Steel product marine ship demand planning method and system based on parameter self-learning

Abstract

The invention relates to a steel product marine ship demand planning method and system based on parameter self-learning, and the method comprises the following steps: obtaining the demand basic data of a steel product and the transport capacity basic data of a marine ship, and respectively constructing a demand pool and a transport capacity pool of ship demands; based on the demand pool and the transport capacity pool, constructing a marine ship demand planning model, solving, outputting a decision variable, and taking the decision variable as a demand planning scheme wherein the marine ship demand planning model comprises a target function and a constraint condition, the constraint condition is autonomously expanded through a machine learning technology, and the target function is used as the demand planning scheme. And performing parameter self-learning on the operation parameters of the marine ship demand planning model through a parameter optimization algorithm to realize parameter autonomous optimization. Compared with the prior art, the method has the advantages of autonomously optimizing parameters, improving ship operation efficiency and operation benefits and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of water transportation scheduling, and particularly to a method and system for planning the demand for steel product water transportation vessels based on parameter self-learning. Background Art

[0002] As an important pillar of the national economy, the steel industry's water transportation scheduling, as the main transportation method for steel finished products, directly affects the competitiveness of the entire industrial chain. Traditional water transportation logistics scheduling relies on manual experience, resulting in problems such as low efficiency and high costs. For upstream steel mills, the traditional scheduling mode mainly focuses on single-base scheduling and fails to utilize the advantages of multi-base collaborative scheduling. At the same time, the small-flow batch collection cycle is relatively long, making it difficult to reduce the inventory cycle. On the other hand, for downstream carriers, manual ship-cargo matching and ship requisition operations are required, and manual intervention is needed for interference events, resulting in problems such as high management costs and low efficiency.

[0003] In the scenarios of direct water transportation scheduling and along-the-way consolidation, based on the inventory and pre-shipment prediction resources of multiple bases, a ship demand planning model is constructed with the factory inventory cycle as the optimization goal. The main difficulties include:

[0004] (1) The constraints of water transportation operations are complex and miscellaneous. Manual collection is incomplete, and some constraint condition data cannot be obtained.

[0005] (2) The efficiency of manually optimizing operation parameters is low, and the operation benefits of the original operation parameters degrade as the business scenario changes. Summary of the Invention

[0006] The purpose of the present invention is to provide a method and system for planning the demand for steel product water transportation vessels based on parameter self-learning to improve the operation efficiency of vessels.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] A method for planning the demand for steel product water transportation vessels based on parameter self-learning includes the following steps:

[0009] Obtain the basic demand data of steel products and the basic capacity data of water transportation vessels, and respectively construct a demand pool and a capacity pool for ship demand;

[0010] Based on the demand pool and the capacity pool, construct a water transportation vessel demand planning model, solve it, and output decision variables. Use the decision variables as the demand planning plan. The water transportation vessel demand planning model includes an objective function and constraint conditions. The constraint conditions are autonomously expanded through machine learning technology, and the operation parameters of the water transportation vessel demand planning model perform parameter self-learning through a parameter optimization algorithm to achieve parameter autonomous optimization.

[0011] Further, the demand pool is used to clean and process the basic demand data of the steel products, and calculate the destination port inventory and the forecasted quantity of ready-to-ship for each flow direction.

[0012] Further, the shipping capacity pool is used to clean and process the basic shipping capacity data of the water transportation vessels, and construct a shipping capacity profile of the vessels, including basic information, frequently traveled routes, vessel trajectories, and real-time positions.

[0013] Further, the objective function aims to minimize the factory cycle time for each flow direction, and the expression of the objective function is:

[0014]

[0015] In the formula, f is the objective value, i is the value range of the ship type number, j is the ship number, k is the departure place, m is the destination, x takes values in the range {0, 1}, and x ijkm is a decision variable, T km is the factory cycle time of the route, and W km is the route weight, representing the operation parameters.

[0016] Further, the decision variables include the transportation route, the requested ship type of the transportation vessel, and the time required to arrive at the port.

[0017] Further, the constraint conditions include basic constraint conditions and extended constraint conditions. The basic constraint conditions include the full load rate constraint condition and the minimum combined loading and departure quantity constraint condition. The extended constraint conditions include the port consolidation rule constraint condition, the variety combined loading constraint condition, and the flow direction ship type constraint condition.

