Intelligent shipping method for port ship

Abstract

The invention discloses an intelligent shipping method for a port ship. The intelligent shipping method comprises the following steps: acquiring information data; dynamically predicting the arrival time of the ship in real time; building a port berth real-time dynamic distribution model based on ship arrival time prediction; solving a port berth real-time dynamic distribution model; and outputting a real-time dynamic port ship berthing result. According to the method, whether the tide and the berthing position meet the ship draft and other constraints are considered in berth dynamic allocation, the berth allocation strategy of the optimal berthing position is designed, the berth utilization rate is increased, the port efficiency is improved, and the method has the advantages of being high in universality and wide in coverage; according to the method, the complex relation between the latitude, the longitude, the COG, the SOG and other characteristics and the arrival time is directly estimated through the machine learning model, real-time prediction of the arrival time is achieved by inputting AIS data of the latest period of time, therefore, dynamic characteristic changes are captured more efficiently, prediction precision is high, convergence is fast, and the workload is small.

Description

Technical Field

[0001] The present invention relates to the technical field of port management, in particular to an intelligent berthing method for port ships. Background Art

[0002] As a comprehensive transportation hub for maritime transport, ports play an important role in the development of the marine economy. The increase in container throughput and the trend towards larger ships have brought a series of challenges to ports. One of the main challenges faced by ports and terminals when formulating berth allocation strategies is the uncertainty of the estimated time of arrival (ETA) of ships. Since berth allocation usually needs to be planned before the ship arrives at the port, actual ship delays will lead to berth vacancies, and early arrivals of ships will also cause port congestion. The uncertainty of ship arrival times not only affects the effective implementation of berth plans, resulting in waste of resources and economic losses, but also causes a decrease in port berth utilization, which indirectly affects port management efficiency. Therefore, accurate prediction of ship arrival times is crucial for berth allocation optimization. At present, many countries around the world, including developed countries, are generally building smart ports, and one of the core applications of building smart ports is the intelligent berthing method for port ships.

[0003] Existing research primarily focuses on predicting ship arrival times or unilaterally optimizing ship scheduling, which is somewhat unreliable. In practice, the uncertainty of ship arrival times can significantly impact intelligent port scheduling. When ships arrive early, they cannot berth according to the port's production schedule. When ships arrive late, berths become unused and wasteful. Both situations result in significant losses for ports and shipping companies. Therefore, developing intelligent port berthing plans based on improved accuracy of ship arrival times is crucial, but this also presents challenges for maritime authorities and port companies. Summary of the Invention

[0004] In order to solve the above problems existing in the prior art, the present invention proposes an intelligent berthing method for ships in ports, which can improve the berth utilization rate and enhance the management efficiency of ports.

[0005] In order to achieve the above object, the technical solution of the present invention is as follows: A method for intelligent berthing of a port ship comprises the following steps:

[0006] A. Collect information data

[0007] Information data is collected through the port network. The collected information data includes AIS data and ship information data. The collected information data is divided into dynamic information, static information and voyage information. The dynamic information is real-time motion state parameters, including longitude and latitude, speed and heading; the static information is ship attribute parameters, including the Maritime Mobile Service Identity (MMSI), ship type, length and width; the voyage information is voyage-related parameters, including departure port, destination port and estimated arrival time;

[0008] Preprocess the collected information data.

[0009] B. Real-time dynamic prediction of ship arrival time

[0010] Based on the collected information data, the ship's arrival time is predicted by a virtual induced route network composed of ship sinking points, navigation marks, maritime regulatory waterways and dynamic buoys.

[0011] C. Establish a real-time dynamic allocation model for port berths based on ship arrival time prediction

[0012] Considering the real-time dynamic allocation problem of port berths under tidal constraints and dynamic ship arrivals, a real-time dynamic allocation model of port berths is established based on the prediction of ship arrival time. The berth allocation plan is optimized by reasonably arranging the entry and exit sequence, berthing time and location of arriving ships.

[0013] D. Solve the real-time dynamic allocation model of port berths

[0014] Genetic algorithm (GA) is used to solve the real-time dynamic allocation model of port berths.

[0015] E. Output real-time dynamic port ship berthing results

[0016] Output real-time dynamic port ship berthing results to provide decision-making information for maritime departments and port management personnel.

[0017] Furthermore, the method for predicting the real-time dynamic ship arrival time in step B comprises the following steps:

[0018] B1. Gradient Boosting

[0019] Let F m (x) is the integrated model of gradient boosting decision tree, f t (x) is the sub-model of the t-th iteration, that is, the weak regression tree or the residual of the fit, then the integrated model of the gradient boosting decision tree is expressed as follows:

[0020]

[0021] Where: is the residual fitting coefficient, t is the residual fitting coefficient number, t=0,1,2,…,m.

[0022] Assume that the loss function after the last residual fitting is L[F m (x), Y], each time a sub-model is added, the following formula is satisfied:

[0023] L[F m (x),Y]<L[F m-1 (x),Y]

[0024] Where: x is the feature vector, Y is the set of true values ​​of the sample scalar, F m-1 (x) is the ensemble model of the penultimate gradient boosting decision tree.

[0025] B2. Predicted ship arrival time

[0026] Assume that the dataset sample used in training is the Qth feature vector x Q and the true value of the Bth sample scalar y B , where x Q ∈R and y B ∈R, the lightweight gradient boosting machine regression prediction model is a strong regression tree through the linear combination of K weak regression trees, and the output is the predicted arrival time of the i-th ship Calculated as follows:

[0027]

[0028] in, Represents input x i The predicted ship arrival time is k, which represents the number of decision trees in the lightweight gradient boosting machine regression prediction model, and f k represents the kth basic classification and regression tree, namely CART tree, and F is the function set of all CART trees.

