Coastal crude oil transportation prediction and early warning system and method based on supply and demand whole chain

Abstract

The application discloses a kind of based on supply and demand whole chain's coastal crude oil transportation prediction early warning system and method, belong to energy and logistics technical field.The system includes: data base layer, can collect multi-source data and carry out pre-processing, to provide standardized, high-accuracy multi-source heterogeneous data;Prediction core layer, can build three-level progressive multidimensional prediction model based on ship, traffic and demand, realize from microcosmic ship trajectory to macroscopic supply and demand matching prediction coverage;Risk control layer, can build three-dimensional progressive early warning evaluation model based on port, sea transport and hinterland, realize from local node risk to whole chain supply chain risk early warning coverage, link prediction and decision-making;Ground application layer, can integrate the technical achievements of the first three layers, convert it into intuitive, operable decision result.The application can realize real-time monitoring, accurate prediction and intelligent early warning of crude oil market through multi-source data fusion, machine learning modeling and dynamic early warning mechanism.

Description

Technical Field

[0001] This invention belongs to the field of energy and logistics technology, specifically relating to a coastal crude oil transportation forecasting and early warning system and method based on the entire supply and demand chain. Background Technology

[0002] As a major energy importer, the stability of maritime transport is crucial to national energy security. In recent years, the global shipping market has experienced frequent short-term disruptions, and the supply and demand balance of crude oil has a critical impact on national economic development and energy security. Traditional methods for forecasting crude oil supply and demand mainly rely on econometric models or time series analysis, which have the following limitations:

[0003] 1. Limited data sources: Relying solely on structured data such as historical production and consumption, lacking integrated analysis of unstructured data such as weather events. Low integration of multi-source data, failing to establish a linkage analysis mechanism between "maritime supply and hinterland demand".

[0004] 2. Insufficient real-time capability: Port transportation forecasts rely heavily on annual historical data, which has a long data acquisition cycle and limited channels, making it difficult to meet the needs of short-term situation forecasting and dynamic assessment and early warning. It is also unable to process massive amounts of dynamic data in real time, making it difficult to capture the immediate impact of sudden market changes (such as adjustments to the OPEC+ production cut agreement, escalation of geopolitical conflicts, etc.) on supply and demand.

[0005] 3. Low prediction accuracy: Traditional models have limited ability to fit nonlinear relationships and complex dynamic patterns. In particular, under the background of structural changes such as accelerated replacement of new energy sources and strengthened carbon constraints, the prediction error increases significantly.

[0006] 4. Lagging early warning mechanism: The system lacks dynamic threshold settings and multi-dimensional risk assessment, making it unable to promptly identify potential risks such as supply-demand gaps and abnormal price fluctuations. The early warning system only covers a single link, lacking comprehensive early warning capabilities across the entire chain of macro-level security, from "ship micro-behavior" to "shipping supply chain" to "industrial supply chain."

[0007] Therefore, there is an urgent need to build a comprehensive system that integrates real-time data processing, intelligent prediction algorithms, and visual decision support to enhance the ability of coastal energy transportation to cope with short-term risks. Summary of the Invention

[0008] The present invention aims to at least partially solve one of the technical problems in the aforementioned related technologies.

[0009] Therefore, the purpose of this invention is to provide a coastal crude oil transportation forecasting and early warning system and method based on the entire supply and demand chain. Through multi-source data fusion, machine learning modeling and dynamic early warning mechanism, it can realize real-time monitoring, accurate forecasting and intelligent early warning of the crude oil market.

[0010] To solve the above-mentioned technical problems, the present invention is implemented as follows:

[0011] This invention provides a coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain. The system includes:

[0012] The data infrastructure layer is configured to collect and preprocess multi-source data, providing standardized, highly accurate multi-source heterogeneous data for subsequent prediction and early warning.

[0013] The core prediction layer is configured to build a three-level progressive multi-dimensional prediction model based on ships, traffic volume and demand, achieving prediction coverage from micro-level ship trajectories to macro-level supply and demand matching, and providing quantitative support for subsequent early warning.

[0014] The risk management layer is configured to construct a three-dimensional, progressive early warning and evaluation model based on ports, shipping, and hinterland, achieving early warning coverage from local node risks to full-chain supply chain risks, and connecting forecasting and decision-making; and,

[0015] The application layer is configured to integrate the technological achievements of the data infrastructure layer, the prediction core layer, and the risk management layer, transforming them into intuitive and actionable decision-making results.

[0016] In addition, the coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain according to the present invention may also have the following additional technical features:

[0017] In some implementations, the multi-source data collected by the data infrastructure layer is full-chain data covering maritime transport, ports, and hinterland, including AIS real-time data, port data, and hinterland industry data.

[0018] In some implementations, the data infrastructure layer preprocesses the data, including triple verification and data cleaning.

[0019] The triple verification process includes spatiotemporal consistency verification, speed logic verification, and historical behavior verification.

[0020] The spatiotemporal consistency verification includes removing abnormal data of ships with the same MMSI code that cross sea areas within a certain time period;

[0021] The speed logic verification includes marking crude oil tanker data whose speed is less than a first judgment threshold or greater than a second judgment threshold when not anchored; the second judgment threshold is greater than the first judgment threshold.

[0022] The historical behavior verification includes triggering manual review when the deviation between the current route and the average route over a recent period exceeds a certain distance.

[0023] In some implementations, the progressive logic of the three-level progressive multidimensional prediction model is: progression from the micro level to the meso level and then to the macro level.

[0024] The micro-layer is configured to accurately locate ship dynamics using a ship trajectory prediction model, enabling minute-level predictions of transit time in key sea areas and arrival time at coastal ports.

[0025] The meso-level layer is configured to predict port volume based on ship trajectory prediction results and use an import volume assessment and prediction model to achieve progressive calculation from single ship cargo volume to overall port arrival volume.

