A port berth optimization scheduling method based on big data

Abstract

The application discloses a port berth optimization scheduling method based on big data, and particularly relates to the technical field of intelligent port scheduling; target port historical ship berthing data is collected, a structured big data sample set is constructed, a berthing duration prediction model is established, and ship predicted berthing time is output; in combination with current and future available time windows of each berth, a berth resource feasible matching matrix is constructed, a minimum operation switching time constraint, a tide influence model and a berth operation preference rule set are introduced, the matrix is subjected to multi-dimensional weighted adjustment, and a weighted matching matrix is obtained; subsequently, a heuristic optimization algorithm is used to solve the weighted matching matrix, and an optimal allocation sequence of ships to berths is obtained; the application can improve berth resource utilization, reduce ship waiting time, and improve port operation efficiency.

Description

Technical Field

[0001] This invention relates to the field of port intelligent scheduling technology, specifically to a port berth optimization scheduling method based on big data. Background Technology

[0002] With the rapid development of global maritime trade, ports, as crucial nodes for cargo distribution, play a key role in the smooth operation of the entire logistics chain through efficient berth resource management and vessel scheduling. Traditional port berth scheduling typically relies on manual experience or static rules for decision-making, making it difficult to adapt to the high-frequency, large-volume, and constantly changing demands of vessel arrivals. Especially against the backdrop of rapid growth in container throughput, the problems of irrational berth resource allocation and low scheduling efficiency are becoming increasingly prominent.

[0003] More seriously, in actual operation, some ports are subject to complex and nonlinear scheduling relationships between berths due to geographical environmental factors such as limited harbor basin shape, unstable tidal flow velocity, and uneven berth water depth, as well as dynamic factors such as diverse ship sizes, large fluctuations in loading and unloading operation time, and channel congestion. This makes it difficult for traditional methods to accurately predict the future availability of each berth, resulting in phenomena such as ship queuing, berthing delays, and even wrong berthing. This not only affects the port's throughput efficiency but also causes shipping companies to incur high port demurrage costs.

[0004] Currently, there is a lack of a comprehensive scheduling method that can integrate historical operational data, real-time dynamic information, and predictive models to achieve global optimization and dynamic allocation of port berth resources. Especially in scenarios with complex port basin layouts and parallel operations of multiple vessel types, how to leverage big data technology to achieve real-time optimal berth scheduling decisions has become a core challenge for intelligent port operations. Summary of the Invention

[0005] The purpose of this invention is to provide a port berth optimization scheduling method based on big data to address the shortcomings of the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a port berth optimization scheduling method based on big data, comprising:

[0007] Collect historical ship berthing data for the target port and form a structured big data sample set;

[0008] A berthing duration prediction model is established for each berth based on the sample set. The model is based on time series regression combined with berth resource state variables, and the predicted berthing time is output.

[0009] Based on the ship's current estimated arrival time and predicted berthing time, and combined with the current and future available time windows of the berth, a berth resource feasibility matching matrix is ​​constructed. Each element in the matrix represents the schedulability score of ship i berthing at berth j.

[0010] Based on the minimum operation switching time between berths, the tidal impact model, and the berth operation preference rule set, the schedulability score in the matrix is ​​weighted and adjusted to generate a weighted matching matrix.

[0011] The weighted matching matrix is ​​solved using a heuristic optimization algorithm to obtain the optimal solution for berth scheduling, which is the optimal allocation sequence of ships to berths, and satisfies the objective functions of minimum queuing time, minimum tidal conflict and minimum resource waste.

[0012] The optimal berth scheduling solution is output to the port scheduling control system and linked with real-time AIS data to dynamically update the ship arrival sequence and berth allocation, and continuously optimize the scheduling.

[0013] Preferably, the historical vessel berthing data includes, but is not limited to: vessel type information, estimated arrival time, actual berthing time, berth number, operation time, departure time, tidal data and meteorological data during berthing.

[0014] Preferably, the step of establishing a berthing duration prediction model for each berth based on a sample set includes:

[0015] The historical berthing data belonging to the same berth number in the sample set are clustered and initially grouped according to ship type, tide class and cargo type to obtain a subset of ships with similar operational characteristics;

[0016] For each subset of ships, a time series data window is constructed. With the expected arrival time as the time axis, S consecutive berthing samples before and after the expected arrival time are extracted to form a sliding window sequence, where S is the length of the sliding window.

[0017] A regression algorithm based on long short-term memory network is used to model the time series. The input features include: ship type code, historical berthing time, average berth operating efficiency, tide time series and wind speed time series.

[0018] The trained berthing duration prediction model is used to output the predicted value of the ship's berthing duration.

[0019] Preferably, the construction of the berth resource feasible matching matrix includes:

[0020] Extract the estimated arrival time and predicted berthing time for each target vessel;

[0021] Obtain the current availability status information of each berth, including current occupancy status, estimated release time, and subsequent reservation schedule;

[0022] Based on the ship’s estimated arrival time and its predicted berthing time, a sliding time window is used on all berth resources to determine whether the berth is idle during the continuous predicted berthing time period starting from the estimated arrival time.