[0018] Further, the steps for obtaining the extended constraint conditions include:

[0019] Obtain the candidate historical ship batch transportation detail data to generate the historical constraint condition mining sample data, including the ship batch number, origin port, destination port, loaded variety, ship type, and ship batch operation revenue;

[0020] Based on the historical constraint condition mining sample data, use the association rule algorithm to analyze the destination port consolidation rules of historical ship batches, and output the consolidatable rule set as the port consolidation rule constraint condition of the water transportation vessel demand planning model;

[0021] Based on the historical constraint condition mining sample data, use the clustering algorithm to cluster the variety distribution of the consolidated ports, and output the combined loading rule set as the variety combined loading constraint condition of the water transportation vessel demand planning model;

[0022] Cluster the transportation ship types of the flow direction vessels, and output the main ship types of the flow direction as the flow direction ship type constraint condition.

[0023] Further, the steps for realizing parameter self-optimization of the operation parameters of the water transportation ship demand planning model include:

[0024] Obtain the review result and implementation benefit of the demand planning scheme output by the water transportation ship demand planning model as the sample data set for parameter self-optimization;

[0025] Set the operation parameter fine-tuning threshold and the frequency of high-frequency operation parameters to generate a set of fine-tuning parameters;

[0026] Based on the sample data set and the set of fine-tuning parameters, use a parameter optimization algorithm to optimize the parameters and output the adjusted operation parameters;

[0027] Conduct an actual benefit evaluation on the adjusted operation parameters and decide whether to update the operation parameters according to the benefit evaluation result. The actual benefit evaluation indicators include the review passing rate of the ship demand, the ship operation efficiency, and the average factory cycle.

[0028] Further, the parameter optimization algorithm includes one of the Epsilon-Greedy algorithm, the UCB algorithm, and the Thompson sampling algorithm.

[0029] The present invention also provides a water transportation ship demand planning system based on parameter self-learning, including:

[0030] A demand pool and a transport capacity pool construction module: used to obtain the basic demand data of steel products and the basic transport capacity data of water transportation ships, and respectively construct a demand pool and a transport capacity pool for ship demand;

[0031] A planning module: used to construct a water transportation ship demand planning model based on the demand pool and the transport capacity pool, solve it, output decision variables, and use the decision variables as the demand planning scheme. The water transportation ship demand planning model includes an objective function and constraint conditions, and the constraint conditions are autonomously expanded through machine learning technology. The operation parameters of the water transportation ship demand planning model are parameter self-learned through a parameter optimization algorithm to realize parameter self-optimization.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] (1) On the basis of the traditional water transportation ship demand scale model based on operations research optimization, the present invention introduces a data-driven self-learning technology, discovers the constraint conditions of the model, and autonomously optimizes the operation parameters in the model, so that the water transportation ship demand scale model can be applied to complex water transportation ship demand scheduling and improve the ship operation efficiency.

[0034] (2) The present invention utilizes historical ship arrangement data to mine the ship arrangement experience of business experts through intelligent algorithms, so as to autonomously generate constraint conditions, reduce the complexity and workload of modeling, and accelerate the modeling speed. At the same time, according to the ship arrangement revenue data, a parameter optimization algorithm is used to optimize the operation parameters, reduce the workload of business operations, and improve the operation revenue of the water transportation ship demand model in the business.

[0035] (3) The water transportation ship demand scale model of the present invention takes minimizing the flow-out factory cycle as the optimization goal, reduces the delivery cycle and transportation cost, improves the ship operation efficiency, and aims to achieve a win-win situation for all parties in logistics coordination. Brief Description of the Drawings

[0036] Figure 1 is a schematic flow chart of the method of the present invention;

[0037] Figure 2 is a schematic structural diagram of the system of the present invention;

[0038] Figure 3 is a flow chart of the mining of stowage-capable terminals based on association rules of the present invention;

[0039] Figure 4 is a flow chart of the mining of stowage-capable terminals based on category distribution clustering of the present invention. Detailed Embodiments

[0040] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and gives the detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0041] Embodiment 1

[0042] This embodiment provides a method for planning the water transportation ship demand of steel products based on parameter self-learning, as Figure 1 shown, the method includes the following steps:

[0043] Step 1: Construct a data-driven framework for the water transportation ship demand planning model.

[0044] 1.1. Construct a framework for water transportation ship demand planning, a constraint mining component, and a parameter optimization component.

[0045] 1.2. Construct a resource demand pool for ship demand, and manage the destination port inventory and pre-shipment forecast volume of each flow direction.

[0046] 1.3. Construct a transport capacity pool for ship demand, manage the transport capacity resources of the platform, and construct a transport capacity portrait of the ship including portrait information such as basic information, frequently traveled routes, ship trajectories, and real-time positions.

[0047] 1.4. Build a water transportation ship demand planning model based on operational research decision-making technology to minimize the optimization objectives, constraints, and decision variables of the flow-to-factory cycle setting model. The constraints are divided into business setting and model learning constraints. The basic business constraints include full load rate constraints, minimum combined shipment departure volume constraints, etc.