[0026] The macro layer is configured to use the port volume forecast results calculated by the meso layer as a basis, combined with hinterland industry data, and to use the hinterland demand forecasting model to predict the demand for imported crude oil by sea.

[0027] In some of these implementations, the ship trajectory prediction model at the micro-layer is based on an improved LSTM and DQN fusion algorithm, and adds a marine feature embedding function to convert meteorological data and waterway attributes into feature vectors, which are then concatenated with the ship's historical trajectory and real-time position and input into the model.

[0028] In some implementations, the mesoscopic prediction step includes:

[0029] S1. Single ship empty / full load determination: The weighted k-means clustering model with adaptive initial center is adopted, with draft as the core feature, combined with speed and heading angle, and calibrated with reference to the ship's design draft range and historical data to output the empty / full load status;

[0030] S2. Four-level system calculation: sequentially perform ship arrival forecasting and output a list of arrivals from 1 to 30 days, combine draft and ship type to assess single ship cargo capacity, assess arrival time sequence, and extrapolate trade scale by correcting deviation coefficients.

[0031] In some implementations, the core steps of the macroscopic layer include:

[0032] Construct a multi-level fluctuation characteristic system, including storage and transportation node characteristics, industrial chain transmission characteristics, and spatiotemporal correlation characteristics;

[0033] Two-stage feature selection: In the first stage, LASSO regression was used to retain features with correlation coefficients > 0.6; in the second stage, grey relational analysis-mutual information fusion scoring was used to select core variables with a score of Scorei > 0.8.

[0034] Dynamic correction: The MAPE error is calculated using newly collected data at regular intervals, and the model parameters are adjusted through online learning to cope with sudden fluctuations in the industry.

[0035] In some implementations, the three-dimensional progressive logic of the risk management layer is as follows:

[0036] Local Nodes: Port Adaptive Early Warning Focusing on Port Unloading Capacity Risks; Specific Contents include: Using import volume forecasts as input, comparing port unloading capacity, identifying congestion and unloading gap risks, including three levels of early warning: Yellow warning is when the predicted arrival volume for the next 7 days > 120% of the port's average daily unloading capacity; Orange warning is when C > 150% and berth occupancy rate > 90%, where C = predicted arrival volume / (port's average daily unloading capacity × T); Red warning is when C > 180% or more vessels waiting at anchor > 20, triggering an emergency unloading plan;

[0037] Transportation segment: International maritime early warning focusing on risks in the maritime transport process; specific content includes: using ship trajectory prediction results as input, combined with sea area risk data, to identify arrival delay and navigation safety risks; early warning trigger conditions include: arrival deviation warning: if the deviation between the predicted arrival time and the planned time is >24 hours, the cause will be analyzed; sea area risk warning: when the risk level of the sea area being traversed rises to high, alternative routes will be automatically recommended, and additional time and costs will be calculated;

[0038] The entire chain focuses on early warning of supply and demand balance in the hinterland, which is crucial for addressing supply and demand matching risks in the supply chain. Specific content includes: using hinterland demand forecasts and import volume forecasts as inputs, combined with real-time inventory, to identify supply chain disruption risks; early warning triggering conditions include: calculating the number of days of available inventory, and triggering an inventory warning when the number of available days is less than the safety threshold; calculating the demand gap, and initiating a supply chain warning when the demand gap is greater than 0.

[0039] This invention also provides a coastal crude oil transportation forecasting and early warning method based on the entire supply and demand chain, implemented using the coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain as described in any of the preceding embodiments. The steps of the method include:

[0040] S1: Collect multi-source data and preprocess it to provide standardized, high-accuracy multi-source heterogeneous data for subsequent prediction and early warning;

[0041] S2: Construct a three-level progressive multi-dimensional prediction model based on ships, traffic volume and demand to achieve prediction coverage from micro-level ship trajectories to macro-level supply and demand matching, providing quantitative support for subsequent early warning;

[0042] S3: Construct a three-dimensional progressive early warning and evaluation model based on ports, shipping, and hinterland to achieve early warning coverage from local node risks to full-chain supply chain risks, and connect forecasting and decision-making;

[0043] S4: Integrate the technological achievements of S1, S2 and S3 and transform them into intuitive and actionable decision-making results.

[0044] In addition, the coastal crude oil transportation forecasting and early warning method based on the entire supply and demand chain according to the present invention may also have the following additional technical features:

[0045] In some implementations, step S4 includes progressive logic for constructing a three-tier architecture;

[0046] The first-layer architecture is based on Hadoop distributed storage and Spark Streaming real-time processing, and introduces time-series databases and edge computing nodes to process multi-source data from S1.

[0047] The second-layer architecture displays the real-time location of global energy transport vessels in terms of ship monitoring, shows the arrival volume trend charts and port percentage pie charts for unloading volume forecasting, realizes dynamic linkage between ship trajectory, meteorological data, arrival volume and inventory in terms of multi-dimensional correlation analysis, and automatically generates decision suggestions for early warning events.

[0048] The third-layer architecture is user-oriented, enabling human-computer interaction and decision implementation.

[0049] Compared with the prior art, the present invention has at least the following beneficial effects:

[0050] In this embodiment of the invention, the coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain provides full-chain data coverage, breaking down information silos and forming a full-chain data network of "maritime supply - port operations - hinterland demand." It achieves, for the first time, the systematic linkage of multi-source heterogeneous data, providing complete data support for subsequent forecasting and early warning. A "ship feature database" is constructed based on the characteristics of crude oil tankers, employing a triple verification mechanism of spatiotemporal consistency, speed logic, and historical behavior to eliminate data such as cross-sea area anomalies, speed anomalies, and route deviations. The processed data accuracy is stable at 92%-95%, far exceeding the accuracy of traditional data processing. Simultaneously, real-time streaming transmission is achieved through message queues and edge computing nodes, improving data processing efficiency by more than 40%.

