A ship voyage cycle estimation method, system and device
By constructing a ship sailing cycle prediction model based on XGBoost and combining multiple features for personalized prediction and probability calibration, the problem of inaccurate sailing time prediction in quarterly sailing schedules has been solved, achieving more efficient and accurate sailing cycle prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI FIRSTTECH SOFTWARE INC
- Filing Date
- 2026-03-05
- Publication Date
- 2026-07-10
AI Technical Summary
Existing technologies lack precise, personalized, and robust methods for providing sailing time estimates for quarterly shipping schedules, which can comprehensively consider the static attributes of ships, the inherent characteristics of shipping routes, and historical seasonal patterns. This leads to inaccurate estimates, resulting in supply chain risks or wasted capacity.
A ship sailing cycle prediction model is constructed using the XGBoost gradient boosting tree algorithm. Combining features such as ship, port, route, and time period, the model is trained using historical data to perform personalized benchmark predictions, and residual analysis and probability calibration are conducted to finally generate a ship sailing cycle prediction value.
It improved the accuracy of sailing time estimation, enhanced the scientific nature and efficiency of quarterly sailing schedules, increasing accuracy by 20%-40% and planning efficiency by 50%.
Smart Images
Figure CN122367701A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ship operation, and in particular to a method, system and equipment for predicting ship sailing cycles. Background Technology
[0002] Currently, steel manufacturing enterprises, especially large integrated steel enterprises, need long-term transportation plans to ensure the supply of raw materials to their manufacturing bases, because the transportation of raw materials such as iron ore from ocean-going vessels involves long voyages.
[0003] Currently, the methods used to provide sailing time estimates for quarterly sailing schedules are mainly the following, all of which have significant drawbacks:
[0004] 1. Fixed-value method based on human experience: Planners assign a fixed number of sailing days to each fixed route (such as the Australia route, the North Africa route, etc.) based on historical experience. This method is simple and crude, and does not consider the impact of key factors such as differences in ship type and seasonal changes. It can be either too optimistic, causing supply chain risks, or too conservative, leading to wasted capacity.
[0005] 2. Real-time data-driven route travel time prediction services: Maritime services such as "Ships.com" use real-time data like AIS to accurately predict the arrival times of departing vessels. Their advantage lies in integrating real-time vessel positions, sea conditions, weather, and congestion information. This service provides predictions for voyages that have already occurred, while quarterly sailing schedules require estimates for voyages that have not yet occurred. During the planning phase, vessels have not yet departed, so this method is ineffective.
[0006] Therefore, there is a lack of a method that can comprehensively consider the static attributes of ships, the inherent characteristics of routes, and historical seasonal patterns, and provide accurate, personalized, and robust sailing time base generation for future quarterly plans. Summary of the Invention
[0007] The purpose of this invention is to provide a method, system, and device for predicting ship sailing cycles, in order to solve the problems mentioned in the background art.
[0008] To achieve the above-mentioned objectives, one aspect of the present invention provides a method for predicting ship sailing cycles, comprising the following steps:
[0009] Step S1: Select parameters that affect the ship's sailing cycle and collect historical data of the parameters;
[0010] Step S2: The XGBoost gradient boosting tree algorithm is used to construct a ship sailing cycle prediction model, and historical data is used to train the model.
[0011] Step S3: Input the planned quarterly ship navigation data into the trained ship navigation cycle prediction model to obtain the personalized baseline prediction value of the planned ship navigation.
[0012] Step S4: Perform residual analysis and probability calibration on the individualized baseline predictions of the planned ship voyage;
[0013] Step S5: Add the personalized baseline prediction value to the probability calibration value to obtain the estimated ship sailing cycle value.