[0023] If the berth is determined to be in an idle state, the corresponding position in the berth resource feasible matching matrix will be assigned a value of 1, indicating that ship i can be scheduled to berth j; otherwise, the value will be assigned a value of 0.

[0024] Preferably, the step of weighting and adjusting the schedulability scores in the matrix to generate a weighted matching matrix includes:

[0025] Obtain the minimum berth operation switchover time ΔT_cut, which represents the minimum safe interval time reserved between two consecutive vessels at the same berth;

[0026] For each schedulable match, if the difference between the estimated berthing time of vessel i and the last operation end time of berth j is less than ΔT_cut, then reduce its matching score.

[0027] The scores are adjusted using a penalty function, and the adjusted scores replace the original matrix values ​​to construct a switching time correction matrix. After correction and fusion processing, a weighted matching matrix is ​​obtained.

[0028] Preferably, the step of solving the weighted matching matrix using a heuristic optimization algorithm includes:

[0029] Using a weighted matching matrix as input, the set of ships and the set of berths are modeled as a bilateral matching problem, and the objective function is set as minimizing ship queuing time, minimizing the number of tidal conflicts, and minimizing berth idling rate.

[0030] A genetic algorithm is used to initialize multiple scheduling candidate solutions, each of which represents a sequence of ship-to-berth allocations.

[0031] The fitness of candidate solutions is calculated, and the fitness function uses the weighted combination of the weighted score matrix value and the objective function as the evaluation index.

[0032] The candidate solutions are iteratively updated through selection, crossover, and mutation operations until the preset number of iterations or fitness convergence conditions are reached, and the optimal berth scheduling solution is output.

[0033] Preferably, the method of linking with real-time AIS data includes:

[0034] Real-time acquisition of location, speed, and estimated arrival time of the target vessel from its AIS system;

[0035] Clean and predictive correct the AIS data to generate the updated arrival time T_real;

[0036] Compare T_real with the original estimated arrival time. If the deviation exceeds the preset threshold of 30 minutes, the scheduling update process is triggered.

[0037] During the update process, the original optimal solution is used as the initial solution, and the heuristic optimization algorithm is run again to quickly generate a new berth scheduling solution.

[0038] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0039] 1. This invention constructs a port berth optimization scheduling method based on big data. By combining historical berthing data with time series prediction models, tidal impact modeling, and multi-factor weighted matching matrices, it significantly improves the accuracy of ship berthing time prediction and the rationality of berth resource allocation. Compared with traditional scheduling methods relying on manual experience or static rules, this invention can dynamically perceive the port operating environment and achieve comprehensive optimization of ship queuing time, tidal conflicts, and berth resource waste.

[0040] 2. This invention constructs a continuously iterative closed-loop optimization system by introducing a heuristic optimization algorithm and a real-time AIS data linkage mechanism. This system can not only modify the scheduling plan at the minute level to ensure scheduling feasibility under emergencies, but also maintain the stability and flexibility of the scheduling solution during continuous operation. Therefore, this invention achieves intelligent, real-time, and efficient port berth scheduling, and has good prospects for engineering applications. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0042] Figure 1 This is a mind map of the method of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] For examples, please refer to Figure 1 As shown in this embodiment, a port berth optimization scheduling method based on big data includes:

[0045] Collect historical ship berthing data for the target port and form a structured big data sample set;

[0046] A berthing duration prediction model is established for each berth based on the sample set. The model is based on time series regression combined with berth resource state variables, and the predicted berthing time is output.

[0047] Based on the current estimated arrival time and predicted berthing time of the vessel, and combined with the current and future available time windows of the berths, a berth resource feasibility matching matrix is ​​constructed. Each element B in the matrix... ij The score represents the schedulability score of vessel i berthing at berth j;

[0048] Based on the minimum operation switching time between berths, the tidal impact model, and the berth operation preference rule set, the schedulability score in the matrix is ​​weighted and adjusted to generate a weighted matching matrix.

[0049] The weighted matching matrix is ​​solved using a heuristic optimization algorithm to obtain the optimal solution for berth scheduling, which is the optimal allocation sequence of ships to berths, and satisfies the objective functions of minimum queuing time, minimum tidal conflict and minimum resource waste.

[0050] The optimal berth scheduling solution is output to the port scheduling control system and linked with real-time AIS data to dynamically update the ship arrival sequence and berth allocation, and continuously optimize the scheduling.

[0051] To optimize the scheduling of port berth resources, it is first necessary to construct a high-quality structured big data sample set to support subsequent model training and decision-making processes.

[0052] The data collection scope covers the berthing information of all vessels arriving at the target port within the past 3 years (or a specified time period), specifically including the following fields:

[0053] Vessel information: including vessel name, IMO number, gross tonnage, maximum draft, length, and cargo type (container, bulk, roll-on / roll-off, etc.).