[0051] In this embodiment of the invention, the coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain is based on an improved LSTM+DQN fusion algorithm. It innovatively adds a "sea area feature embedding layer" to achieve ship trajectory prediction within 1 week to 1 month. The prediction error is ≤30 meters within 2 hours. The transit time of key sea areas and the arrival time of coastal ports can be predicted at the "minute level", laying an accurate foundation for transportation volume assessment.

[0052] In this embodiment of the invention, the coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain innovatively adopts a weighted k-means clustering model with adaptive initial center to determine the empty and full load status of ships (draft depth weight 0.6, combined with speed and tonnage calibration), combined with historical arrival punctuality rate correction, and finally the arrival volume prediction error is ≤8%, supporting the refined allocation of port resources;

[0053] In this embodiment of the invention, the coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain is constructed based on short-term data of the hinterland industry. It constructs a multi-level fluctuation characteristic system of "storage and transportation nodes - industrial chain transmission - spatiotemporal correlation". It selects core variables through LASSO regression + grey relational degree - mutual information fusion, and dynamically corrects the model parameters with online learning algorithms at regular intervals. It can quickly respond to sudden fluctuations in the industry and achieve accurate prediction of the demand for imported crude oil by sea.

[0054] In this embodiment of the invention, the coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain provides the first-ever full-chain early warning system covering "ships-ports-hinterland," breaking through the limitations of traditional single-link early warning and constructing a three-dimensional early warning system. Relying on edge computing nodes (millisecond-level processing of AIS data) and Spark Streaming real-time computing (throughput ≥1000 messages / second), the early warning response time is shortened to within 5 minutes, and each early warning event is associated with "triggering cause-quantified impact-handling suggestions," forming a closed loop of "early warning triggering-solution generation-effect feedback," avoiding the problem of traditional early warnings that "only provide a prompt, but no countermeasures."

[0055] The coastal crude oil transportation forecasting and early warning method based on the entire supply and demand chain of the present invention is implemented using the aforementioned coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain. Therefore, it possesses at least all the features and advantages of the aforementioned coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain, which will not be repeated here. Additional aspects and advantages of the present invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0056] Figure 1 This is a flowchart of a coastal crude oil transportation forecasting and early warning method based on the entire supply and demand chain, disclosed in one embodiment of the present invention. It describes the access paths of AIS data, port operation data, and hinterland industry data, as well as the specific execution steps of the three-fold verification of cargo categories (spatiotemporal consistency verification / ship speed logic verification / historical behavior verification). It illustrates the linkage mechanism from port adaptive early warning, international shipping early warning to hinterland demand early warning, and marks the key indicators (such as port congestion index, arrival deviation rate, and inventory safety threshold) and triggering conditions of each early warning dimension. Detailed Implementation

[0057] 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, not all, of the embodiments of the present invention. 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.

[0058] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings and specific examples and application scenarios.

[0059] In some embodiments of this invention, a short-term forecasting and early warning technology and simulation system for coastal crude oil transportation based on the entire supply and demand chain is provided. Based on a closed-loop approach of "real-time data acquisition - short-term transportation demand forecasting - threshold comparison - early warning triggering - coordinated response," it couples AIS forecasting data with multi-source data such as port operations and industry demand to establish a full-chain early warning and evaluation system from micro-level ship behavior to macro-level supply chain security. By constructing a multi-source heterogeneous data fusion framework, integrating AIS real-time data, port facility and operation data, hinterland industry data, etc., and employing a fusion algorithm of deep learning and reinforcement learning, it achieves accurate prediction of ship trajectories, dynamic assessment of transport volume, and full-chain risk early warning. A three-layer architecture visualization system enables dynamic data display and decision support. This addresses the problems of insufficient short-term forecasting accuracy, weak multi-source data integration capabilities, and incomplete early warning systems in existing technologies, significantly improving the short-term supply chain resilience and industrial chain resilience of coastal energy transportation, and providing scientific support for port planning and emergency decision-making. This includes:

[0060] 1. Multi-source data acquisition and preprocessing

[0061] Multi-source data access unit: collects AIS real-time data (update frequency 1.7-2.5 seconds), port operation data, global port geographic information, hinterland industry data (such as refining and chemical enterprise processing volume and inventory), etc.

[0062] AIS data cleaning unit: Employs a triple verification mechanism (spatiotemporal consistency verification, speed logic verification, and historical behavior verification) to refine target vessel data from crude oil tankers, improving data accuracy to over 90%.

[0063] 2. Prediction Model Construction

[0064] Ship trajectory prediction model: Based on the fusion algorithm of LSTM and reinforcement learning, the model is constructed by taking into account the historical trajectory of the ship, its real-time location and marine meteorological data, and constructing a motion prediction model for crude oil ships on international routes. It can provide minute-level forecasts of the time of passage through key sea areas and arrival at Chinese coastal ports within 1 week to 1 month.

[0065] Import volume assessment and prediction model: Establish a four-level system of "ship arrival prediction - single ship cargo volume assessment - arrival time series assessment - trade scale projection", and combine the ship draft and load mapping model to dynamically calculate the arrival and unloading volume.

[0066] Hinterland Demand Forecasting Model: Based on short-term data such as output, sales, and inventory of key regional industries, fluctuation characteristics are extracted and correlated. Combined with the inherent characteristics of the crude oil storage and transportation system, such as storage capacity, transportation efficiency, and pipeline topology, a multi-level fluctuation characteristic system is constructed. A two-stage feature screening mechanism is adopted to achieve accurate prediction of short-term demand for imported crude oil via sea.

[0067] 3. Construction of Early Warning Evaluation Model

[0068] Port adaptability early warning evaluation method: Integrating arrival forecast data and port berth resource data to provide early warning of congestion risk and unloading capacity gap;

[0069] International maritime early warning assessment method: Based on ship trajectory prediction, monitor the deviation of ship arrival time, and combine the ship's flag and the risk level of key international sea areas it passes through to trigger navigation safety warnings and recommend detour routes;

[0070] Feederland supply and demand balance early warning evaluation method: Based on the forecast of feederland demand, the assessment and forecast of import volume, and the output and inventory data of related industries, when the number of days of crude oil inventory available is lower than the safety threshold of 15 days, the supply chain disruption early warning is activated.