[0029] Furthermore, the steps of establishing the real-time dynamic allocation model of port berths in step C are as follows:

[0030] C1. Determine the operating conditions of the real-time dynamic allocation model of port berths

[0031] C11. All ships must pass through the channel at a constant time and are not allowed to dock or wait at will. When leaving the channel, they must immediately berth according to the berthing plan.

[0032] C12. The ship's head direction is pointing to the 0 scale direction of the shoreline.

[0033] C13. Large ships need to take advantage of the tides to enter and exit the port, but small ships do not.

[0034] C14. To prevent collision during navigation, the two vessels in front and behind should maintain a safe distance when sailing.

[0035] C15. Does not involve specific loading and unloading operations of ships.

[0036] C16. Vessels shall not be dispatched repeatedly during the planning period.

[0037] C17. No berthing operation will be performed after the vessel berths.

[0038] C2. Establish a real-time dynamic allocation model for port berths

[0039] The objective function of the real-time dynamic allocation model of port berths is expressed as follows:

[0040]

[0041] Constraints:

[0042]

[0043]

[0044] 0≤x i ≤Ll i i∈I (14)

[0045]

[0046] x i -x j +l i ≤(1-y ij )Li∈I,j∈I,i≠j (16)

[0047] x ij +x ji +y ij +y ji ≥1i∈I,j∈I,i≠j (17)

[0048] x ij +x ji ≤1i∈I,j∈I,i≠j (18)

[0049] y ij +y ji ≤1i∈I,j∈I,i≠j (19)

[0050] Where: I represents the set of |I| container ships arriving at the port; i represents the ship number, j represents the serial number of another ship; I in Indicates that ships need to enter the port during the high tide. outIndicates that ships need to leave the port during the high tide; L is the length of the coastline; n is the total number of times a set of high and low water levels occurs; V is the duration of the high or low water level; T is the total planning time, T = 2Vn; a i Indicates the estimated arrival time of ship i; l i represents the length of ship i; h i represents the operation time of ship i; C represents the time required to enter or exit the channel; represents the starting position of ship i suitable for berthing; Indicates the end position of ship i suitable for berthing; M is a sufficiently large number; x i Indicates the starting position of the berthing operation of ship i; Indicates the moment when the waiting ship i enters the channel; Indicates the time when the operation vessel i enters the channel; Indicates the start time of berthing operation for ship i; Indicates the time when ship i ends its operation; represents the time when ship i leaves the port; in Indicates that the value is 1 when ship i enters the channel for berthing at the nth high tide, otherwise it is 0; μ in The value is 1 when ship i completes the operation and leaves the channel at the nth high tide, otherwise it is 0; i,b If ship i is docked at shoreline b, it is 1, otherwise it is 0; ij In the two-dimensional graph, the value is 1 when ship i is completely below ship j, otherwise it is 0; ij In the two-dimensional graph, the value is 1 when ship i is completely to the left of ship j, and 0 otherwise.

[0051] Furthermore, the steps of the genetic algorithm in step D are as follows:

[0052] D1. Chromosome encoding

[0053] According to the characteristics of the berth allocation problem, integer coding is used to generate the initial solution of the real-time dynamic allocation model of port berths. Each chromosome represents the berthing service order of ships. Each chromosome consists of XN genes, where I represents the number of ships. The genes on each chromosome of the initial population are randomly generated.

[0054] D2. Generate initial solution

[0055] A reasonable berth allocation scheme is constructed based on the chromosomes in the population. The berthing time and berth position of each ship in the port channel affected by tidal constraints and berthing water depth are determined, and then the fitness function is calculated. The specific steps are as follows:

[0056] D21. Set the time when ship i starts entering the channel as the estimated arrival time, and determine whether the ship needs to enter the port during the high tide. If the conditions for entering the port are met, execute step D22; otherwise, wait for the start time of the next high tide to enter the port.

[0057] D22. Set the berthing position of ship i to the starting position that meets the berthing conditions. The berthing time is the sum of the time of entering the channel and the time of passing the channel. Determine whether it overlaps with other ships on the berthing time two-dimensional graph. If there is no overlap, execute step D23. Otherwise, increase the berthing position by one unit until it moves to the end position of the berthing conditions. If it still overlaps with other ships, increase the time by one unit on the time axis and return to step D22.

[0058] D23. After the ship operation is completed, it is determined whether ship i needs to leave the port with the tide. If the conditions for leaving the port are met, step D24 is executed; otherwise, the ship waits at the berth for the start of the next high tide to leave the port.

[0059] D24. Let i = i + 1 and search for the optimal berth for the next ship.

[0060] D3. Calculate the fitness function

[0061] The inverse of the objective function is used as the fitness function, meaning that lower fitness values ​​represent better berth allocation plans. This ensures that individuals with shorter waiting times have higher fitness values, making them more likely to be retained and propagated during the genetic process, thereby guiding the algorithm to converge towards the optimal solution.

[0062] D4. Perform selection operation

[0063] First, using an elite retention strategy, the fitness values ​​of all chromosomes in the current population are calculated and sorted in ascending order. Chromosomes with higher fitness are retained according to the set elite ratio and directly copied to the next generation, ensuring that the best individuals in the next generation are even better than those in the previous generation. Subsequently, a roulette wheel selection method is used for the remaining individuals, with individuals with higher fitness having a higher probability of being selected. The selected individuals undergo crossover and mutation operations, and elite individuals are also inserted into these operations to ensure that the population size remains at the preset number.

[0064] D5. Perform cross operation

[0065] The two-point crossover operator is used to perform a crossover operation on chromosomes. Two parent chromosomes are selected, two crossover points are randomly generated, and the chromosome segments between the two crossover points are exchanged while retaining and mapping the remaining genes to ensure the validity of the chromosomes, thereby generating two new daughter chromosomes.