[0054] Estimated arrival time: the time when the vessel is scheduled to arrive at the port, recorded by the AIS system;

[0055] Actual berthing time: This refers to the actual start time when the vessel berths at the dock, as recorded by the port scheduling system.

[0056] Berth number: The specific berth number corresponding to the actual berth where the vessel is berthed;

[0057] Operation time: the total operation time from berthing to completion of unloading / loading;

[0058] Departure time: The actual time when a vessel leaves its berth and departs from the port area;

[0059] Tidal data: Tidal level change curves during berthing, including the highest and lowest tide values, with a recording time accuracy of 30 minutes;

[0060] Meteorological data: including meteorological indicators such as wind speed, wind direction, rainfall intensity, and visibility; sampling time accuracy is the same as above.

[0061] Port congestion index: The system calculates this index based on the average number of ships queuing per hour, serving as an auxiliary feature for scheduling pressure.

[0062] Berth operation efficiency parameters: These are calculated based on the average loading and unloading volume per unit time at the berth and are used to assess the intensity of berth resource utilization.

[0063] The specific steps of data structuring are as follows:

[0064] Delete records with missing values, time logic conflicts (such as departure time being earlier than berthing time), or abnormal identification codes (such as duplicate IMO numbers);

[0065] Convert all time fields to a unified timestamp format (Unix time) for easier subsequent processing;

[0066] All vessel records are normalized. For example, wind speed is normalized to a decimal between 0 and 1, and the ship length is normalized using the maximum ship length.

[0067] For each record, mark the berth usage status, operation success rate, and whether there is a delay (based on whether the difference between the estimated arrival time and the actual berthing time exceeds the threshold).

[0068] Using berthing time as the central window, sliding matching and splicing of tidal and meteorological data are performed to form a time-series feature matrix.

[0069] All discrete variables such as ship type and berth number are processed using one-hot encoding to generate samples in a uniform format.

[0070] Ultimately, the constructed structured sample set D is a multi-dimensional array structure, where each sample data represents the complete berthing process of a ship and its environmental variables, with a total of no less than 25 dimensions of features. This sample set will serve as the input basis for subsequent prediction model training (such as the berthing duration prediction model) and scheduling optimization model.

[0071] Static threshold definition: To determine whether a delay has occurred, a time difference threshold of 30 minutes is set. That is, if the "actual berthing time minus the estimated arrival time" is greater than 30 minutes, the record is marked as "delayed". The tide level anomaly threshold is set to 80% of the port's historical average lowest tide level. When it is below this threshold, it is marked as "low tide risk". The wind speed threshold is set to 12 meters per second. When it is exceeded, it is marked as "strong wind risk", which facilitates the extraction of key features by the subsequent risk perception model.

[0072] Extract a set of historical berthing records associated with a specific berth number from the constructed structured sample set, denoted as D. k Where k represents the berth number index. Since different ship types, tidal conditions, and types of cargo significantly affect berthing operation time, to enhance the model's predictive accuracy and generalization ability, D needs to be adjusted. k Preliminary clustering and grouping processing is performed.

[0073] The specific steps are as follows:

[0074] Extracting feature vectors: for D k Three key clustering features were extracted from each ship record:

[0075] Ship type (e.g., container ship, roll-on / roll-off ship, bulk carrier, etc.);

[0076] Tidal rating (divided into high, medium, and low levels based on the tide level at the time of berthing);

[0077] Operational cargo types (such as loading, unloading, two-way operations, etc.).

[0078] Clustering method selection: Density-based clustering algorithms (such as DBSCAN) are used to perform unsupervised clustering on the above features, resulting in several subsets of ships with similar operational characteristics, denoted as C1, C2, ..., C6. n In this context, n represents the total number of similar job features.

[0079] Each ship subset C i This represents a sample of historical operations under similar tidal conditions and operating methods, providing a data foundation for subsequent time series modeling.

[0080] After completing the initial grouping, for each subset C i The samples within the dataset are sorted in ascending order according to the "estimated arrival time" field to establish a time series structure. Subsequently, with each target sample as the center, S historical records are extracted backward to construct a sliding window sequence of length S, forming the time series input dataset.

[0081] The sliding window length S is a positive integer, with a recommended default value between 5 and 10, dynamically set according to the vessel traffic density of the target port. For example, if the average daily vessel arrivals at the port are large, S can be set to 10 to enhance the model's time-series memory capability.

[0082] For each sliding window sequence, the following time series features are extracted:

[0083] Ship type coding: The ship type category is processed using a unique hot coding method;

[0084] Historical berthing time: The time required for the previous S historical berthings (in minutes).

[0085] Berth average operating efficiency: average loading and unloading volume per unit time (tons / hour).

[0086] Tide level time series: Tide level values ​​sampled every 30 minutes within a window period;

[0087] Wind speed time series: the average wind speed value every 30 minutes within the corresponding time period.