[0071] 4. Design of a Prediction and Early Warning Visualization System

[0072] It adopts a three-tier B / S architecture of "data middle platform - application module - interactive interface", integrates functions such as ship trajectory heat map, cargo volume spatiotemporal analysis, and early warning status display, and supports real-time access on PC web page.

[0073] This invention also discloses a short-term forecasting and early warning technology and simulation system implementation method for coastal crude oil transportation based on the entire supply and demand chain, including the following steps:

[0074] Step 1. Data Acquisition and Preprocessing: Integrate multi-source data and clean and verify it according to crude oil commodity type;

[0075] Step 2. Predictive Model Training: Train the trajectory prediction and cargo volume assessment model using historical AIS data and ship ledger data;

[0076] Step 3. Real-time forecast: Input real-time data and output the forecast results of ship arrivals and unloading volume at my country's coastal ports within 1 week to 1 month;

[0077] Step 4. Full-chain early warning: Based on the coupled analysis of prediction results and multi-source data, trigger tiered early warnings;

[0078] Step 5. Visualization: Present the prediction and early warning information through a three-tier architecture system.

[0079] Example 1:

[0080] This embodiment provides a method for implementing a short-term forecasting and early warning technology and simulation system for coastal crude oil transportation based on the entire supply and demand chain. The detailed implementation details are described below.

[0081] First, the detailed implementation of data acquisition and preprocessing will be explained.

[0082] 1. For multi-source data acquisition and preprocessing, please refer to [link / reference]. Figure 1 As shown.

[0083] AIS Big Data: AIS data is collected through shore-based base stations (update frequency 1.7-2.5 seconds) and satellite receiving terminals (update frequency 4.5-5.5 seconds), covering 28 fields such as ship MMSI code, latitude and longitude, speed, heading, and draft.

[0084] Port Big Data: Connects to the port big data system (to obtain berth information, channel information, vessel ledger data, berth occupancy rate, and loading / unloading efficiency data), customs database (to obtain imported oil product categories and declaration volumes), and industry association platform (to obtain daily processing volume and inventory data of refining and chemical enterprises), and realizes real-time data streaming transmission through Kafka message queue.

[0085] Hinterland Industry Operation Data: Establish a data collection network covering key industries in the crude oil industry chain within the hinterland, regularly acquiring short-term data from core entities such as refineries, chemical manufacturing companies, and crude oil reserves. This data includes, but is not limited to, daily / weekly production of crude oil and its derivatives (such as the output of major products like gasoline, diesel, and ethylene), terminal sales (including actual sales data from online and offline channels), and inventory changes (increases and decreases in crude oil and refined oil inventories). Simultaneously, the collected data undergoes cleaning and processing. Outlier detection algorithms (such as the Z-score method) are used to remove extreme error data, interpolation is employed to supplement missing values, and the data time granularity and statistical standards are standardized to form a standardized short-term dataset.

[0086] 2. AIS data cleaning for crude oil tankers

[0087] To address the specialized characteristics of crude oil tankers, a ship characteristic database was established (e.g., VLCC oil tankers have a draft range of 20-25 meters, while LNG carriers typically have a speed of 16-19 knots).

[0088] The following verification rules shall be adopted:

[0089] Spatiotemporal consistency verification: Remove abnormal data from vessels with the same MMSI code that cross sea areas within 10 minutes;

[0090] Speed ​​logic check: An anomaly is marked when the crude oil tanker's speed is >25 knots or <3 knots (not at anchor).

[0091] Historical behavior verification: Manual review is triggered when the ship’s average route over the past 3 months is compared with the current route and the deviation exceeds 50 nautical miles.

[0092] After processing, the data accuracy stabilized at 92%-95%, meeting the input requirements of the prediction model.

[0093] Secondly, the algorithm implementation of the predictive analysis model will be explained in detail.

[0094] 1. Ship trajectory prediction model based on improved LSTM and reinforcement learning fusion algorithm

[0095] An improved LSTM neural network is employed, adding a "sea area feature embedding layer" to the traditional model. This transforms meteorological data (wind speed, wave height) and channel attributes (width, water depth) of the traversed sea areas into feature vectors. The training dataset contains AIS data (120 million records) of major global energy transport routes for a specific period (e.g., 2021-2023). Samples are generated using a sliding window method (inputting the trajectory of the previous 48 hours, outputting the trajectory of the next 72 hours). The DQN algorithm from reinforcement learning is introduced to optimize the predicted path. When the predicted trajectory approaches a restricted area, the weights are automatically adjusted to keep the prediction error within 30 meters within 2 hours.

[0096] In some embodiments of the present invention, the feature input matrix is ​​constructed as follows:

[0097] Assume the ship's historical trajectory data is a time series. Each time step t i The eigenvectors are defined as follows:

[0098]

[0099] In the formula, Lon i , Lat i Latitude and longitude coordinates (degrees) Speed i For speed (knots) Course i For heading (degrees) Wind i , Wave i For sea area wind speed (m / s) and wave height (m); Draft i Draft (m).

[0100] The feature matrix is ​​standardized as follows:

[0101]

[0102] in, , These are the characteristic mean and standard deviation, calculated based on historical AIS data from the past two years.

[0103] In some embodiments of the present invention, the improved LSTM model structure is as follows:

[0104] A marine feature embedding layer is introduced to transform static channel attributes (such as water depth D and width W) into feature vectors. After being concatenated with the temporal features, it is input into the LSTM unit:

[0105]

[0106] in, H t For the hidden layer state, the weights of key time steps are reinforced through an attention mechanism:

[0107]

[0108]

[0109] W a The attention weight matrix is ​​used as the final output layer to obtain trajectory predictions for the next 1-30 days through a linear transformation.