[0066] D6. Perform mutation operations

[0067] During berth allocation, the docking layout of the leading ship affects the berth allocation of the trailing ship. A better berth allocation plan means a more reasonable layout for the leading ships. Therefore, mutation operations are performed sequentially on the chromosomes that have undergone elite retention and crossover operations. Mutation operations retain a certain percentage of genes from the parent chromosome, while the remaining genes are randomly generated to form a new chromosome.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] 1. The present invention takes into account the constraints of tides and whether the berthing position meets the ship's draft depth in the dynamic allocation of berths, designs a berth allocation strategy for the optimal berthing position, improves berth utilization, improves port efficiency, and has the characteristics of strong versatility and wide coverage;

[0070] 2. This paper proposes a method that uses a machine learning model to directly estimate the complex relationship between latitude, longitude, course over ground (COG), speed over ground (SOG), and arrival time. This method uses the latest AIS data as input to achieve real-time arrival time prediction, effectively capturing dynamic feature changes, achieving high prediction accuracy, fast convergence, and minimal workload.

[0071] 3. Since the present invention uses genetic algorithms to solve the real-time dynamic allocation model of port berths, it can provide accurate intelligent berthing plans for port ships, which is highly scientific, has high calculation accuracy and good reliability;

[0072] 4. Since the present invention utilizes the relevant optimization methods of operations research to solve the intelligent berthing plan of port ships, it can effectively reduce the operating costs of shipping companies and has good economic efficiency.

[0073] 5. By adopting the intelligent berthing method for port ships proposed in the present invention, maritime administration departments and port enterprises can dynamically plan the production plan of ships in the port area, thereby achieving macro-management of the ships in the entire port. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The present invention has the following Figure 18 Zhang, among which:

[0075] Figure 1 It is the technical roadmap of the present invention.

[0076] Figure 2 This is a satellite image of the port area.

[0077] Figure 3 It is a prediction flow chart.

[0078] Figure 4 It is a schematic diagram of the time window when ships entering and leaving the port are affected by tides.

[0079] Figure 5 This is a schematic diagram of chromosome encoding.

[0080] Figure 6 It is a flow chart of berth allocation strategy.

[0081] Figure 7 This is a schematic diagram of intersection selection.

[0082] Figure 8 This is a schematic diagram of gene segment exchange.

[0083] Figure 9 This is a schematic diagram of the mapping relationship between genes.

[0084] Figure 10 This is a schematic diagram of the crossover results.

[0085] Figure 11 It is a schematic diagram of mutation operation.

[0086] Figure 12 It is a comparative diagram of different distance intervals and models.

[0087] Figure 13 This is a schematic diagram of the comparison of prediction results.

[0088] Figure 14 It is a schematic diagram of eigenvalues ​​in the LightGBM model.

[0089] Figure 15 It is the coordinated scheduling convergence graph.

[0090] Figure 16 is the optimal berth allocation map.

[0091] Figure 17 This is the FCFS berth allocation map.

[0092] Figure 18 This is a schematic diagram of the impact of ETA prediction error on berth allocation. DETAILED DESCRIPTION

[0093] The present invention will be further described below in conjunction with the accompanying drawings. Figure 1 As shown, a method for dynamic berthing of port ships is suitable for both inland ports and coastal ports, including the following steps:

[0094] 1. Information data collection

[0095] The present invention collects basic information of a port, such as Figure 2As shown. The container terminal of this port has a continuous berth with a coastline length of 1652 meters. Ships can dock at any position. The continuous berth space is discretized and divided into multiple stages. The minimum unit length is set to 50 meters, and 1652 meters is discretized into 33 units. 2880 minutes is the set planning period (48 hours). Therefore, the space-time range composed of the berth and time of the example of the present invention is 33×2880. The present invention designates 1 to 10 unit berth lengths as special berths that meet the berthing requirements of large tide-riding ships, while ordinary ships are not restricted by berth types. The time required for a ship to pass through the channel is 1 hour. Considering that the tidal cycle is 12 hours, in order to ensure the safety and smoothness of the ship berthing, ships 2 and 15 must carry out berthing and unberthing operations within the high tide time window.

[0096] The present invention also collected and analyzed the AIS data of container ships arriving at the port from May 27 to June 27, 2024. The AIS data of ships are mainly collected through shore-based facilities and ship-borne equipment. The shore-based facilities are responsible for receiving and storing the AIS message information sent by the ship, while the ship-borne equipment is responsible for sending this information. These data can be used to query the position and trajectory of the ship in real time. According to the nature of the data, AIS data can be divided into three categories: dynamic information, static information and voyage information. Among them, dynamic information includes real-time motion status parameters such as longitude and latitude, speed, and heading; static information includes parameters describing the attributes of the ship such as the Maritime Mobile Communication Service Identification Code (MMSI), ship type, length and width of the ship; voyage information covers voyage-related parameters such as departure port, destination port, and estimated arrival time, as shown in Table 1.

[0097] Table 1 Basic information of ships

[0098]

[0099]

[0100] 2. Real-time dynamic ship arrival time prediction

[0101] (1) Data preprocessing

[0102] In order to ensure the quality and reliability of AIS data, the raw data needs to be preprocessed. The specific processing methods are as follows:

[0103] 1) Vessel arrival track screening

[0104] When using MMSI to filter data for the same ship, all AIS data points marked with this MMSI will be grouped together. These data points contain the ship's trajectory information at different departure ports and destination ports. Based on the AIS information providing the ship's departure and destination port information, dynamic information between the same departure and destination ports can be filtered out.

[0105] 2) Duplicate value processing

[0106] For duplicate data, they are grouped according to MMSI. By comparing the time and geographical information of two adjacent data items in each voyage, if they are the same, the duplicate data is deleted and only one record is saved.

[0107] 3) Error value handling

[0108] According to statistical analysis, the speed of container ships is basically distributed between 0 and 25 knots. Speeds exceeding 25 knots are considered outliers. For numerical anomalies in the original data, the present invention uses rule verification to handle them. The rule criteria are shown in the last column of Table 2. Values ​​exceeding their attribute range are considered outliers and need to be deleted.