[0088] These features together constitute the input tensor, which is used by the LSTM model for time series modeling.

[0089] LSTM (Long Short-Term Memory) is an improved recurrent neural network suitable for processing data sequences with long-term dependencies. In this invention, a single-layer LSTM network is used as the backbone structure for regression modeling. Its input is multi-dimensional time series features within a sliding window, and its output is the predicted berthing time of the current target vessel.

[0090] The specific network parameter settings are as follows:

[0091] Number of hidden units in LSTM: 64;

[0092] Time step: equal to the sliding window length S (e.g., S=5);

[0093] Activation function: Hyperbolic tangent function (tanh);

[0094] Output layer: A single fully connected neuron that outputs the predicted value;

[0095] Optimizer: Adam optimizer, with an initial learning rate of 0.001;

[0096] Loss function: Mean squared error (MSE), which is the squared mean of the difference between the actual berthing time and the predicted value.

[0097] Weights are updated using backpropagation and gradient descent.

[0098] To ensure the model's stability and generalization ability, the training process for each berth model is set up as follows:

[0099] All input features are normalized to 0 to 1 to avoid bias in the model convergence process caused by features of different dimensions;

[0100] Time sorting and sliding window batch generation ensure that the input data is strictly sorted in ascending order of time to generate a window sequence;

[0101] Five-fold cross-validation was used to train and test the LSTM model for each berth to reduce the risk of overfitting.

[0102] If the average prediction error of all folds in the cross-validation results is less than 30 minutes, the model is considered to have converged well and can proceed to the production deployment stage.

[0103] Another set of berthing samples that were not used in training was selected as the validation set for the final prediction accuracy test.

[0104] If the prediction mean error is controlled within ±30 minutes and the maximum error does not exceed 60 minutes, the model will be deployed to the scheduling system.

[0105] Considering the highly dynamic nature of port operations, such as adverse weather conditions, sudden congestion, and equipment malfunctions that could affect the timeliness of model predictions, this invention introduces an adaptive incremental update mechanism:

[0106] The real-time monitoring system compares the predicted values ​​with the actual berthing time.

[0107] If the deviation exceeds 60 minutes and occurs more than 3 times consecutively, the "model retraining" mechanism will be triggered.

[0108] New samples are automatically incorporated into the sample set, and the model is retrained in an offline environment to achieve dynamic evolution.

[0109] Within the scheduling cycle, the system extracts the estimated arrival time T for each vessel i waiting to berth. 1i and predicted berthing duration .in:

[0110] Estimated arrival time T 1i Provided by the ship's AIS system or dispatch system, in minutes;

[0111] Predicted berthing time The output of the LSTM regression model described in this invention represents the operation time required from the start of berthing to departure from the berth, also in minutes.

[0112] This timeframe provides a window boundary for subsequent determination of berth vacancy intervals.

[0113] For all berth resources j (j = 1, 2, ..., m) within the port, the scheduling system synchronizes their status data in real time, specifically including the following:

[0114] Current Occupancy Status: If berth j is currently occupied by a vessel, the system records its estimated operation completion time T. 2j ;

[0115] Subsequent Reservation Schedule: The system records the work reservation time periods for berths within the next 24 hours, including the reservation start time T. 3jk End time T 4jk (k is the schedule number);

[0116] The information is provided uniformly by the port's Terminal Operating System (TOS) or resource scheduling platform, with an accuracy down to the minute level.

[0117] For each target vessel i, the system uses its estimated arrival time T 1i Construct a continuous time interval starting from the window. A sliding judgment is performed on the time axis of each berth j to determine whether the time period is occupied or conflicts with a reserved task.

[0118] The judgment logic is as follows:

[0119] If the current time <T 2j And T 2j >T 1i This indicates that the berth has not yet been released, so T 1i The berth is not available if it is not within the available area.

[0120] Iterate through all reservation tasks k for berth j, if the window interval of ship i is... With any reservation time period [T 3jk T 4jk If there is an intersection, that is, an overlapping part, the berth is unavailable;

[0121] If neither of the above two conditions is met, berth j is determined to be vacant during this time period and can be used by vessel i for scheduling.

[0122] Based on the above judgment results, construct an m-column × n-row feasible matching matrix B to represent the scheduling feasibility between n ships and m berths:

[0123] If the i-th ship can be at the j-th berth If the entire task is completed within the interval without any conflicts, then assign the value B. ij =1;

[0124] If there is a conflict of occupancy or reservation, then assign the value B. ij =0.

[0125] Matrix B is a binary matrix, and each element B ij All of these have a clear physical meaning, namely, "whether ship i can be dispatched to berth j".

[0126] After constructing the initial berth resource feasible matching matrix B, this invention further introduces a multi-dimensional weighted adjustment mechanism to more precisely reflect the actual operational constraints and priority scheduling rules of the port. The initial matrix B only indicates whether a ship can be physically and temporally matched with berth resources, with a value of 0 or 1, and is a binary matrix.