[0048] Among them, the optimization objective function is:

[0049]

[0050] In the formula, f is the target value, i is the value range of the ship type number, j is the ship number, k is the departure place, m is the destination, x takes values in the range {0, 1}, x ijkm is the decision variable, T km is the line factory cycle, W km is the line weight, representing the operation parameters.

[0051] The decision variable is the variable that needs to be solved in the water transportation ship demand planning model, that is, the ship use planning scheme, including the output transportation line, the ship type of the requested ship for transportation, and the arrival time at the port.

[0052] 1.5. The basic library for solving the water transportation ship demand planning model includes a general solver and a heuristic algorithm library, supporting exact solution algorithms such as branch and bound, branch and price, Benders decomposition, etc., and heuristic algorithms such as genetic algorithms and particle swarm algorithms to solve the planning model.

[0053] 1.6. Configure an optimization constraint mining component and a parameter optimization component according to the application scenario, learn the constraints based on historical data, and expand the constraints of the basic planning model and optimize the operation parameters. The mined constraints are introduced into the planning model by directly generating constraint modeling formulas or calculating the similarity threshold of the constraints.

[0054] 1.7. Build a simulation and online evaluation component for the planning model, supporting the evaluation of the simulation benefits of the planning model based on historical data and the real-time benefit curve of the evaluation strategy based on online data.

[0055] Step 2: Build a constraint mining component based on machine learning.

[0056] 2.1. Generate historical constraint mining sample data based on the candidate historical ship batch transportation details data, including data such as ship batch number, origin port, destination port, loaded product type, ship type, and ship batch operation benefits.

[0057] 2.2. For the historical constraint mining sample data generated in 2.1, analyze the port combination rules of the destination ports of historical ship batches through the association rule algorithm of the constraint mining component, and output the port combination rule set as the port combination rule constraints of the planning model.

[0058] 2.3. Mine the sample data of historical constraint conditions generated in 2.1. Through the clustering algorithm of the constraint mining component, cluster the variety distribution of consolidated ports, and output the consolidated loading rule set as the variety consolidated loading constraint conditions of the planning model. Cluster the transport ship types of the ships flowing to the port, and output the main ship types of the flow direction as the flow direction ship type constraint conditions of the planning model.

[0059] This embodiment is based on the Figure 3 shown associated-rule-based loadable terminal mining technology, combined with the historical loadable terminal stay duration to mine the variety consolidated loading constraint conditions to determine whether the terminal is loadable.

[0060] Figure 3 : Associated-rule-based loadable terminal mining technology

[0061] 1). Based on the historical ship batch transport route data, count the loading frequencies of the consolidated terminal set to generate the frequent item set of loadable terminals

[0062] 2). Calculate the support, confidence, and lift of the frequent item set of loadable terminals based on the apiori algorithm.

[0063] 3). Based on the apiori algorithm process, calculate the association rules of the candidate terminal consolidated terminal frequent item set to generate the candidate association rule loadable terminals

[0064] 4). Based on the apiori algorithm process, connect and prune the candidate association rule loadable terminal frequent item set, and output the set of loadable terminals as the parameters of the constraint conditions of the loadable terminals.

[0065] 5). In the application stage, the mined historical consolidated terminal rules are used to adjust the real-time consolidated terminal rule parameters through terminal congestion identification.

[0066] 2.4. Add the constraint condition rules output by the constraint mining components in 2.2 and 2.3 to the constraint mining set of the water transportation ship demand planning model framework.

[0067] This embodiment is based on the Figure 4 shown category distribution clustering-based loadable terminal mining technology, combined with the historical route main ship type variety consolidated loading data, to mine the variety consolidated loading constraint conditions to judge the matching result of the main ship type variety consolidated loading of the route.

[0068] Figure 4 : Category distribution clustering-based loadable terminal mining technology

[0069] 1). Based on the historical ship batch transport variety consolidated loading data, including routes, category sets, and category consolidated loading weights

[0070] 2). Construct the variety distribution construction vector of the line, set the k-means clustering algorithm, and set the stop threshold through the within-class scatter WCSS index.

[0071] 3). Traverse the number of k-means class centers, and k-means iteratively calculate the within-class scatter. If the threshold is not reached, continue the calculation.

[0072] 4). When the WCSS stop threshold is met, output the variety distribution cluster center of the line as the variety distribution constraint parameter.

[0073] Step 3: Construct the parameter optimization component of the water transportation ship demand planning model.

[0074] 3.1. Record the manual review results and implementation strategy benefits of the output plan of the water transportation ship demand planning model as the sample data set for the operation parameter optimization strategy.