[0109] 4) Missing value processing

[0110] For missing data, this paper uses mean filling to fill missing values ​​in SOG, COG, and heading. Missing values ​​in vessel status are filled with the most common '0' state. Considering that stationary anchored vessels are not helpful for analyzing navigation characteristics, further data screening is performed by filtering out vessels with speeds below 1 knot as anchored vessels, thereby retaining only dynamic AIS data in navigational status in the dataset, as shown in Table 2.

[0111] Table 2 AIS data attributes and abnormal data judgment rules

[0112]

[0113] Data processing is a key step in the data analysis process. It can obtain more accurate data mining results and improve the quality of the researched data. In the establishment of a predictive model, the processed data can improve the accuracy of the model and the reliability of the prediction.

[0114] 5) Feature Engineering

[0115] Feature engineering refers to the process of extracting effective features from raw data to improve the model's prediction ability. In the present invention, the actual arrival time (ATA) minus the current timestamp is used as the output variable of the prediction model, and attributes such as the ship's position longitude, ship's position latitude, ground speed SOG, ground course COG, ship length (Length) and ship width (width) are selected as basic features. In addition, according to the longitude and latitude of the two trajectory points, the distance between the two points on the sphere is calculated, and then the average speed and the remaining distance to the destination are constructed to predict the characteristics of the ship's arrival time. The processed partial feature data table is shown in Table 3. Finally, the data set contains 8 input features and 1 output feature, and the specific variable information is shown in Table 4.

[0116] Table 3 Variable information in the dataset

[0117]

[0118] Table 4 Partial characteristic data

[0119]

[0120]

[0121] (2) Prediction method

[0122] ETA is an important basis for container terminals to formulate production operation plans. It is necessary to estimate the arrival time of each ship within the planning range. Therefore, the ETA prediction of ships is a regression problem. The prediction process is as follows: Figure 3 shown.

[0123] 1) Random forest regression prediction model

[0124] Random Forest (RF) is an integrated learning method. Assume that the training data set is: S n ={(X1,Y1),(X2,Y2),…(X n ,Y n )},X∈R m Represents the feature vector of each sample, containing m features, and Y∈R represents the target value of each sample. First, the bagging algorithm is used to extract the original data set S n Randomly draw q bootstrap sample sets with replacement from , denoted as: A new training set is formed. During the training process, a decision tree is trained for each training set. The decision trees under different training sets are independent of each other, and these decision trees constitute a random forest. Secondly, in the process of constructing the decision tree, a part of the features are randomly selected from all the attribute features of the training set samples to form a random feature subspace, which is used as the split feature set of the current node of the decision tree. The optimal split point is found among these features to divide the data into left and right subtrees. This process is iterated until the predetermined stopping condition is reached. Based on q sample sets, q regression tree models can be generated. Each tree is called a base classifier, which is recorded as formula (20):

[0125] h(X,θ k ) k=1,2…q (20)

[0126] Among them, h(X,θ k ) represents the prediction function of a tree, X is the input feature vector, which represents the feature data you use to predict, θ kIt is the parameter of the tree, which describes the structure of the tree. Each tree is trained based on different bootstrap samples, and its parameter θ k are different. The ensemble algorithm is used to make independent predictions for each tree in the training set. The prediction results are shown in formula (21):

[0127]

[0128] Finally, the predicted arrival time of the ship is output by taking the mean of all the regression tree prediction results and aggregating them. Formula (22):

[0129]

[0130] 2) Light Gradient Boosting Machine (LightGBM) method

[0131] LightGBM is a machine learning algorithm based on gradient boosting decision tree. Its basic idea is to linearly combine K weak regression trees into a strong regression tree. Assume that the dataset sample used in training is the Qth eigenvector x Q and the true value of the Bth sample scalar y B , where x Q ∈R and y B ∈R, Y is the set of sample scalar true values, and the model includes K weak regression trees. The prediction model output is as follows:

[0132]

[0133] in, Represents input x i The predicted arrival time of ship i, K represents the number of decision trees in the model, f k represents the kth basic CART tree, and F is the function set of all CART trees. GBDT has the functional characteristics of gradient boosting and decision tree, and has the advantages of good training effect and low overfitting. Assume f i (x) is the sub-model of the i-th iteration (weak regression tree, i.e., the residual of the fitting), and the integrated model is formula (24):

[0134]

[0135] The loss function is L[F m(x), Y], each time a new sub-model is added, the loss function is continuously optimized. The prediction model will select those variables with higher information content to update the model based on the current loss function and the gradient information of the features, so as to more effectively reduce the value of the loss function. In this way, the model can gradually improve its prediction ability and ultimately achieve better performance. Each time a sub-model is added, formula (25) is satisfied:

[0136]

[0137] 3) K-Nearest Neighbors (KNN) algorithm

[0138] KNN algorithm is an instance-based non-parametric learning method. Let the training set T = {(x1,y1),(x2,y2),...,(x N ,y N )}, where each eigenvector x i =(x i1 ,x i2 ...,x in ) contains m features, and the corresponding output is the arrival time y of the i-th ship i , N is the total number of feature vectors. When a given query feature vector x is input, first calculate x and each training feature vector x i The Euclidean distance is calculated as follows:

[0139]

[0140] Then sort the distance values ​​from small to large. If a distance is ranked nth, the corresponding instance is called the nth nearest neighbor of x, and its output is y n (x), the KNN regression prediction formula is as follows (27):

[0141]

[0142] The KNN weighted regression prediction formula is as follows (28):

[0143]

[0144] in, is the predicted ship arrival time, y i is the true numerical label of the k nearest neighbors, w i is the sample weight, formula (27) means taking the average of the k nearest neighbor sample labels as the predicted value, and formula (28) means taking the weighted average of the k nearest neighbor sample labels as the predicted value.