[0014] Furthermore, the parameters mentioned in step S1 include ship characteristics, port and route characteristics, time cycle characteristics, and operational characteristics, wherein:
[0015] Ship characteristics include ship type and gross tonnage;
[0016] Port and shipping route characteristics include loading and unloading port pairs, route markings, and voyage distance;
[0017] Time cycle characteristics include the planned departure month, planned departure quarter, planned departure day of the week, and whether it is peak season;
[0018] Operational characteristics include the historical average sailing cycle.
[0019] Furthermore, the formula for the personalized baseline prediction model in step S3 is as follows:
[0020] ,
[0021] in, This is the personalized baseline prediction value for the i-th route; The input feature vector includes ship type, port pair, planned departure quarter, etc. The trained XGBoost model; This is the k-th regression tree; F is the set of all regression trees.
[0022] Furthermore, in step S2, the training method involves dividing the historical dataset into a training set and a test set according to time order; labeling and encoding categorical features; standardizing numerical features; and training the ship sailing cycle prediction model using the constructed feature set as input and the target variable, route time, as output.
[0023] Furthermore, the residual analysis and probabilistic modeling in step S4 include the following steps:
[0024] Step S401: Calculate the residual between the actual flight time and the model prediction; where the residual formula is:
[0025] ,
[0026] in: The residual between the actual flight time and the model prediction for the i-th voyage; This refers to the actual sailing time. Personalized baseline predictions;
[0027] Step S402: Analyze the residuals according to the higher-order quantile formula, which is as follows:
[0028] ,
[0029] in: R is the quantile in the residual distribution corresponding to the confidence level α; P(R≤r) is the empirical probability that the residual is less than or equal to r.
[0030] Furthermore, the formula for calculating the estimated ship sailing period in step S5 is as follows:
[0031] ,
[0032] in: This is the final output standard navigation period; Personalized baseline forecasts for new routes; The 80th percentile of the historical residuals represents the upper limit of acceptable normal fluctuations.
[0033] Furthermore, in step S1, for multiple segments of ship navigation, a method of defining target variables segment by segment is used to predict the ship navigation period.
[0034] A second aspect of the present invention provides a ship sailing period prediction system, comprising a data preparation module, a model training module, a benchmark prediction module, a residual analysis module, and a period synthesis module, wherein:
[0035] The data preparation module is used to select parameters that affect the ship's sailing cycle and collect historical data of these parameters;
[0036] The model training module uses the XGBoost gradient boosting tree algorithm to build a ship sailing cycle prediction model and uses historical data for model training.
[0037] The baseline prediction module is used to input the planned quarterly ship navigation plan data into the pre-trained ship navigation cycle prediction model to obtain the personalized baseline prediction value of the planned ship navigation.
[0038] The residual analysis module is used to perform residual analysis and probability calibration on the personalized baseline predictions of planned ship voyages;
[0039] The cycle synthesis module is used to add the personalized baseline prediction value to the probability calibration value to obtain the estimated value of the ship's sailing cycle.
[0040] A third aspect of the present invention provides an electronic device comprising: a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implementing the steps of the above-described method for predicting ship sailing cycles.
[0041] A fourth aspect of the present invention provides a computer-readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the above-described method for estimating the ship's sailing cycle.
[0042] Compared with existing technologies, this system and method have the following advantages:
[0043] 1. This invention provides a method and system for determining the standard cycle of ship navigation based on the fusion of machine learning and probabilistic statistics, which solves the problems of inaccurate navigation time estimation and lack of personalized benchmarks in the formulation of ocean shipping plans.
[0044] 2. The standard cycle provided by this invention comprehensively considers systematic differences in ships, routes, seasons, and historical fluctuation patterns. Compared to fixed values based on human experience, its accuracy as a planning input (average deviation from subsequent actual occurrences) can be improved by 20%-40%, significantly enhancing the accuracy and scientific rigor of planning.
[0045] 3. This invention frees managers from the arduous task of relying on vague experience and manually adjusting plans, improving the efficiency of the shipping schedule planning process by at least 50%. Attached Figure Description
[0046] Figure 1 This is a schematic diagram of a method for predicting ship sailing cycles based on machine learning and probability statistics.
