at time , is the output layer parameter, is the LSTM activation function, and the Sigmoid activation function is used; The mean squared error (MSE) is used as the loss function of the freight rate prediction model to train the freight rate prediction model: , where is the historical actual freight rate data at time .

[0016] As a preferred solution of the artificial intelligence-based shipping container freight rate prediction method described in the present invention, in step S5, the steps of combining market supply and demand data and fuel price data and dynamically adjusting the parameters of the freight rate prediction model using the Bayesian optimization method are as follows: Define the final freight rate prediction input by integrating market supply and demand data and fuel price data : , where represents the final freight rate prediction input at time , define to represent the predicted freight rate calculated in step S4, represents the market supply and demand data, represents the fuel price data; Dynamically adjust the model parameters using Bayesian optimization to minimize the prediction error: , where represents the optimized model parameters, represents the expected value of the mean squared error (MSE), and optimize the model based on historical freight rate data; The final freight rate prediction model after adjusting the parameters based on Bayesian optimization is: , where represents the final predicted freight rate at time , represents the trained LSTM prediction model, and the parameters have been adjusted by Bayesian optimization; The final calculation formula of the LSTM prediction model is expanded as: , , where represents the optimized LSTM hidden state, is the LSTM network weight after Bayesian optimization, is the parameter of the finally optimized output layer.

[0017] The beneficial effects of the present invention are as follows: In the present invention, a port state prediction model is constructed using a graph neural network (GNN). Based on port throughput, ship queuing situation, berth utilization rate, and transportation scheduling situation, the port congestion index is calculated. By constructing a port network graph and describing the connection relationship between ports with an adjacency matrix, information propagation is carried out using GNN, and the updated port features are calculated. The LSTM regression model is trained with the port congestion index to predict the future port state, thereby dynamically capturing the mutual influence between ports and improving the prediction accuracy.

[0018] In the present invention, based on the predicted port state, a cargo flow prediction model is constructed by training an agent using reinforcement learning (RL). The Markov decision process (MDP) is used to define the cargo flow state, and the deep Q-network (DQN) is used to optimize the cargo flow path, enabling the agent to adjust the cargo flow strategy under different port states, thus adapting to market changes and improving the prediction reliability.

[0019] In the present invention, a freight rate prediction model based on the attention mechanism is constructed. The attention mechanism assigns dynamic weights to port state, cargo flow prediction results, and route capacity data to capture the influence of different factors on the freight rate. Subsequently, the time series characteristics are modeled through an LSTM network to achieve more accurate freight rate prediction. Based on the freight rate prediction, combined with market supply and demand data and fuel price data, the Bayesian optimization method is used to dynamically adjust the model parameters to minimize the mean square error (MSE) and improve the adaptability and stability of the prediction results.

[0020] In summary, the present invention effectively solves the defects of traditional methods that rely on lagged data, do not fully consider the impact of port congestion, emergencies, and market dynamic changes, enabling the freight rate prediction to more accurately reflect the complex relationships among market supply and demand, port state, and cargo flow, improving the response ability to short-term freight rate fluctuations, and providing a more reliable decision-making basis for the shipping market. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0022] Figure 1 It is a schematic flow chart of the shipping container freight rate prediction method based on artificial intelligence of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings in the specification.

[0024] In the following description, many specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Persons skilled in the art may make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0025] Secondly, as used herein, "one embodiment" or "an embodiment" refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other.