[0145] 4) ANN regression prediction model

[0146] Artificial Neural Network (ANN) is a nonlinear modeling tool that simulates the structure and working mechanism of human brain neurons. It constructs a function f(·) that transforms the input feature vector x=(x1, x2…, x n ) is mapped to a continuous output variable, i.e., the predicted value of the ship arrival time That is, formula (29):

[0147]

[0148] ANN usually consists of an input layer, several hidden layers, and an output layer. The output of each hidden layer neuron can be expressed as formula (30):

[0149]

[0150] Among them, w ij Represents the input feature vector x i To the hidden layer neuron h j The connection weight, b j is the bias term, and φ(·) is the activation function, commonly used functions include ReLU and tanh. In the regression problem, the output layer is usually a linear activation function, and its output is the predicted ship arrival time of formula (31):

[0151]

[0152] The model is trained using supervised learning. Forward propagation is used to predict the ship's arrival time y, and the mean squared error (MSE) is used as the loss function to measure the difference between the predicted result and the true value. Subsequently, a backpropagation algorithm is used to calculate the gradient of the loss function with respect to the network weights. This is then combined with an optimization algorithm to update the network parameters, thereby continuously optimizing model performance.

[0153] In order to evaluate the accuracy and stability of the model, the present invention selected three indicators: goodness of fit (R2), mean square error (MSE) and mean absolute error (MAE).

[0154] 3. Real-time dynamic allocation model of port berths based on ship arrival time prediction

[0155] (1) Problem description

[0156] The present invention considers the problem of berth allocation for continuous terminals and dynamic ship arrivals under tidal constraints, with the goal of rationally arranging the order of entry and exit, as well as the time and location of berthing, and achieving high service quality to meet the needs of customers at the terminal. The continuous terminal layout allows ships to berth at any location, especially with the trend of large-scale container ships. In busy hub ports, ships of different sizes frequently berth. Continuous berth allocation provides greater flexibility and can more effectively adapt to the needs of various ships, thereby optimizing the berth allocation plan. In addition, tidal factors have a significant impact on the operation of container ports. When large ships pass through the waterway, they must ensure that the water depth of the waterway meets their draft requirements. Figure 4 This study describes the impact of tides on the entry and exit of large vessels. Tides are cyclical, with alternating high and low water level windows of roughly equal duration. When a large vessel enters a port, it must confirm whether the water level meets the required conditions. If the water level meets the requirements, the vessel can enter the port; otherwise, the vessel must wait in an anchorage outside the port for a high water level window and then enter the port at high tide. Taking tides into account can maximize the berthing capacity of large vessels. Estimated arrival times (ETAs) are a crucial basis for terminal operations planning. Shipping companies typically report detailed arrival information to the terminal 24 hours before container ships enter the terminal waters. Therefore, the arrival time of a vessel is uncertain until it actually arrives. However, berth scheduling must be conducted in advance, so only arrival time estimates can be considered. Using the aforementioned regression prediction algorithm, the arrival time prediction results were applied to berth optimization scheduling to verify their effectiveness in real-world applications.

[0157] (2) Determine the operating conditions of the real-time dynamic allocation model of port berths

[0158] The same as “determining the operating conditions of the real-time dynamic allocation model of port berths” in the invention content.

[0159] (3) Establishing a real-time dynamic allocation model for berths

[0160] The present invention takes minimizing the total waiting time of all ships within the planning period as the objective function and constructs a real-time dynamic berth allocation model, which is as described in step C3 of the invention.

[0161] 4.GA optimization algorithm

[0162] Considering that the berth allocation problem is an NP-hard problem, the present invention introduces a genetic algorithm to solve the berth allocation problem.

[0163] (1) Chromosome encoding

[0164] In genetic algorithms, encoding is a key step in representing potential solutions to a problem in the form of chromosomes, which directly affects the algorithm's search capability and solution efficiency. Based on the characteristics of the berth allocation problem, the present invention uses integer encoding to generate the initial solution. Each chromosome represents the order in which ships will be served at berths. Each chromosome consists of I genes, where I represents the number of ships. Figure 5 As shown in Figure 1, this chromosome segment represents the order in which ships arrive at berths. The genes on each chromosome of the initial population are randomly generated. This random initialization method helps maintain the diversity of the population.

[0165] (2) Generate initial solution

[0166] Algorithm flow questions such as Figure 6 As shown, the content is the same as "generating an initial solution" in the invention content.

[0167] (3) Fitness function calculation

[0168] The role of the fitness function is to evaluate the quality of each individual and reflect its survival ability in the population. The higher the fitness value, the better the individual is, and the greater the possibility of being retained and further optimized during the iterative process. The present invention takes minimizing the total waiting time of ships in the port as the optimization goal. In order to make the subsequent selection operations consistent with the optimization goal, the objective function is taken in the form of the inverse as the fitness function, as shown in formula (32), that is, a lower fitness value represents a better berth allocation plan:

[0169]

[0170] Among them, f i Represents the fitness value of chromosome i, Z i Represents the objective function value of the i-th chromosome, and N is the population size.

[0171] (4) Select operation

[0172] The purpose of the selection process is to select high-performing individuals from the current population for participation in the reproduction process. This method employs a combination of elite retention and roulette wheel selection. First, using the elite retention strategy, the fitness values ​​of all chromosomes in the current population are calculated and sorted in ascending order. Chromosomes with high fitness are retained according to a set elite ratio and directly replicated to the next generation, ensuring that the best individuals in the next generation are even better than those in the previous generation. Subsequently, a roulette wheel selection method is used for the remaining individuals, with individuals with higher fitness being more likely to be selected.

[0173] The specific steps of the roulette selection method are as follows:

[0174] 1) Calculate the fitness value f for each individual in the population i (i=1,2,3,...,N, N is the population size).

[0175] 2) The selection probability P of each individual i It is determined by the ratio of its fitness to the total fitness value of all individuals. As shown in formula (33), the higher the fitness of an individual, the larger the "roulette area" it occupies and the higher the probability of being selected.