[0047] Figure 2 The histogram and kernel density estimation plot are for the residual set R. Detailed Implementation
[0048] 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, and 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.
[0049] like Figure 1The diagram shown illustrates the method flow of this invention. This embodiment provides a method for predicting ship sailing periods. The overall process is as follows: data preparation → feature engineering → model training → personalized benchmark prediction → residual calculation and analysis → probability distribution modeling → standard period synthesis. Specifically, it includes the following steps:
[0050] Step S1: Select the parameters that affect the ship's sailing cycle and collect historical data of the parameters.
[0051] The actual route data is segmented based on the number of unloading ports, thus exhibiting a multi-segment characteristic.
[0052] Target variable definition:
[0053] For voyages with multiple port calls, they are broken down into several independent segments for modeling.
[0054] The travel time for each segment is defined as follows:
[0055] Actual berthing time at the port of discharge for this segment – Completion time at the port of loading for this segment.
[0056] The total sailing period is the sum of the sailing times for all segments.
[0057] For example: For the voyage "Loading port A → Unloading port B → Unloading port C":
[0058] First segment time = B berthing time – A completion time.
[0059] Second leg time = C berthing time – B completion time.
[0060] Total sailing period = First leg time + Second leg time.
[0061] Select ship navigation cycle parameters, including the following data characteristics:
[0062] Ship characteristics: ship type, gross tonnage;
[0063] Port and shipping route characteristics: loading and unloading ports, shipping route markings, and voyage distance;
[0064] Time cycle characteristics: planned departure month, planned departure quarter, planned departure day of the week, whether it is peak season;
[0065] Operational characteristics: historical average sailing cycle.
[0066] Step S2: The XGBoost gradient boosting tree algorithm is used to construct a ship sailing cycle prediction model, and historical data is used to train the model.
[0067] Because the gradient boosting tree algorithm excels at handling tabular data, mixed-type features, and nonlinear relationships, we chose the XGBoost regression model. The collected historical dataset was divided into training and test sets (in chronological order); categorical features (such as ship type, loading / unloading port pairs, and route identifiers) were labeled and encoded; numerical features were standardized; and the XGBoost regression model was trained using the constructed feature set as input and the target variable, route time, as output. The model will learn the complex mapping relationship from input features to sailing time.
[0068] Step S3: Input the planned quarterly ship navigation data into the trained ship navigation cycle prediction model to obtain the personalized baseline prediction value of the planned ship navigation.
[0069] By utilizing machine learning models, the systematic impact of static and planned factors such as ship attributes, port sequences, and time cycles on sailing time is learned and quantified from historical data. This generates a personalized baseline prediction value for any new planned route, eliminating real-time interference. The personalized baseline prediction value refers to the basic sailing time value predicted by the machine learning model, excluding real-time interference. The formula for the personalized baseline prediction model is as follows:
[0070]
[0071] in: This is the personalized baseline prediction value for the i-th route; The input feature vector includes ship type, port pair, planned departure quarter, etc. The trained XGBoost model; This is the k-th regression tree; F is the set of all regression trees.
[0072] For a new quarterly plan (planned to send a ship from the Yangtze River base to Brazil in Q1 2025), only factors such as ship type, loading and unloading ports, and planned departure quarter are extracted. This set of features is then input into a pre-trained XGBoost model, and the model output is the personalized baseline prediction value for the planned voyage.
[0073] Step S4 involves residual analysis and probability calibration of the personalized baseline forecasts for planned ship voyages. This step uses XGboost to forecast historical data and calculates the residual for each sample. A probability distribution analysis is then performed on the set of historical residuals, and a higher-order quantile (80th quantile) is calculated. This quantile represents the upper limit of acceptable normal fluctuations in operations after excluding known systematic factors. This quantifies and integrates inherent volatility in historical voyages that cannot be explained by the personalized baseline model, upgrading point forecasts to standard periods that incorporate operational tolerances.