[0026] Embodiment 1, referring to Figure 1 , this embodiment provides an artificial intelligence-based shipping container freight rate prediction method, including the following steps: Step S1, collect shipping data and construct a standardized data set; The shipping data includes ship trajectory data, port berth data, transportation scheduling data, historical freight rate data, and real-time weather information; In step S1, clean, deduplicate, and normalize the shipping data to construct a standardized data set; The ship trajectory data includes the position, speed, heading, and destination port information of the ships on the route, which is obtained by the Automatic Identification System (AIS), and the ship trajectory data is calibrated by combining the data of the Vessel Traffic Service (VTS); The port berth data includes the real-time berth occupancy, estimated berthing time, and operation completion time, which are collected by the Port Management System (PMS) and the Port Operation Scheduling System (TOS); The transportation scheduling data includes route adjustments, liner delays, and changes in shipping capacity; Step S2, based on the standardized data set, construct a port status prediction model to predict the port status; In step S2, use the Graph Neural Network (GNN) to analyze the port throughput, ship queuing situation, berth utilization rate, and transportation scheduling situation, calculate the port congestion index, and predict the port status based on the port congestion index; In step S2, the steps of using the Graph Neural Network (GNN) to analyze the port throughput, ship queuing situation, berth utilization rate, and transportation scheduling situation, calculate the port congestion index, and predict the port status are as follows: Construct a port network graph, with ports as nodes and shipping routes as edges, and use the Graph Neural Network (GNN) for feature extraction and prediction, Define the port network graph as : , Among them, represents the set of port nodes, represents the th port, is the total number of ports, represents the set of edges, represents the shipping connection between port and port , represents the port feature matrix, represents the eigenvector of port , Extract the port features and define the eigenvector of port as : , Among them, represents the port throughput, represents the number of ships in the queue, represents the berth utilization rate, represents the amount of transportation scheduling adjustment, Define the adjacency matrix : If there is a direct route between port and port , , otherwise , Among them, is the -dimensional adjacency matrix, represents the route relationship between port and , Use GNN to propagate the port status information and calculate the updated port features: , Among them, represents the port features of the th layer, represents the port features of the th layer, represents the neighbor set of port , is the degree of port , is the degree of port , is the trainable weight matrix of GNN, is the activation function, and ReLU is used for non-linear mapping, Calculate the port congestion index, and define the port congestion index as : , where represents the congestion index of the port , is the corresponding weight parameter, The weight parameter is optimized by least squares regression, and the formula is: , where is the true congestion index calculated from historical data, Predict the port status based on the port congestion index. Use the port congestion index as the input to train a regression model to predict the future port status: , where is the predicted status of the port , is the LSTM regression model, is the final output of the GNN; Specifically, in this step, a port status prediction model based on the graph neural network GNN is constructed. The port network diagram is used for modeling. The port throughput, ship queuing situation, berth utilization rate, and transportation scheduling situation are used as node features, and the connection relationship between ports is described through an adjacency matrix. The model uses the graph convolution GNN propagation mechanism to calculate the port features and finally forms the port congestion index; the port congestion index is used to train the LSTM regression model to predict the future port status, which can dynamically capture the mutual influence between ports and improve the accuracy of port status prediction; Step S3: Based on the prediction result of the port status, construct a cargo flow prediction model to predict the cargo flow status; In step S3, the cargo flow prediction model uses reinforcement learning RL to train an intelligent agent to learn the cargo flow patterns under different port statuses and combines historical freight rate data to predict the cargo flow status; In step S3, the steps to construct the cargo flow prediction model are The cargo flow status is modeled using a Markov decision process MDP, and the state space is defined as : , where is the cargo flow status at time , is the vector of congestion indices of all ports, is the cargo flow path information, is the historical freight rate data, Define the action space for the agent's decision-making as : , where is the allocation ratio of goods on each shipping route, is the goods reallocation strategy, Define the revenue function of goods circulation as : , where is time the reward at time is the revenue of port for handling goods, is the congestion cost of port , is the weight parameter, and the weight parameter is optimized using gradient descent: , where is the historical revenue data, represents the optimization variable of the weight parameter for goods circulation revenue, represents the optimization variable of the weight parameter for port congestion cost; Use the deep Q-network DQN for training, and the Q-value update formula: , where is the Q-value of the current state-action pair, is the learning rate, is the discount factor, is