[0176]

[0177] 3) Calculate the cumulative probability Q i , used for subsequent random selection, as shown in formula (34). At this time, the entire roulette wheel is divided into several intervals, each interval corresponding to an individual.

[0178]

[0179] 4) Generate a random number r between (0,1) and find the value that satisfies Q i-1 ≤r≤Q i When , individual i is selected. Repeat this process until a certain number of individuals are selected.

[0180] (5) Crossover operation

[0181] The crossover operation simulates gene recombination in biological genetics, gradually improving the quality of the solution while maintaining population diversity. The present invention uses a two-point crossover operator to perform a crossover operation on chromosomes. The specific steps are as follows:

[0182] 1) Select two parent chromosomes A and B from the population and randomly generate two crossover points pos1 and pos2, so that 1≤pos1<pos2≤|I|, such as Figure 7 shown.

[0183] 2) Exchange the chromosome segments between the two crossover points to obtain offspring chromosomes A and B, while ensuring that the genes at other positions remain unchanged, such as Figure 8 shown.

[0184] 3) Correction operations may be required after exchanging genes to ensure that the chromosome meets the problem constraints (ships cannot be re-scheduled), so the remaining genes need to be mapped and adjusted, such as Figure 9 shown.

[0185] 4) The final sub-chromosomes are as follows Figure 10 shown.

[0186] (6) Mutation operation

[0187] During the berth allocation process, the berthing layout of the ship in front will affect the berth allocation of the ship behind. A better berth allocation plan means that the layout of the ship in front is more reasonable. Therefore, the chromosomes of elite retention and crossover operations are mutated in turn. The mutation operation will retain a certain proportion of genes in the parent chromosome, and the remaining genes will be randomly generated to form a new chromosome. Figure 11 shown.

[0188] 5. Based on the predicted results of ship arrival time, output real-time dynamic ship berthing results.

[0189] (1) Ship arrival time prediction

[0190] 1) Hyperparameter Search

[0191] The present invention uses the Optuna framework to optimize the key hyperparameters of the model, and selects the root mean square error (RMSE) as the core indicator to measure the performance of the model. The RMSE average value obtained by cross-validation is returned to the main function as the optimization objective function value. After executing the optimization process multiple times, the parameter combination corresponding to the minimum objective function value is selected as the optimal hyperparameter combination of the model. In this way, the performance of the model under different hyperparameter combinations can be comprehensively evaluated. The optimal parameter combinations of each model are shown in Tables 7 and 8.

[0192] Table 7 RF model parameters

[0193]

[0194] Table 8 LightGBM model parameters

[0195]

[0196] 2) Result analysis

[0197] The present invention uses Python's scikit-learn machine learning library to implement the model, and selects two typical and popular regression models, KNN and ANN, for comparison, as shown in Table 9 and Figure 12 The results show that LightGBM has the best prediction effect. This paper uses the LightGBM algorithm to predict the arrival time of ships on June 28, 2024, which is used in the following berth allocation example to provide a more accurate scheduling plan for the port. The comparison between the prediction results and the actual arrival time is shown in the figure below. Figure 13 shown.

[0198] Table 9 Prediction performance of the model on the test dataset

[0199]

[0200] 3) Importance of feature variables

[0201] The feature importance of the trained LightGBM model is analyzed and explained. The value of the feature variable importance of the LightGBM model refers to the number of times a feature is used in the inference step. The higher the score, the more important the feature is to the final prediction (i.e., the greater its contribution). The importance of each feature and its score are shown as follows: Figure 14 shown.

[0202] (2) Intelligent parking result output

[0203] 1) Intelligent parking results

[0204] The model based on the present invention is implemented by using the GA optimization algorithm, and the present invention is programmed using MATLAB r2021a software. The present invention sets the population size to 50, the number of elites retained to 5, the selection rate to 0.9, the crossover rate to 0.6, and the mutation rate to 0.3. The maximum number of iterations, i.e., the termination criterion, is set to 400. By running MATLAB, the approximate optimal solution of the example can be obtained, and the optimal solution of each generation is recorded. The relevant convergence diagram is shown as follows: Figure 15 As shown in the figure, the horizontal axis represents the number of iterations, and the vertical axis represents the minimum value of the waiting time of all ships in the port during each iteration.

[0205] By running the above example, the optimal berth allocation scheme is obtained, and the optimal objective function value of this scheme is 1556. As shown in Table 10, this scheme clarifies the best order for ships to enter the port for berthing services. In addition, the visualization results of the optimal berth allocation are shown in Figure 16 As shown in the figure, 16 ships are assigned to berths, and their berthing times do not overlap, effectively avoiding conflicts. For large ships such as Ship 2 and Ship 15, which need to enter and exit the port under specific tidal conditions, the scheduling plan ensures that they can enter and exit the port safely within the high-water time window. At the same time, their berthing positions also fully consider the draft depth restrictions of the ships, ensuring the safety and feasibility of berthing. This scheduling plan, which comprehensively considers various factors, not only improves the utilization efficiency of berth resources but also optimizes the overall operation process of the port.

[0206] Table 10 Optimal scheduling solution

[0207]

[0208] 2) Verification of solution effectiveness

[0209] This paper studies the problem of berth dynamic allocation optimization. To verify the effectiveness of the proposed scheme, this scheme is compared with the "first come, first served" (FCFS) strategy commonly used in ports. That is, berths are allocated in order according to the priority of the ship's arrival time. The specific content of the first come, first served berth allocation scheme is shown in Table 11, and the berth allocation diagram is shown in Figure 11. Figure 17 shown.