[0074] Specifically, the following steps are included:
[0075] Step S401: For the i-th voyage in the historical dataset, its residual Ri is defined as the difference between the actual flight time of the voyage and the personalized baseline prediction.
[0076] The residual formula is:
[0077] ,
[0078] in: The residual between the actual flight time and the model prediction for the i-th voyage; This refers to the actual sailing time. This is a personalized baseline prediction.
[0079] Step S402: Calculate the probability calibration value, which refers to the higher-order quantile obtained through historical residual analysis, used to quantify uncontrollable fluctuations. This method does not use traditional methods based on the normal distribution assumption, such as "mean ± 2 standard deviations," but directly analyzes the empirical distribution of the residuals and calculates a specified higher-order quantile from the residual set R.
[0080] The formula for higher-order quantiles is as follows:
[0081] ,
[0082] in: Let be the quantile of the residual distribution corresponding to the confidence level α; P(R≤r) is the empirical probability that the residual is less than or equal to r. In actual calculations, it can be obtained by taking the ⌈α⋅N⌉th value after sorting the residuals. ⌈α⋅N⌉ means rounding up to ensure that the actual sample position corresponding to the αth proportion is obtained. For example, if N=100 and α=0.8, then the 80th value is taken.
[0083] like Figure 2The image shows a histogram and kernel density estimate of the residual set R. The residual distribution should approximate a normal distribution centered at 0 or a t-distribution, indicating that the model has no systematic bias. Calculate the basic statistics of the residuals: mean (which should be close to 0), standard deviation, skewness, and kurtosis. This helps in understanding the amplitude and distribution pattern of fluctuations.
[0084] Step S5: Add the personalized baseline prediction value to the probability calibration value to obtain the estimated ship sailing cycle value.
[0085] The estimated standard cycle value for ship navigation is the sum of the personalized baseline forecast and the probability calibration value, and is used as the input baseline for quarterly sailing schedules.
[0086]
[0087] in: This is the final output standard navigation period; Personalized baseline forecasts for new routes;
[0088] The 80th percentile of the historical residuals represents the upper limit of acceptable normal fluctuations.
[0089] In traditional methods, the P75 (75th percentile, meaning that 75% of the sample values are less than or equal to this value) of the original sailing time is directly taken as the benchmark value. This method ignores systematic differences such as ships, routes, and seasons, and conflates systematic biases with random fluctuations, resulting in a benchmark value that is not scientific or personalized enough.
[0090] It should be noted that the method of this embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this embodiment, and the multiple devices will interact with each other to complete the above method.
[0091] Corresponding to the above embodiments, the present invention also proposes a ship sailing period prediction system, including a data preparation module, a model training module, a benchmark prediction module, a residual analysis module, and a period synthesis module, wherein:
[0092] The data preparation module is used to select parameters that affect the ship's sailing cycle and collect historical data of these parameters;
[0093] The model training module uses the XGBoost gradient boosting tree algorithm to build a ship sailing cycle prediction model and uses historical data for model training.
[0094] The baseline prediction module is used to input the planned quarterly ship navigation plan data into the pre-trained ship navigation cycle prediction model to obtain the personalized baseline prediction value of the planned ship navigation.
[0095] The residual analysis module is used to perform residual analysis and probability calibration on the personalized baseline predictions of planned ship voyages;
[0096] The cycle synthesis module is used to add the personalized baseline prediction value to the probability calibration value to obtain the estimated value of the ship's sailing cycle.
[0097] The present invention also provides an electronic device, comprising: a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implementing the steps as described in any of the above embodiments.
[0098] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, the present invention also provides a computer-readable storage medium storing computer instructions for causing a computer to perform the methods of any of the above embodiments.