the maximum Q-value of the next state, The agent optimizes the goods circulation path according to the port state and freight rate data: , where is the predicted goods circulation path, Specifically, reinforcement learning RL is used here to construct a goods circulation prediction model. The state space is defined by the Markov decision process MDP, and the agent decides the allocation ratio of goods on each shipping route and the reallocation strategy. The goal of the agent is to maximize the goods circulation revenue function, where the weight parameters of the revenue and port congestion cost are optimized by the gradient descent method to make the prediction results more consistent with the historical data. The agent is trained using the deep Q-network DQN, and the strategy is optimized using the Q-value update formula. Finally, the agent learns the optimal goods circulation mode and combines the historical freight rate data to generate the prediction result of the goods circulation state; effectively adapting to different port states; Step S4: Based on the predicted result of the cargo flow status in step S3, construct a freight rate prediction model to predict the freight rate. In step S4, the freight rate prediction model uses an attention mechanism to assign dynamic weights to the port status in step S2, the predicted result of the cargo flow in step S3, and the route transport capacity data. In step S4, the steps of constructing a freight rate prediction model to predict the freight rate are as follows: Define the input features of the freight rate prediction : , where represents the feature vector at time , define to represent the predicted port status value calculated in step S2, and define to represent the predicted cargo flow value calculated in step S3, to represent the route transport capacity data, Use the attention mechanism for feature weighting and calculate the weights: , where represents the dynamic weight of feature at time , is a learnable parameter of the attention mechanism, is the value of feature at time , and the subscript represents the dimension of the parameter matrix of the attention mechanism, After feature weighting, obtain a new feature vector: , where represents the weighted feature vector; Use a long short-term memory network (LSTM) to predict the freight rate and define the update of the hidden state: , , where represents the LSTM hidden state at time , are the weight parameters of the LSTM network, represents the predicted freight rate at time , are the parameters of the output layer, is the LSTM activation function, and the Sigmoid activation function is used; The mean square error (MSE) is used as the loss function of the freight rate prediction model to train the freight rate prediction model: , where is the historical actual freight rate data at time ; Specifically, the attention mechanism is used here to weight the input features to construct the freight rate prediction model. The input features include the port status prediction result, the cargo flow prediction result, and the route capacity data. The dynamic weights are calculated through the attention mechanism and weighted before being input into the LSTM network for freight rate prediction. The hidden state of the LSTM captures the time series features and finally outputs the predicted freight rate. The mean square error (MSE) loss function is used for model training to minimize the error between the historical freight rate and the predicted freight rate, effectively improving the dynamic adaptability to different factors; Step S5: Based on the freight rate prediction in step S4, combined with the market supply and demand data and the fuel price data, the Bayesian optimization method is used to dynamically adjust the parameters of the freight rate prediction model to determine the final freight rate prediction result; In step S5, the steps of dynamically adjusting the parameters of the freight rate prediction model by combining the market supply and demand data and the fuel price data using the Bayesian optimization method are as follows: Define the final freight rate prediction input by integrating the market supply and demand data and the fuel price data : , where represents the final freight rate prediction input at time , define to represent the predicted freight rate calculated in step S4, represents the market supply and demand data, represents the fuel price data; Use Bayesian optimization to dynamically adjust the model parameters to minimize the prediction error: , where represents the optimized model parameters, represents the expected value of the mean square error (MSE), and the model is optimized based on the historical freight rate data; The final freight rate prediction model after adjusting the parameters based on Bayesian optimization is: , where represents the final predicted freight rate at time , represents the trained LSTM prediction model, and the parameters have been adjusted through Bayesian optimization; The final calculation formula of the LSTM prediction model is expanded as follows: , , wherein, represents the optimized LSTM hidden state, is the LSTM network weight optimized by Bayesian optimization, is the parameter of the finally optimized output layer; Specifically, Bayesian optimization is used here to dynamically adjust the parameters of the freight rate prediction model. Taking market supply and demand data and fuel price data as additional inputs, the network parameters of LSTM are optimized. The objective function of Bayesian optimization is the expected value of the mean square error MSE. The optimal parameters are obtained through iterative optimization. The trained LSTM prediction model uses the optimized parameters for the final freight rate prediction. The prediction results can more accurately reflect the influence of market supply and demand and fuel prices and can more dynamically adapt to market changes.