[0110]

[0111] :future k The predicted latitude and longitude values ​​for each moment;

[0112] W o , b o Output layer weights and biases.

[0113] The model loss function uses the mean squared error with a penalty term:

[0114] .

[0115] 2. Import volume assessment and prediction model

[0116] In some embodiments of the present invention, the construction of the "ship-cargo-port" mapping model is as follows:

[0117] ① Construction of a basic database of single-ship loading capacity

[0118] Establish a basic database of single-ship loading capacity based on ship type (e.g., crude oil tankers are classified into 200,000-ton, 250,000-ton, and 300,000-ton classes according to their tank capacity);

[0119] ② Construction of an evaluation model for empty and fully loaded crude oil tankers

[0120] Considering the "either empty or full" characteristic of international crude oil tankers, and combining real-time draft data from AIS, the empty / full load status of crude oil tankers is determined. A weighted k-means clustering model algorithm based on adaptive initial centers is proposed to determine the empty / full load status of crude oil tankers at sea. The steps include:

[0121] Step 1. Data Acquisition and Preprocessing:

[0122] Collect dynamic characteristics of the target crude oil tanker: real-time draft (core feature), speed (auxiliary feature, speed is usually slightly lower when fully loaded than when unloaded), and rate of change of heading angle (reflects navigation stability, more stable when fully loaded).

[0123] Collect static characteristics: tonnage corresponding to the ship's MMSI code, design draft range (design value for no-load / full-load), and hull type (differences in tanker type affect the baseline of draft).

[0124] The collected data is standardized to eliminate dimensional differences, and a feature matrix F=[ f 吃水 , f 航速 , f 航向 , f 吨位 ].

[0125] Step 2. Construct a weighted k-means clustering model based on adaptive initial centers:

[0126] Initial cluster center optimization: Abandoning the traditional subjective selection method of "the deepest center draft," this method combines the design draft range from the ship's static characteristics. The unloaded and fully loaded design draft values ​​are used as the baseline values ​​for the initial cluster centers. Further calibration is performed using the median draft of similar ships over the past three months to obtain the initial centers. , (Where α and β are historical data correction coefficients, ranging from 0.9 to 1.1).

[0127] Weighted clustering calculation: The weight of the draft feature is set to 0.6 (core influence), the weights of the speed and wave-corrected draft are set to 0.2 and 0.15 respectively, and the total weight of other features is 0.05. The similarity between the sample and the cluster center is calculated by weighted Euclidean distance, and the clustering results are iteratively optimized.

[0128] Clustering result verification: If the distance between the centers of the two clusters is less than 30% of the difference between the ship's design draft when it is empty and when it is fully loaded, then the feature weights are dynamically adjusted (the speed feature weight is increased to 0.3), and the clustering is repeated until the distance threshold is met.

[0129] Step 3. Use an individual difference dynamic calibration mechanism to calibrate the clustering model results:

[0130] To address the differences in draft characteristics among crude oil tankers of different tonnages (such as VLCCs and Suezmaxes), a "tonnage-draft baseline database" is established: the vessels are divided into three levels according to tonnage (100,000-200,000 tons, 200,000-250,000 tons, and 250,000-300,000 tons), and each level stores the 95% confidence interval of the no-load / full-load draft of vessels of the same level over the past year.

[0131] The clustering results are calibrated a second time: if the draft of a ship's clustering result exceeds the confidence interval corresponding to its tonnage class, the state discrimination result is corrected by combining its speed change trend (the speed after leaving the port under full load shows a stable downward trend, and the speed before arriving at the port under empty load shows a stable upward trend). Finally, the result is stored through the State attribute (0=empty load, 1=full load, with a confidence score of 0-100).

[0132] Step 4. Iterative optimization of evaluation results in real time:

[0133] A reinforcement learning mechanism is introduced: using actual port loading and unloading records (such as crude oil loading and unloading documents) as labels, historical discrimination results are fed back for training, and feature weights and cluster centers are dynamically updated.

[0134] The "tonnage-draft baseline database" is updated monthly to incorporate new ship data and correction parameters under extreme weather conditions, ensuring long-term accuracy and stability.

[0135] ③ Short-term import volume forecasting based on a four-level forecasting system

[0136] Ship arrival forecast: Based on the determined fully loaded crude oil vessels, output a list of vessels arriving in port in the next 1-30 days (including estimated arrival time and MMSI code).

[0137] Single vessel cargo capacity assessment: Calculate the single vessel's cargo capacity by combining draft data and vessel type;

[0138] Arrival timing assessment: Summarize arrival volumes by date to identify peak periods (e.g., the 5th-10th of each month is the peak period for imported crude oil arrivals).

[0139] Trade volume projection: The forecast results are revised by taking into account the deviation coefficient between the arrival volume and the actual import volume over the past three months (average 1.08).

[0140] In some embodiments of the present invention, the short-term arrival volume is represented by a time-series superposition model of a four-level forecasting system as follows:

[0141]

[0142] Qi ( t ): No. i A ship in t Cargo capacity at any given moment;

[0143] P i ( t ): No. i A ship in t The probability of arrival at a specific time is calculated based on historical on-time arrival rates.

[0144] In some embodiments of the present invention, the prediction error control formula is as follows:

[0145] .

[0146] 3. Hinterland Demand Forecasting Model

[0147] Based on the short-term data fluctuation characteristics of key industries related to crude oil production in the hinterland, a multi-dimensional correlation analysis model is constructed to achieve accurate prediction of short-term demand for imported crude oil via sea. The steps are as follows:

[0148] ① Fluctuation feature extraction and correlation analysis

[0149] Based on short-term data from key industries related to crude oil production in the hinterland, and combined with the inherent characteristics of the crude oil storage and transportation system, such as storage capacity, transportation efficiency, and pipeline topology, a multi-level fluctuation characteristic system is constructed:

[0150] Storage and transportation node characteristics: Calculate inventory saturation rate and change acceleration (identify replenishment threshold), and extract pipeline transportation volatility and pressure fluctuation coefficient (capture sudden changes in transportation capacity).