[0210] Table 11 FCFS scheduling scheme

[0211]

[0212]

[0213] By comparing Figure 16 and Figure 17 Tables 10 and 11 show that this scheme takes optimization objectives into consideration during scheduling. It does not follow the order of ship arrival completely, but performs intelligent optimization to generate a better allocation strategy, which makes the overall berth utilization rate higher. FCFS directly arranges according to the order of arrival, which may lead to uneven utilization of berth resources. The comparison results of the two berth allocation schemes are shown in Table 12. The sum of the waiting time of ships under the FCFS scheme is 2175 minutes, which is greater than the target value of 1556 minutes of the optimized scheme of the present invention, and the relative deviation is 28%. Therefore, the scheme of the present invention can effectively optimize the allocation of berth resources and avoid waste of resources. However, in the absence of a flexible scheduling mechanism, the FCFS scheme requires ships to wait passively, which increases the waste of ship waiting time.

[0214] Table 12 Results for 16 ships using GA and FCFS.

[0215]

[0216] 3) Verification of algorithm effectiveness

[0217] To verify the effectiveness of the genetic algorithm, this paper used Gurobi and the genetic algorithm to solve problems with varying numbers of arriving ships. Example C5-1 represents a problem with five ships, including one large ship. The experimental results were compared and analyzed, where OBJ represents the objective function value and CT represents the model solution time. In the experiment, Gurobi's runtime was capped at 7200 seconds. The experimental results, shown in Table 13, demonstrate the performance of the two methods for problems of varying scales.

[0218] Table 13 Algorithm effectiveness comparison results.

[0219]

[0220] 4) Sensitivity analysis

[0221] Sensitivity analysis is an uncertainty research method based on quantitative analysis. It is used to evaluate the sensitivity of model output results to changes in input parameters. By systematically adjusting input parameters and observing changes in output results, sensitivity analysis can evaluate the robustness of the model and identify key factors that have a significant impact on system performance.

[0222] This paper integrates the prediction process with allocation modeling to propose a sensitivity analysis method for changes in the estimated arrival time. The impact of ETA prediction errors of ±1h, ±3h, and ±5h on berth allocation planning is considered for different ship sizes. With a planning period of T = 48 hours and the number of berths remaining unchanged, the above optimization algorithm is used to calculate ship waiting times under different case scenarios and analyze the impact of ETA prediction errors on berth allocation results. The experimental results are shown in Figure 2. Figure 18 shown.

[0223] In summary, the present invention predicts the ship's travel time based on real-time dynamic predictions, and establishes a port intelligent berthing model in combination with continuous terminals under tidal constraints. Finally, it uses genetic algorithms and other related optimization algorithms to solve the problem and determine the dynamic berth allocation. It has the characteristics of wide coverage and strong applicability. The intelligent berthing of the port will affect the arrival time of the ship, the order of entering and leaving the port, and the berth assignment, and will also affect the rational formulation of the port production scheduling operation plan. Different from the traditional port berthing, the present invention has designed and developed a port intelligent berthing model that combines influencing factors such as tides and ship arrival time. It has low workload, good economy and strong versatility. The present invention only needs to make a real-time dynamic prediction of the ship's arrival time, and establish a port ship intelligent berthing model in combination with continuous terminals under tidal constraints. Then, the dynamic berth allocation can be calculated according to the relevant optimization algorithms, so that the maritime administration and port enterprises can realize the real-time dynamic macro-management of the ships in the port area.

Claims

1. A method for intelligent berthing of a ship in a port, characterized by: The following steps are involved: A. Collect information data Information data is collected through the port network. The collected information data includes AIS data and ship information data. The collected information data is divided into dynamic information, static information and voyage information. The dynamic information is real-time motion state parameters, including longitude and latitude, speed and heading; the static information is ship attribute parameters, including the Maritime Mobile Service Identity (MMSI), ship type, length and width; the voyage information is voyage-related parameters, including departure port, destination port and estimated arrival time; Preprocessing the collected information data; B. Real-time dynamic prediction of ship arrival time Based on the collected information data, the ship's arrival time is predicted using a virtual guided route network consisting of ship sinking points, navigation marks, maritime regulatory waterways and dynamic buoys; C. Establish a real-time dynamic allocation model for port berths based on ship arrival time prediction Considering the real-time dynamic allocation of port berths under tidal constraints for continuous terminals and dynamic ship arrivals, a real-time dynamic allocation model for port berths is established based on the ship arrival time prediction. This model can rationally arrange the entry and exit sequence, berthing time and location for arriving ships, and optimize the berth allocation plan. D. Solve the real-time dynamic allocation model of port berths Genetic algorithm (GA) is used to solve the real-time dynamic allocation model of port berths. E. Output real-time dynamic port ship berthing results Output real-time dynamic port ship berthing results to provide decision-making information for maritime departments and port management personnel.

2. The intelligent berthing method for a port ship according to claim 1, characterized in that: The method for predicting the real-time dynamic ship arrival time described in step B comprises the following steps: B1. Gradient Boosting Let F m (x) is the integrated model of gradient boosting decision tree, f t (x) is the sub-model of the t-th iteration, that is, the weak regression tree or the residual of the fit, then the integrated model of the gradient boosting decision tree is expressed as follows: Where: is the residual fitting coefficient, t is the residual fitting coefficient number, t=0,1,2,…,m; Assume that the loss function after the last residual fitting is L[F m (x), Y], each time a sub-model is added, the following formula is satisfied: L[F m (x),Y]<L[F m-1 (x),Y] Where: x is the feature vector, Y is the set of true values ​​of the sample scalar, F m-1 (x) is the ensemble model of the penultimate gradient boosting decision tree; B2. Predicted ship arrival time Assume that the dataset sample used in training is the Qth feature vector x Q and the true value of the Bth sample scalar y B , where x Q ∈R and y B ∈R, the lightweight gradient boosting machine regression prediction model is a strong regression tree through the linear combination of K weak regression trees, and the output is the predicted arrival time of the i-th ship Calculated as follows: in, Represents input x i The predicted ship arrival time is k, which represents the number of decision trees in the lightweight gradient boosting machine regression prediction model, and f k represents the kth basic classification and regression tree, namely CART tree, and F is the function set of all CART trees.