[0027] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A shipping container freight rate prediction method based on artificial intelligence, characterized in that: including Step S1: Collect shipping data and construct a standardized data set; Step S2: Based on the standardized data set, construct a port status prediction model to predict the port status; Step S3: According to the prediction result of the port status, construct a cargo flow prediction model to predict the cargo flow status; Step S4: Based on the cargo flow status prediction result of Step S3, construct a freight rate prediction model to predict the freight rate; Step S5: On the basis of the freight rate prediction in Step S4, combine market supply and demand data and fuel price data, and use the Bayesian optimization method to dynamically adjust the parameters of the freight rate prediction model to determine the final freight rate prediction result.

2. The method for predicting shipping container freight rates based on artificial intelligence according to claim 1, wherein: The shipping data includes ship trajectory data, port berth data, transportation scheduling data, historical freight rate data and real-time weather information; In Step S1, clean, de-duplicate and normalize the shipping data to construct a standardized data set.

3. The method for predicting the freight rate of shipping containers based on artificial intelligence according to claim 2, characterized in that: The ship trajectory data includes the position, speed, course and destination port information of the ships on the route, which is obtained by the Automatic Identification System (AIS), and the ship trajectory data is calibrated by combining the data of the Vessel Traffic Service (VTS); The port berth data includes the real-time berth occupancy, estimated berthing time and operation completion time, which are collected by the Port Management System (PMS) and the Port Operation Scheduling System (TOS); The transportation scheduling data includes route adjustment, liner delay and capacity change.

4. The method for predicting the shipping container freight rate based on artificial intelligence according to claim 3, wherein: In Step S2, use the Graph Neural Network (GNN) to analyze the port throughput, ship queuing situation, berth utilization rate and transportation scheduling situation, calculate the port congestion index, and predict the port status based on the port congestion index.

5. The method for predicting shipping container freight rates based on artificial intelligence according to claim 4, characterized in that: In Step S2, the steps of using the Graph Neural Network (GNN) to analyze the port throughput, ship queuing situation, berth utilization rate and transportation scheduling situation, calculate the port congestion index, and predict the port status are as follows: Construct a port network graph, with ports as nodes and shipping routes as edges, and use the Graph Neural Network (GNN) for feature extraction and prediction; Define the port network diagram as : ， Among them, represents the set of port nodes, represents the th port, is the total number of ports, Represents the edge set, represents the port and the port the shipping connection between them Indicates the port feature matrix, Represents the port Characteristic vector of Extract port features and define the port The feature vector of is ， Among them, represents the port throughput, represents the number of ships queuing, represents the berth utilization rate, represents the adjustment amount of transportation scheduling, Define the adjacency matrix : If there is a port and the port has a direct shipping route, otherwise , Among them, is a dimensional adjacency matrix Indicates a port and the shipping route relationship between Use GNN for port status information propagation and calculate the updated port features: ， Among them, represents the port features of the th layer, represents the port features of the th layer, represents the neighbor set of the port , is the degree of the port , is the degree of the port , is the trainable weight matrix of the GNN, is the activation function, and ReLU is used for non-linear mapping, Calculate the port congestion index, and define the port congestion index as : ， Among them, represents the congestion index of the port , and is the corresponding weight parameter. Weight parameter Optimized by least squares regression, the formula is: ， Among them, is the true congestion index calculated from historical data. Predict the port status based on the port congestion index, use the port congestion index as input, and train a regression model to predict the future port status; ， Among them, is the prediction status of the port , is the LSTM regression model, is the final output of the GNN.

6. The method for predicting shipping container freight rates based on artificial intelligence according to claim 5, characterized in that: In Step S3, the cargo flow prediction model uses Reinforcement Learning (RL) to train an agent to learn the cargo flow patterns under different port states, and combines historical freight rate data to predict the cargo flow status.