[0127] Specifically, the weighted adjustment includes three levels: constraints on the minimum operation switchover time between berths, correction of the tidal impact model, and embedding of berth operation preference rules. After multiple levels of adjustment, the final generated matrix is ​​denoted as B_final, where each element is a weighted value between 0 and 1. The larger the value, the higher the feasibility and priority of matching the corresponding vessel with the berth.

[0128] In actual port operations, once a vessel has completed its work and departed, the berth cannot be immediately used for the next vessel. A series of preparatory tasks are typically required, such as resetting unloading equipment, relocating personnel, and clearing the channel. This interval is called the minimum operation switchover time, denoted as ΔT_cut.

[0129] Threshold definition: The value of ΔT_cut is obtained based on historical port data statistics. For example, the average value for a container terminal is approximately 30 minutes. If the statistical standard deviation is large, it can be dynamically adjusted according to the ship type and size. The switching time for the largest ship type can be set to 60 minutes, while for small and medium-sized ships it can be maintained at 30 minutes.

[0130] Judgment method:

[0131] For each schedulable match B ij Extract the estimated arrival time T of vessel i. 1i and predicted berthing duration Simultaneously extract the last operation end time T of berth j. 2j .

[0132] If T 1i Less than T 2j +ΔT_cut indicates that if the minimum switching time requirement is not met when the ship arrives at the port, its matching score will be reduced.

[0133] If T is satisfied 1i Greater than or equal to T 2j If +ΔT_cut is applied, the switching condition is met, and a high score is maintained.

[0134] Weighting method: Introduce a smoothing penalty function to adjust the original value 1 to: Where ΔT_actual is the actual time interval, and α is the adjustment coefficient, typically ranging from 0.05 to 0.1. When ΔT_actual is much smaller than ΔT_cut, W1 approaches 0; when ΔT_actual is greater than or equal to ΔT_cut, W1 approaches 1. The switching time correction matrix B1 is generated in this way.

[0135] Berth availability is constrained not only by time and equipment but also by tidal conditions. In particular, vessels with deep drafts cannot berth safely during low tide, thus requiring the construction of a tidal impact model M2 to correct the matrix score.

[0136] Model input and output:

[0137] Input: Planned berthing time range for the vessel Tide level sequence within; ship draft d i The minimum safe water depth h of berth j j .

[0138] Output: Tide Adaptation Factor T ij The value ranges from 0 to 1, with a larger value indicating more suitable tidal conditions.

[0139] Margin calculation method: Within the time interval, calculate the minimum tide level h_min, and define the safety margin as: M = h_min - d i If M is greater than or equal to 0, it means that the draft requirement is met; if M is less than 0, there is a risk of stranding.

[0140] Tidal factor definition: When M is greater than or equal to 0.5 meters, T ij Set to 1;

[0141] When M is between 0 and 0.5 meters, T ij Decreasing according to a linear function, for example, T ij =M÷0.5;

[0142] When M is less than 0, T ij Set it to 0.

[0143] Matrix correction method: Multiply each element in B1 by T ij We obtain the tidal correction matrix B2, i.e.: B 2ij =B 1ij ×T ij This method can effectively eliminate matching schemes with a high risk of low tide.

[0144] Different berths have different operational preferences; for example, some berths are more suitable for container operations, while others are more suitable for bulk or liquid cargo. To incorporate long-term port operation experience into the model, this invention introduces a set of operational preference rules R.

[0145] The rule set R includes the following parameters:

[0146] Ship type compatibility: Based on historical statistics, the operational efficiency of berth j for ship type i is scored;

[0147] Cargo type preference: Historical average efficiency of berths for different cargo types such as containers, bulk cargo, and roll-on / roll-off (Ro-Ro);

[0148] Quay crane resource matching degree: Whether the number and capacity of quay cranes corresponding to the location of the berth meet the needs of ships.

[0149] Preference factor definition: For each ship and berth combination, a preference factor P is defined. ij The value ranges from 0 to 1. If the preference condition is fully met, then P... ij =1; If there is a partial match, then assign a value based on the historical efficiency ratio. For example, if the efficiency is only 70% of the standard, then P ij = 0.7; if there is no perfect match, then P ij = 0.

[0150] Matrix fusion method: Multiply B2 by the preference factor to obtain the final weighted matrix: B_final ij =B 2ij ×P.

[0151] To ensure the comparability of scores across different vessels, this invention normalizes B_final so that all its elements fall within the range of 0 to 1. The normalization method uses max-min scaling, i.e., B_norm. ij =(B_final ij The formula is: (﹣min)÷(max﹣min), where min and max are the minimum and maximum values ​​of all elements in the matrix, respectively. The resulting normalized matrix B_norm serves as the input to subsequent scheduling optimization algorithms (such as integer programming, heuristic search, or reinforcement learning). The algorithm will prioritize berth allocation schemes with high scores, low risks, and that conform to preferences, thereby improving the overall throughput capacity of the port while satisfying both safety and efficiency.