[0151] Supply chain transmission characteristics: The three-level inventory lag response coefficient is calculated by using the VAR model, and the processing load fluctuation index is introduced to quantify the impact of unplanned shutdowns on demand.

[0152] Spatiotemporal correlation characteristics: Based on geographical distribution, a transportation radius weight matrix is ​​constructed, and cross wavelet transform is used to identify the time-frequency correlation of fluctuations in storage and transportation nodes.

[0153] A two-stage screening method was adopted: In the first stage, redundant variables were eliminated using the L1 regularization property of LASSO regression (with a penalty coefficient λ=0.01), retaining core features with a correlation coefficient >0.6 with import demand (such as the acceleration of inventory changes in reserves and the refinery processing load fluctuation index); in the second stage, a grey relational analysis-mutual information fusion scoring model was constructed, with the calculation formula as follows: (Score) i =0.7×GRI i +0.3×MI i Select Score i The core variable is greater than 0.8.

[0154] Among them, GRI i MI represents the grey relational degree of the i-th feature (values ​​0-1). i The mutual information value (in bits) is α, and the weighting coefficient is 0.7 (optimized to 0.7 via grid search). The final score is selected. i Features with a value greater than 0.8 are included in the prediction model to ensure that the input variables have both linear and nonlinear explanatory power.

[0155] ② Demand Forecasting and Dynamic Adjustment

[0156] The latest short-term industry data is input into the trained model, which outputs short-term (e.g., 7-day, 15-day, 30-day) forecasts for seaborne crude oil import demand. Simultaneously, a real-time feedback mechanism is introduced. Every 24 hours, based on newly collected industry data and actual import data, the prediction error (e.g., Mean Absolute Percentage Error, MAPE) is calculated. Online learning algorithms dynamically adjust model parameters and correct prediction results, ensuring that even in the event of sudden fluctuations in industry data (e.g., a sudden drop in raw material demand due to temporary maintenance at a large refinery), the predicted values ​​can quickly respond to and reflect changes in actual demand.

[0157] The above methods can accurately map the micro-data fluctuations of hinterland industries to the macro-demand for imported crude oil via sea freight, providing data support for short-term import scheduling and transportation planning, and improving the responsiveness of the crude oil supply chain and the rationality of resource allocation.

[0158] The following section explains the construction of the early warning evaluation model and evaluation system.

[0159] 1. Port Adaptability Early Warning Evaluation Method

[0160] Set three warning thresholds:

[0161] Yellow alert: When the predicted arrival volume in the next 7 days is greater than 120% of the port's average daily unloading capacity (e.g., Qingdao Port's average daily crude oil unloading capacity is 500,000 tons, and the predicted arrival volume is 600,000 tons).

[0162] Orange alert: Meeting both the criteria of excessive arrival volume and berth occupancy rate > 90%;

[0163] Red alert: If there are more than 20 vessels waiting to be anchored, activate the emergency unloading plan (such as coordinating with other ports for diversion).

[0164]

[0165] Q pred Forecast of port arrivals over the next 7 days;

[0166] Q portThe port's average daily unloading capacity;

[0167] T : Predicted number of days.

[0168] Warning threshold:

[0169] Yellow alert: 120% < C ≤ 150%;

[0170] Orange alert: 150% < C ≤ 180% and berth occupancy rate ≥ 90%;

[0171] Red alert: C > 180% or ≥ 20 vessels waiting at anchor.

[0172] 2. International Maritime Early Warning Assessment Methods

[0173] Real-time monitoring of ship navigation status:

[0174] Arrival Deviation Warning: When the predicted arrival time deviates from the planned time by more than 24 hours, analyze the reasons (such as the situation in the Red Sea causing the detour around the Cape of Good Hope) and update the subsequent voyage forecast;

[0175] Marine risk warning: Based on the safety level of the sea area through which the ship passes, when the risk level rises to "high", the system automatically recommends alternative routes and calculates the additional sailing time (approximately 15 days) and cost (increased fuel consumption by 22%).

[0176] Formula for calculating arrival deviation rate:

[0177] In the formula, t pred To predict arrival time; t plan For the planned arrival time.

[0178] Combining marine risk coefficient R (0~1, with 1 for high risk), Comprehensive Early Warning Index:

[0179] W =0.7 D +0.3 R

[0180] when W When the delay rate is ≥20%, a delay warning will be triggered and a detour route will be recommended. The detour time increment will be estimated using the following formula:

[0181]

[0182] L alt , L ori This indicates the distance (nautical miles) between the alternative route and the original route.

[0183] V avg This indicates the average speed (knots).

[0184] 3. Early Warning and Evaluation Method for Supply and Demand Balance in Hinterland

[0185] Establish a "stock-arrival-processing" linkage model:

[0186] ① Obtain real-time inventory data of refineries in the hinterland (such as crude oil inventory of Dalian Petrochemical).

[0187] ② Calculate the number of days of inventory available = current inventory / average daily processing volume;

[0188] ③ When the number of days of crude oil inventory available is less than 15 days and the predicted arrival volume in the next 10 days is less than 70% of the demand gap, a supply chain warning is triggered, and suggestions for multi-channel transportation are pushed (such as activating reserve warehouses and increasing land transportation supplements).

[0189] Hinterland Demand Early Warning Index:

[0190]

[0191] In the formula, I current Indicates the current inventory level; C daily This represents the average daily consumption (such as the daily processing volume of a refining enterprise); 15 days is the safety stock days threshold (for crude oil).

[0192] When S < 100%, an inventory warning is triggered, calculated in conjunction with the arrival volume gap:

[0193] , ;

[0194] When G > 0, the multi-channel transportation plan will be activated.