7. The method for predicting the shipping container freight rate based on artificial intelligence according to claim 6, characterized in that: In Step S3, the steps of constructing the cargo flow prediction model are as follows: The cargo flow status is modeled using the Markov Decision Process (MDP), and the state space is defined as : ， Among them, is the cargo flow status at time moment, is the congestion index vector of all ports, is the cargo flow path information, is the historical freight rate data, Define the action space for the agent's decision-making as : ， Among them, is the allocation ratio of goods on each shipping route, is the goods reallocation strategy, Define the revenue function for the flow of goods as : ， wherein, is the reward at time moment, is the revenue from handling goods at the port , is the port congestion cost, is the weight parameter, and the weight parameter is optimized by gradient descent: ， Among them, is historical revenue data, represents the optimization variable of the cargo turnover revenue weight parameter, represents the optimization variable of the port congestion cost weight parameter; Use the Deep Q-Network (DQN) for training, and the Q-value update formula: ， wherein, is the Q value of the current state-action pair, is the learning rate, is the discount factor, is the maximum Q-value for the next state, The agent optimizes the cargo flow path according to the port status and freight rate data; ， Among them, is the predicted goods flow path.

8. The method for predicting shipping container freight rates based on artificial intelligence according to claim 7, characterized in that: In Step S4, the freight rate prediction model uses the attention mechanism to assign dynamic weights to the port status in Step S2, the cargo flow prediction result in Step S3 and the route capacity data.

9. A method for predicting shipping container freight rates based on artificial intelligence according to claim 8, characterized in that: In Step S4, the steps of constructing the freight rate prediction model to predict the freight rate are as follows: Define the input features for freight rate forecasting : ， Among them, represents the feature vector at a specific time, and is defined as represents the predicted port state value calculated in step S2, and is defined as represents the predicted cargo flow value calculated in step S3, represents the route capacity data, Use the attention mechanism for feature weighting and calculate the weights: ， Among them, represents the feature at time the dynamic weight at the moment, is a learnable parameter of the attention mechanism, is the time feature at the moment value, and the subscript represents the dimension of the parameter matrix of the attention mechanism After feature weighting, obtain a new feature vector: ， Among them, represents the weighted feature vector; Use the Long Short-Term Memory Network (LSTM) for freight rate prediction, and define the update of the hidden state: ， ， Among them, represents the LSTM hidden state at the time instant, is the weight parameter of the LSTM network, represents the predicted freight rate, is the output layer parameter, is the LSTM activation function, and the Sigmoid activation function is adopted; The mean square error (MSE) is used as the loss function of the freight rate prediction model to train the freight rate prediction model: ， Among them, is the time historical actual freight rate data at the moment.

10. The method for predicting shipping container freight rates based on artificial intelligence according to claim 9, wherein: In step S5, the steps of dynamically adjusting the parameters of the freight rate prediction model by combining market supply and demand data and fuel price data are as follows: Define the final freight rate prediction input by integrating market supply and demand data and fuel price data : ， Among them, represents the final freight rate prediction input at the time point, which is defined as represents the predicted freight rate calculated in step S4, represents the market supply and demand data, represents the fuel price data; Use Bayesian optimization to dynamically adjust the model parameters and minimize the prediction error: ， Among them, represents the optimized model parameters, Represents the expected value of the mean squared error (MSE), and optimizes the model based on historical freight rate data; The final freight rate prediction model based on Bayesian optimization for parameter adjustment is: ， Among them, represents time the predicted final freight rate, represents the trained LSTM prediction model, and the parameters have been adjusted through Bayesian optimization; The final calculation formula of the LSTM prediction model is expanded as: ， ， Among them, represents the optimized LSTM hidden state, is the weight of the LSTM network optimized by Bayesian optimization, is the parameter of the finally optimized output layer.

Images