[0152] After obtaining the weighted matching matrix B_final, this invention employs a heuristic optimization algorithm to solve for it, thereby obtaining the optimal allocation sequence of ships to berths. Compared with traditional exhaustive or linear programming methods, the heuristic method can converge quickly in complex, nonlinear environments with many constraints, generating near-optimal or optimal solutions. By rationally designing the fitness function, search operator, and dynamic update mechanism, the scheduling optimization method of this invention not only meets multi-objective requirements but also possesses strong real-time performance and scalability.

[0153] The input is a weighted matching matrix B_final, with dimensions n rows and m columns, where n is the number of ships waiting to berth and m is the number of port berths. Matrix elements B_final ij The comprehensive matching score for vessel i berthing at berth j ranges from 0 to 1.

[0154] Define decision variable X ij When vessel i is assigned to berth j, X ijThe value is 1 if it is true and 0 otherwise. Each row can have only one element with a value of 1, ensuring that each ship is assigned only one berth.

[0155] The objective function is a weighted combination of three objectives:

[0156] Minimize ship queuing time;

[0157] Minimize the number of tidal conflicts;

[0158] Minimize berth vacancy rate.

[0159] The three objectives are weighted by weight coefficients W1, W2, and W3, where W1+W2+W3=1. The weights are set according to the actual needs of the port operator. For example, to prioritize reducing queuing, W1=0.5, W2=0.3, and W3=0.2.

[0160] This invention employs an improved genetic algorithm as the core heuristic for solving the problem, combined with a local search mechanism to improve the quality of the solution and the convergence speed. The process includes the following steps:

[0161] Several ship-berth allocation sequences are randomly generated, and each sequence is a candidate solution. To improve the initialization quality, a greedy strategy can be introduced on top of the randomness, such as prioritizing the allocation of berths to ships with the highest matching scores.

[0162] Fitness function design

[0163] For each candidate solution, calculate the fitness score F. The fitness function is defined as: F = W1 × (1 ÷ C1) + W2 × (1 ÷ C2) + W3 × (1 ÷ C3); where: C1 is the cost of ship queuing time; C2 is the cost of tidal conflict; and C3 is the cost of wasted berth resources.

[0164] Genetic algorithms continuously generate new generations of candidate solutions through selection, crossover, and mutation operators. The selection operation prioritizes solutions with higher fitness, the crossover operation swaps berth allocation fragments from different solutions, and the mutation operation randomly adjusts the berth allocation of a particular ship to maintain the diversity of the solution space.

[0165] After each iteration, local perturbations are performed on the solutions with the highest fitness. For example, two ships in the same time period are swapped, some berthing order is adjusted, or waiting time is inserted. If the fitness is improved after the perturbation, it is retained; otherwise, it is rolled back.

[0166] The iteration terminates and the optimal solution is output when the preset number of iterations N is reached or the fitness improvement of K consecutive generations is less than the threshold ε.

[0167] Queuing time cost C1: For each vessel i, queuing time is defined as the difference between its actual start berthing time and its estimated arrival time. C1 is the sum of the queuing times for all ships.

[0168] Tidal conflict cost C2: If vessel i's lowest tide level during its berthing period is lower than the sum of its draft and the berth's safe depth, it is counted as one conflict. C2 is the weighted sum of all conflict counts, with the weights determined by the berth's risk level.

[0169] Resource waste cost C3: Calculates the proportion of idle time of berths within the scheduling cycle. If the continuous idle time of a berth exceeds the threshold ΔT_idle (e.g., 60 minutes), it is included in the waste cost. C3 is the sum of the waste proportions of all berths.

[0170] All cost items are normalized after calculation to ensure that the values ​​are between 0 and 1, which facilitates combined calculations.

[0171] For example, the specific implementation of the algorithm includes:

[0172] Population size: 50 to 100;

[0173] Maximum number of iterations: 500;

[0174] Crossover probability: 0.8;

[0175] Mutation probability: 0.2;

[0176] Elite retention rate: The top 10% of solutions directly advance to the next generation.

[0177] After each iteration, the top 5% of candidate solutions are selected for local perturbation. Local perturbation includes:

[0178] Vessel swapping: exchanging berths between two vessels;

[0179] Order adjustment: delaying or advancing the berthing of a certain vessel;

[0180] Time insertion: Add waiting time for vessels at high tidal risk.

[0181] Real-time monitoring of the actual arrival time T_real of the vessel, when it differs from the estimated arrival time T_real... 1i If the deviation exceeds 30 minutes, a recalculation is triggered. During the recalculation, the genetic algorithm is run again with the current optimal solution as the initial solution, shortening the convergence time of the re-optimization.