3. The intelligent berthing method for a port vessel according to claim 1, characterized in that: The steps for establishing the real-time dynamic allocation model of port berths described in step C are as follows: C1. Determine the operating conditions of the real-time dynamic allocation model of port berths C11. All ships must pass through the channel at a fixed time and are not allowed to dock or wait at will. Ships must immediately berth according to the berthing plan after leaving the channel. C12, the ship's head direction is pointing to the 0 scale direction of the shoreline; C13. Large ships need to use the tides to enter and exit the port, but small ships do not. C14. To prevent collisions during navigation, the preceding and following vessels shall maintain a safe distance when sailing. C15. Does not involve specific loading and unloading operations of ships; C16. Vessels shall not be repeatedly dispatched during the planning period; C17. No berthing operation shall be carried out after the vessel berths; C2. Establish a real-time dynamic allocation model for port berths The objective function of the real-time dynamic allocation model of port berths is expressed as follows: Constraints: 0≤x i ≤L-l i i∈I (14) x i -x j +l i ≤(1-y ij )Li∈I,j∈I,i≠j (16) x ij +x ji +y ij +y ji ≥1 i∈I,j∈I,i≠j (17) x ij +x ji ≤1 i∈I,j∈I,i≠j (18) and ij +y ji ≤1 i∈I,j∈I,i≠j (19) Where: I represents the set of |I| container ships arriving at the port; i represents the ship number, j represents the serial number of another ship; I in Indicates that ships need to enter the port during the high tide. out Indicates that ships need to leave the port during the high tide; L indicates the length of the coastline; n indicates the total number of times a set of high and low water levels occurs; V indicates the duration of the high or low water level; T represents the total planning time, T = 2Vn; a i Indicates the estimated arrival time of ship i; l i represents the length of ship i; h i represents the operation time of ship i; C represents the time required to enter or exit the channel; represents the starting position of ship i suitable for berthing; Indicates the end position of ship i suitable for berthing; M is a sufficiently large number; x i Indicates the starting position of the berthing operation of ship i; Indicates the moment when the waiting ship i enters the channel; Indicates the time when the operation vessel i enters the channel; Indicates the time when the berthing operation of ship i starts; Indicates the time when ship i ends its operation; represents the time when ship i leaves the port; in Indicates that the value is 1 when ship i enters the channel for berthing at the nth high tide, otherwise it is 0; μ in The value is 1 when ship i completes the operation and leaves the channel at the nth high tide, otherwise it is 0; i,b If ship i is docked at shoreline b, it is 1, otherwise it is 0; ij In the two-dimensional graph, the value is 1 when ship i is completely below ship j, otherwise it is 0; ij In the two-dimensional graph, the value is 1 when ship i is completely to the left of ship j, and 0 otherwise.

4. The intelligent berthing method for a port vessel according to claim 1, characterized in that: The steps of the genetic algorithm described in step D are as follows: D1. Chromosome encoding According to the characteristics of the berth allocation problem, the initial solution of the real-time dynamic allocation model of port berths is generated by integer coding. Each chromosome represents the order of ship berthing service. Each chromosome consists of XN genes, where I represents the number of ships. The genes on each chromosome of the initial population are randomly generated. D2. Generate initial solution A reasonable berth allocation scheme is constructed based on the chromosomes in the population. The berthing time and berth position of each ship in the port channel affected by tidal constraints and berthing water depth are determined, and then the fitness function is calculated. The specific steps are as follows: D21. Set the time when ship i starts entering the channel as the estimated arrival time, and determine whether the ship needs to enter the port during the high tide. If the conditions for entering the port are met, execute step D22; otherwise, wait for the start time of the next high tide to enter the port; D22. Set the berthing position of vessel i to the starting position that meets the berthing conditions. The berthing time is the sum of the time it enters the channel and the time it passes through the channel. Determine whether it overlaps with other vessels on the berthing time two-dimensional graph. If not, proceed to step D23. Otherwise, increment the berthing position by one unit until it reaches the end position of the berthing conditions. If it still overlaps with other vessels, increment the berthing position by one unit on the time axis and return to step D22. D23: After the ship operation is completed, it is determined whether ship i needs to leave the port at high tide. If the conditions for leaving the port are met, step D24 is executed; otherwise, the ship waits at the berth for the next high tide to leave the port. D24, let i = i + 1, and perform berth optimization for the next ship; D3. Calculate the fitness function The inverse of the objective function is taken as the fitness function, that is, a lower fitness value represents a better berth allocation plan; This ensures that individuals with shorter waiting times correspond to higher fitness values ​​and are more likely to be retained and propagated in the genetic process, thereby guiding the algorithm to converge to the optimal solution; D4. Perform selection operation First, using the elite retention strategy, the fitness values ​​of all chromosomes in the current population are calculated and sorted in ascending order. Chromosomes with higher fitness are retained according to the set elite ratio and directly copied to the next generation to ensure that the best individuals in the next generation are better than those in the previous generation. Subsequently, the roulette wheel selection method is used for the remaining individuals, and the individuals with greater fitness have a higher probability of being selected. The selected individuals will undergo crossover and mutation operations, and elite individuals will also be inserted into these operations to ensure that the population size is always the preset number. D5. Perform cross operation The two-point crossover operator is used to perform a crossover operation on chromosomes. Two parent chromosomes are selected, two crossover points are randomly generated, and the chromosome segments between the two crossover points are exchanged while retaining and mapping the remaining genes to ensure the validity of the chromosomes, thereby generating two new daughter chromosomes. D6. Perform mutation operations During the berth allocation process, the docking layout of the front ship will affect the berth allocation of the rear ship; a better berth allocation plan means that the layout formed by the front ship is more reasonable; therefore, the chromosomes of elite retention and crossover operations are mutated in turn. The mutation operation will retain a certain proportion of genes in the parent chromosome, and the remaining genes are randomly generated to form a new chromosome.

Images