[0195] The technical implementation of visualization is explained below.

[0196] Adopting a three-tier B / S architecture of "data platform - application module - interactive interface," it achieves the integration of distributed real-time processing and edge intelligence. The visualization system module breaks through the limitations of traditional "data display," constructing a three-in-one interactive visualization platform of "real-time perception - intelligent analysis - decision inference." Through technological architectural innovation and functional upgrades, it achieves a leap from "data presentation" to "decision empowerment." This visualization system not only realizes the dynamic presentation of data but also, through technological innovation, becomes an integrated carrier of "prediction - early warning - decision-making," significantly improving the short-term risk response capability of the coastal energy transportation supply chain.

[0197] 1. Data Platform Layer

[0198] Building upon Hadoop distributed storage (capacity ≥ 10TB) and Spark Streaming real-time processing (throughput ≥ 1000 records / second), a time-series database (such as InfluxDB) and edge computing nodes are introduced. The time-series database is optimized for storing high-frequency time-series data such as ship trajectories and weather data, improving query speed by 3 times. Edge nodes are deployed at port base stations to preprocess AIS real-time data (update frequency 1.7-2.5 seconds), achieving millisecond-level visualization response and solving the latency problem of traditional centralized processing.

[0199] 2. Application Module Layer

[0200] It includes "Ship Monitoring", "Unloading Volume Prediction", "Multi-dimensional Correlation Analysis Module" and "Early Warning Center".

[0201] Ship monitoring module: Real-time display of the location of energy transport vessels at sea around the world, supporting filtering by ship type.

[0202] The unloading volume forecasting module supports trend charts of arrival volume over the past 7, 14, and 30 days, as well as pie charts showing the proportion of arrivals by port.

[0203] Multi-dimensional correlation analysis module: Breaking through the limitations of single charts, it supports dynamic linkage of "ship trajectory - meteorological data - arrival volume". For example, when a VLCC oil tanker is selected, its predicted ship trajectory and sailing path can be automatically overlaid, and the port berth occupancy rate and hinterland refinery inventory trend on the ship's expected arrival day can be displayed simultaneously, revealing the transmission relationship of "weather affecting navigation - arrival delay - inventory fluctuation".

[0204] The intelligent decision-making module for early warning automatically generates a decision suggestion library for triggered early warning events (such as red port congestion warnings). Based on a recommendation algorithm trained on historical data, it displays quantitative options such as "additional costs of diverting to nearby ports" and "a list of refineries for priority unloading," and uses heatmaps to indicate the risk level of each option (e.g., the pirate risk index for detouring around the Cape of Good Hope). It uses a knowledge graph to link "ship MMSI code - port berth" entities, supporting cross-module data penetration (e.g., clicking on an early warning event directly traces to associated ship trajectories and inventory data).

[0205] 3. Interactive Interface Layer

[0206] It supports adaptive display on PC and tablet, and provides functions such as trajectory playback (can trace the ship's navigation path for the past 3 months), early warning details pop-up window (showing the triggering reason and suggested measures), and data export (supports downloading the forecast report for the past 7 days in Excel format).

[0207] An upgraded experience from "passive viewing" to "active exploration".

[0208] Based on the Vue3 framework and Echarts, adaptive interaction is implemented. Core innovations include:

[0209] Immersive spatial and temporal exploration: Supports four levels of zoom: "Global - Regional - Port - Ship". You can intuitively view the sailing attitude of ships in complex waterways (such as the Strait of Malacca). When tracing the trajectory of the past 3 months, abnormal segments are automatically highlighted and the reasons are marked (such as the impact of strong typhoons).

[0210] Customizable analysis workbench: Provides "drag-and-drop dimension configuration" function, allowing users to freely combine analysis dimensions (such as splitting arrival volume by "ship type + month + port"). The system automatically generates dynamic charts (line charts, heat maps, Sankey diagrams, etc.) and supports one-click export of analysis reports with algorithm confidence scores (Excel / PDF format).

[0211] Early warning response linkage and interaction: When an early warning event is triggered, the interface will automatically pop up the "Early Warning Handling Panel", which displays event details on the left (such as the arrival volume exceeding the standard data and berth occupancy rate in the red port congestion warning).

[0212] Any part of this invention not described in detail can be referred to in the prior art or in the art known to those skilled in the art. This embodiment does not limit such part and will not describe it in detail here.

[0213] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain, characterized in that, The system includes: The data infrastructure layer is configured to collect and preprocess multi-source data, providing standardized, highly accurate multi-source heterogeneous data for subsequent prediction and early warning. The core prediction layer is configured to build a three-level progressive multi-dimensional prediction model based on ships, traffic volume and demand, achieving prediction coverage from micro-level ship trajectories to macro-level supply and demand matching, and providing quantitative support for subsequent early warning. The risk management layer is configured to construct a three-dimensional, progressive early warning and evaluation model based on ports, shipping, and hinterland, achieving early warning coverage from local node risks to full-chain supply chain risks, and connecting forecasting and decision-making; and, The application layer is configured to integrate the technological achievements of the data infrastructure layer, the prediction core layer, and the risk management layer, and transform them into intuitive and actionable decision-making results. The progressive logic of the three-level progressive multi-dimensional prediction model is: the progression from the micro level to the meso level and then to the macro level. The micro-layer is configured to accurately locate ship dynamics using a ship trajectory prediction model, enabling minute-level predictions of transit time in key sea areas and arrival time at coastal ports. The meso-level layer is configured to predict port volume based on ship trajectory prediction results and use an import volume assessment and prediction model to achieve progressive calculation from single ship cargo volume to overall port arrival volume. The macro layer is configured to use the port volume forecast results calculated by the meso layer as a basis, combined with hinterland industry data, and to use the hinterland demand forecasting model to predict the demand for imported crude oil by sea.