[0182] After the algorithm finishes running, it outputs the optimal berth scheduling solution L, where L is a sequence mapping in the form {ship i → berth j}. This sequence guarantees that the following constraints are satisfied:

[0183] Each vessel is assigned to only one berth;

[0184] There are no overlapping or conflicting time intervals for each berth;

[0185] Queuing time, tidal conflicts, and resource waste are all minimized.

[0186] The final scheduling solution will be written into the port scheduling system and issued to the operations department as a real-time execution instruction.

[0187] After solving the weighted matching matrix and obtaining the optimal berth scheduling solution L using a heuristic optimization algorithm, this invention further outputs the scheduling solution to the port scheduling control system and dynamically updates it by linking with real-time AIS data, thereby ensuring continuous optimization of the scheduling plan and high adaptability to the actual environment. This process consists of three core steps: scheduling solution output and verification, AIS data linkage update, and dynamic optimization of vessel arrival sequence and berth allocation.

[0188] The optimal berth scheduling solution L is a sequence mapping in the form of {ship i → berth j, berthing start time t_start, berthing end time t_end}. The system converts this into a structured scheduling instruction, including the following fields:

[0189] Unique identification of a vessel (such as an IMO number);

[0190] Berth number;

[0191] Berthing start time;

[0192] End time of berthing;

[0193] Auxiliary resource parameters such as the number of quay cranes and tugboats required.

[0194] The scheduling instructions are written to the port's TOS (Terminal Operating System) via API interface or message queue. The TOS then distributes these instructions to the quay crane control subsystem, tugboat scheduling subsystem, and operations personnel management subsystem to form a complete execution plan.

[0195] Before outputting, the system performs a consistency check on the scheduling solution:

[0196] Verify berth status: If the berth specified in the scheduling instruction is still occupied, mark it as conflicted;

[0197] Time conflict check: If the newly assigned ship berthing time overlaps with the confirmed task time, it is marked as abnormal;

[0198] Verify auxiliary resources: If the required quay crane or tugboat resources are insufficient, a warning message will be generated.

[0199] If a verification anomaly is found, the system triggers a partial recalculation process, which only reallocates the affected vessels to ensure the executability of the output results.

[0200] The target vessel's AIS system continuously broadcasts its position, speed, and estimated time of arrival (ETA). The system collects data at a fixed interval ΔT_AIS (e.g., every minute) and stores it in the scheduling data warehouse.

[0201] The collected AIS data may contain noise or outliers and needs to be processed as follows:

[0202] Filter out invalid data where the velocity is zero but the position remains unchanged;

[0203] Interpolation correction is applied to points where the trajectory jumps abnormally;

[0204] The ETA field is smoothed, and the weighted average of the most recent 5 ETAs is taken as the correction value T_real.

[0205] The corrected T_real is compared with the original estimated arrival time T1 in the system. If the absolute difference is greater than the 30-minute threshold, the arrival time of the vessel is considered to have changed significantly, and the system immediately triggers the scheduling update process. The 30-minute threshold is an empirical value and can be flexibly adjusted according to the port's operational rhythm.

[0206] When an update is triggered, the heuristic optimization algorithm is rerun with the current optimal scheduling solution L as the initial solution. Since the initial solution is close to optimal, the optimization process can converge within a few iterations, thus completing the recalculation in minutes.

[0207] When the T_real values ​​of multiple vessels change simultaneously, the system sorts all vessels awaiting berthing based on the latest AIS data, generating a new dynamic arrival sequence Q_new. The sorting rule is to arrange them in ascending order of their estimated arrival time; if the times are the same, vessels with confirmed ETA reliability are prioritized.

[0208] The dynamic arrival sequence Q_new is compared with the original sequence Q_old to identify the set S_affect where the order has changed. Optimization is performed only on the S_affect set and its associated berth allocations to avoid global recalculation and reduce computational load.

[0209] In local optimization, a lightweight algorithm based on neighborhood search is used:

[0210] Neighborhood is defined as the ability to swap the berths of two ships in the S_affect set, or to adjust their berthing order.

[0211] Calculate the new fitness function value; if it is better than the original solution, retain it.

[0212] Repeat the neighborhood search until no improvement is made or the iteration limit is reached.

[0213] The local optimization results are combined with the allocation schemes of unaffected ships to form a new optimal solution L_new, ensuring the continuity of the scheduling solution in a dynamic environment.

[0214] The system sets a scheduling optimization cycle threshold ΔT_opt, for example, 10 minutes. Every time ΔT_opt is reached, the system automatically runs an optimization algorithm to reallocate berths to ships that have not yet begun berthing, correcting for potential efficiency losses.

[0215] In the event of an emergency (such as a vessel delay exceeding 2 hours, berth equipment malfunction, or extreme weather), the system immediately triggers an emergency optimization process. Unlike periodic optimization, emergency optimization recalculates the allocation of all vessels that have not yet begun berthing, ensuring that the scheduling plan remains executable even in unforeseen circumstances.