[0099] The aforementioned computer-readable storage medium can be any available medium or data storage device that a computer can access, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MOs), etc.), optical storage (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).
[0100] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for predicting ship sailing cycles, characterized in that, Includes the following steps: Step S1: Select parameters that affect the ship's sailing cycle and collect historical data of the parameters; Step S2: The XGBoost gradient boosting tree algorithm is used to construct a ship sailing cycle prediction model, and historical data is used to train the model. Step S3: Input the planned quarterly ship navigation data into the trained ship navigation cycle prediction model to obtain the personalized baseline prediction value of the planned ship navigation. Step S4: Perform residual analysis and probability calibration on the individualized baseline predictions of the planned ship voyage; Step S5: Add the personalized baseline prediction value to the probability calibration value to obtain the estimated ship sailing cycle value.
2. The method for estimating ship sailing cycles according to claim 1, characterized in that, The parameters mentioned in step S1 include ship characteristics, port and route characteristics, time period characteristics, and operational characteristics, wherein: Ship characteristics include ship type and gross tonnage; Port and shipping route characteristics include loading and unloading port pairs, route markings, and voyage distance; Time cycle characteristics include the planned departure month, planned departure quarter, planned departure day of the week, and whether it is peak season; Operational characteristics include the historical average sailing cycle.
3. The method for predicting ship sailing cycles according to claim 1, characterized in that, The formula for the personalized baseline prediction model in step S3 is as follows: ， in, This is the personalized baseline prediction value for the i-th route; The input feature vector includes ship type, port pair, planned departure quarter, etc. The trained XGBoost model; This is the k-th regression tree; F is the set of all regression trees.
4. The method for predicting ship sailing cycles based on machine learning and probability statistics according to claim 1, characterized in that, In step S2, the training method involves dividing the historical dataset into a training set and a test set according to time sequence; and encoding the categorical features with labels. Standardize numerical features; The ship sailing cycle prediction model is trained by using the constructed feature set as input and the target variable, route time, as output.
5. The method for predicting ship sailing cycles according to claim 1, characterized in that, Step S4, residual analysis and probabilistic modeling, includes the following steps: Step S401: Calculate the residual between the actual flight time of the voyage and the model prediction. Step S402: Analyze the residuals according to the higher-order quantile formula, which is as follows: ， in: R is the quantile in the residual distribution corresponding to the confidence level α; P(R≤r) is the empirical probability that the residual is less than or equal to r.
6. The method for predicting ship sailing cycles based on machine learning and probability statistics according to claim 1, characterized in that, The formula for calculating the estimated ship sailing period in step S5 is as follows: ， in: This is the final output standard navigation period; Personalized baseline forecasts for new routes; The 80th percentile of the historical residuals represents the upper limit of acceptable normal fluctuations.
7. The method for predicting ship sailing cycles according to claim 1, characterized in that, In step S1, for multiple segments of ship navigation, the method of defining target variables segment by segment is used to predict the ship navigation cycle.
8. A ship navigation cycle prediction system, characterized in that, It includes a data preparation module, a model training module, a benchmark prediction module, a residual analysis module, and a periodic synthesis module, among which: The data preparation module is used to select parameters that affect the ship's sailing cycle and collect historical data of these parameters; The model training module uses the XGBoost gradient boosting tree algorithm to build a ship sailing cycle prediction model and uses historical data for model training. The baseline prediction module is used to input the planned quarterly ship navigation plan data into the pre-trained ship navigation cycle prediction model to obtain the personalized baseline prediction value of the planned ship navigation. The residual analysis module is used to perform residual analysis and probability calibration on the personalized baseline predictions of planned ship voyages; The cycle synthesis module is used to add the personalized baseline prediction value to the probability calibration value to obtain the estimated value of the ship's sailing cycle.
9. An electronic device, characterized in that, include: A processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implementing the steps of the ship sailing cycle prediction method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the ship sailing cycle prediction method as described in any one of claims 1 to 7.