2. The coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain as described in claim 1, characterized in that, The multi-source data collected by the data infrastructure layer covers the entire chain of maritime transport, ports, and hinterland data, including AIS real-time data, port data, and hinterland industry data.

3. The coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain as described in claim 1, characterized in that, The data infrastructure layer preprocesses the data, including triple verification and data cleaning. The triple verification process includes spatiotemporal consistency verification, speed logic verification, and historical behavior verification. The spatiotemporal consistency verification includes removing abnormal data of ships with the same MMSI code that cross sea areas within a certain time period; The speed logic verification includes marking crude oil tanker data whose speed is less than a first judgment threshold or greater than a second judgment threshold when not anchored. The second judgment threshold is greater than the first judgment threshold; The historical behavior verification includes triggering manual review when the deviation between the current route and the average route over a recent period exceeds a certain distance.

4. The coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain as described in claim 1, characterized in that, The ship trajectory prediction model at the micro level is based on an improved LSTM and DQN fusion algorithm, and adds a marine feature embedding function, which transforms meteorological data and waterway attributes into feature vectors, and splices them with the ship's historical trajectory and real-time position into the model.

5. The coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain as described in claim 1, characterized in that, The prediction steps for the mesoscopic layer include: S1. Single ship empty / full load determination: The weighted k-means clustering model with adaptive initial center is adopted, with draft as the core feature, combined with speed and heading angle, and calibrated with reference to the ship's design draft range and historical data to output the empty / full load status; S2. Four-level system calculation: sequentially perform ship arrival forecasting and output a list of arrivals from 1 to 30 days, combine draft and ship type to assess single ship cargo capacity, assess arrival time sequence, and extrapolate trade scale by correcting deviation coefficients.

6. The coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain as described in claim 1, characterized in that, The core steps of the macroscopic layer include: Construct a multi-level fluctuation characteristic system, including storage and transportation node characteristics, industrial chain transmission characteristics, and spatiotemporal correlation characteristics; Two-stage feature selection: Based on short-term data, fluctuation features are extracted and correlation analysis is performed. Combining inherent characteristics including the storage capacity, transportation efficiency, and pipeline topology of the crude oil storage and transportation system, a multi-level fluctuation feature system is constructed. A two-stage feature selection mechanism is adopted to achieve accurate prediction of short-term demand for imported crude oil via sea. In the first stage, LASSO regression is used to retain features with correlation coefficients > 0.

6. In the second stage, a grey relational analysis-mutual information fusion score is used to select the score. i Core variables > 0.8; Dynamic correction: The MAPE error is calculated using newly collected data at regular intervals, and the model parameters are adjusted through online learning to cope with sudden fluctuations in the industry.

7. The coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain as described in claim 1, characterized in that, The three-dimensional progressive logic of the risk management layer is as follows: Local Nodes: Port Adaptive Early Warning Focusing on Port Unloading Capacity Risks; Specific Contents include: Using import volume forecasts as input, comparing port unloading capacity, identifying congestion and unloading gap risks, including three levels of early warning: Yellow warning is when the predicted arrival volume for the next 7 days > 120% of the port's average daily unloading capacity; Orange warning is when C > 150% and berth occupancy rate > 90%, where C = predicted arrival volume / (port's average daily unloading capacity × T); Red warning is when C > 180% or more vessels waiting at anchor > 20, triggering an emergency unloading plan; Transportation segment: International maritime early warning focusing on risks in the maritime transport process; specific content includes: using ship trajectory prediction results as input, combined with sea area risk data, to identify arrival delay and navigation safety risks; early warning trigger conditions include: arrival deviation warning: if the deviation between the predicted arrival time and the planned time is >24 hours, the cause will be analyzed; sea area risk warning: when the risk level of the sea area being traversed rises to high, alternative routes will be automatically recommended, and additional time and costs will be calculated; The entire chain focuses on early warning of supply and demand balance in the hinterland, which is crucial for addressing supply and demand matching risks in the supply chain. Specific content includes: using hinterland demand forecasts and import volume forecasts as inputs, combined with real-time inventory, to identify supply chain disruption risks; early warning triggering conditions include: calculating the number of days of available inventory, and triggering an inventory warning when the number of available days is less than the safety threshold; calculating the demand gap, and initiating a supply chain warning when the demand gap is greater than 0.

8. A method for forecasting and early warning of coastal crude oil transportation based on the entire supply and demand chain, characterized in that, The method is implemented using the coastal crude oil transportation forecasting and early warning system based on the entire supply and demand chain as described in any one of claims 1-7, wherein the steps of the method include: S1: Collect multi-source data and preprocess it to provide standardized, high-accuracy multi-source heterogeneous data for subsequent prediction and early warning; S2: Construct a three-level progressive multi-dimensional prediction model based on ships, traffic volume and demand to achieve prediction coverage from micro-level ship trajectories to macro-level supply and demand matching, providing quantitative support for subsequent early warning; S3: Construct a three-dimensional progressive early warning and evaluation model based on ports, shipping, and hinterland to achieve early warning coverage from local node risks to full-chain supply chain risks, and connect forecasting and decision-making; S4: Integrate the technological achievements of S1, S2 and S3 and transform them into intuitive and actionable decision-making results.

9. The coastal crude oil transportation forecasting and early warning method based on the entire supply and demand chain as described in claim 8, characterized in that, Step S4 includes the progressive logic for constructing a three-tier architecture; The first-layer architecture is based on Hadoop distributed storage and Spark Streaming real-time processing, and introduces time-series databases and edge computing nodes to process multi-source data from S1. The second-layer architecture displays the real-time location of global energy transport vessels in terms of ship monitoring, shows the arrival volume trend charts and port percentage pie charts for unloading volume forecasting, realizes dynamic linkage between ship trajectory, meteorological data, arrival volume and inventory in terms of multi-dimensional correlation analysis, and automatically generates decision suggestions for early warning events. The third-layer architecture is user-oriented, enabling human-computer interaction and decision implementation.

Images