[0216] The new scheduling solution L_new generated by periodic and emergency optimization will immediately replace the original plan and distribute execution instructions through the port scheduling control system. To ensure the continuity of operations for the executing departments, the replacement process follows these principles:

[0217] The allocation of vessels that have already begun operations will not be adjusted;

[0218] If a vessel that has not yet started operations is adjusted due to optimization, a new plan will be sent to the execution end, along with an explanation of the reason for the change.

[0219] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A port berth optimization scheduling method based on big data, characterized in that: include: Collect historical ship berthing data for the target port and form a structured big data sample set; A berthing duration prediction model is established for each berth based on the sample set. Time series regression combined with berth resource state variables is used for modeling, and the predicted berthing time is output, including: The historical berthing data belonging to the same berth number in the sample set are clustered and initially grouped according to ship type, tide class and cargo type to obtain a subset of ships with similar operational characteristics; For each subset of ships, a time series data window is constructed, with the expected arrival time as the time axis, and S consecutive berthing samples before and after are extracted to form a sliding window sequence; A regression algorithm based on long short-term memory network is used to model the time series. The input features include: ship type code, historical berthing time, average berth operating efficiency, tide time series and wind speed time series. The trained berthing duration prediction model is used to output the predicted value of the ship's berthing duration. Based on the current estimated arrival time and predicted berthing time of the vessel, and combined with the current and future available time windows of the berths, a berth resource feasibility matching matrix is ​​constructed. Each element B in the matrix... ij The score represents the schedulability score of vessel i berthing at berth j; The construction of the berth resource feasibility matching matrix includes: extracting the estimated arrival time and predicted berthing time of each target vessel; obtaining the current availability status information of each berth, including the current occupancy status, estimated release time, and subsequent reservation schedule; based on the estimated arrival time and predicted berthing time of the vessel, sliding a time window across all berth resources to determine whether the berth is in an idle state during the continuous predicted berthing time period starting from the estimated arrival time; if it is determined to be in an idle state, then adding position B in the berth resource feasibility matching matrix. ij A value of 1 indicates that vessel i can be dispatched to berth j; otherwise, a value of 0 is assigned. Based on the minimum inter-berth operation switchover time, the tidal impact model, and the berth operation preference rule set, the schedulability scores in the matrix are weighted and adjusted to generate a weighted matching matrix, including: Obtain the minimum berth operation switchover time ΔT_cut, which represents the minimum safe interval time reserved between two consecutive vessels at the same berth; Iterate through each schedulable match B ij If the difference between the estimated berthing time of vessel i and the last operation end time of berth j is less than ΔT_cut, its matching score will be reduced. The score is adjusted using a penalty function, expressed as follows: Where ΔT_actual is the actual interval time, α is the adjustment coefficient, W1 is the adjusted score, and ΔT_cut is the minimum job switching time; the adjusted score replaces the original matrix value to construct the switching time correction matrix, and the weighted matching matrix is ​​obtained after correction and fusion processing; The weighted matching matrix is ​​solved using a heuristic optimization algorithm to obtain the optimal solution for berth scheduling, which is the optimal allocation sequence of ships to berths, and satisfies the objective functions of minimum queuing time, minimum tidal conflict and minimum resource waste. The optimal berth scheduling solution is output to the port scheduling control system and linked with real-time AIS data to dynamically update the ship arrival sequence and berth allocation, and continuously optimize the scheduling.

2. The port berth optimization scheduling method based on big data according to claim 1, characterized in that: The historical vessel berthing data includes, but is not limited to: vessel type information, estimated arrival time, actual berthing time, berth number, operation time, departure time, tidal data and meteorological data during berthing.

3. The port berth optimization scheduling method based on big data according to claim 1, characterized in that: The method of solving the weighted matching matrix using a heuristic optimization algorithm includes: Using a weighted matching matrix as input, the set of ships and the set of berths are modeled as a bilateral matching problem, and the objective function is set as minimizing ship queuing time, minimizing the number of tidal conflicts, and minimizing berth idling rate. A genetic algorithm is used to initialize multiple scheduling candidate solutions, each of which represents a sequence of ship-to-berth allocations. The fitness of candidate solutions is calculated, and the fitness function uses the weighted combination of the weighted score matrix value and the objective function as the evaluation index. The candidate solutions are iteratively updated through selection, crossover, and mutation operations until the preset number of iterations or fitness convergence conditions are reached, and the optimal berth scheduling solution is output.

4. The port berth optimization scheduling method based on big data according to claim 3, characterized in that: The above is linked with real-time AIS data, including: Real-time acquisition of location, speed, and estimated arrival time of the target vessel from its AIS system; Clean and predictive correct the AIS data to generate the updated arrival time T_real; Compare T_real with the original estimated arrival time. If the deviation exceeds the preset threshold of 30 minutes, the scheduling update process is triggered. During the update process, the original optimal solution is used as the initial solution, and the heuristic optimization algorithm is run again to quickly generate a new berth scheduling solution.

Images