[0075] 3.2. Construct the Exploration and Exploitation strategy for operation parameter optimization, including

[0076] The operation parameter optimization strategy supports the Epsilon-Greedy algorithm, UCB algorithm, and Thompson sampling algorithm.

[0077] 3.3. Construct the fine-tuning rule strategy and algorithm strategy for operation parameters to conduct the exploration of optimized operation parameters. Support setting the operation parameter fine-tuning threshold and the frequency of high-frequency operation parameters, and generate the set of fine-tuned operation parameters. The fine-tuning algorithm strategy outputs the adjusted operation parameters based on the EE algorithm in Section 3.2.

[0078] 3.4. Construct the operation parameter optimization benefit evaluation strategy to evaluate the actual business benefits of the adjusted operation parameters output in 3.3. The evaluation index set includes the review passing rate of ship demand, ship operation efficiency, average factory cycle, etc.

[0079] 3.5. Update the operation parameters of the water transportation ship demand planning model based on the optimized operation parameters. Conduct an online experiment on the operation parameter optimization. If there is an actual optimization in the evaluation index, add the operation parameter and its evaluation index to the current optimal operation parameter set and update the current online operation parameters.

[0080] Figure 2 It is a data-driven water transportation ship demand planning system constructed by a steel product water transportation ship demand planning method with parameter self-learning of a certain steel logistics platform. The entire application is divided into three parts.

[0081] The first part is the basic data source. By connecting with the logistics platform and TMS (Transportation Management System), it obtains the basic data of the capacity pool and demand pool, conducts data cleaning and data processing, and calculates the resource volume and capacity profile of each flow direction.

[0082] The second part is the main calculation framework of the water transportation ship demand planning model, which includes a framework basic support component and a scheduling strategy component. The basic support component provides a unified solution algorithm, computing resources, and experimental evaluation for the scheduling strategy component. The scheduling strategy component constructs a pluggable scheduling framework, which can configure optimization objectives / constraint conditions / decision variables for ship demand planning. The scheduling strategies to be implemented include multi-loading and multi-unloading, counterflow, strong constraint, profit optimization, etc. At the same time, based on the constraint mining component and parameter optimization component, the scheduling strategy component can realize the autonomous mining of constraint conditions for ship demand planning and the autonomous optimization of operation parameters on the basis of the scheduling framework.

[0083] The third part is the front-end application layer, including a scheduling large screen and scheduling page management. Based on the main calculation framework of the water transportation ship demand planning model, it outputs business functions such as demand pool and capacity pool management, review and adjustment of ship demand, adjustment of operation parameters, and transportation operation tracking for scheduling business personnel.

[0084] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0085] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for planning the demand for steel product water transport vessels based on parameter self-learning, characterized in that: The following steps are involved: Obtain the basic demand data for steel products and the basic capacity data for water-based ships, and build a demand pool and a capacity pool for ship use respectively; A water transport vessel demand planning model is constructed based on the demand pool and the capacity pool, and is solved to output decision variables, and the decision variables are used as a demand planning scheme, wherein the water transport vessel demand planning model includes an objective function and constraints, and the constraints are autonomously expanded through machine learning technology, and the operating parameters of the water transport vessel demand planning model are self-learned through a parameter optimization algorithm to achieve autonomous parameter optimization.

2. The method for planning the demand for steel product water transport vessels based on parameter self-learning according to claim 1 is characterized in that: The demand pool is used to clean and process the basic demand data of the steel products, and calculate the inventory at the destination port and the predicted quantity for shipment in each flow direction.

3. The method for planning the demand for steel product water transport vessels based on parameter self-learning according to claim 1 is characterized in that: The capacity pool is used to clean and process the basic capacity data of the water transport vessel, and to construct a capacity profile of the ship, including basic information, frequently-run routes, ship tracks and real-time locations.

4. The method for planning the demand for steel product water transport vessels based on parameter self-learning according to claim 1 is characterized in that: The objective function aims to minimize the flow to factory cycle, and the expression of the objective function is: In the formula, f is the target value, i is the number of ship types, j is the ship number, k is the departure place, m is the destination, x is in the range of {0, 1}, x ijkm is the decision variable, T km is the line factory cycle, W km is the line weight, which represents the operation parameter.

5. The method for planning the demand for steel product water transport vessels based on parameter self-learning according to claim 1 is characterized in that: The decision variables include the transport route, the type of transport vessel and the time required to arrive at the port.

6. The method for planning the demand for steel product water transport vessels based on parameter self-learning according to claim 1 is characterized in that: The constraints include basic constraints and extended constraints. The basic constraints include full load rate constraints and minimum loading and departure quantity constraints. The extended constraints include port rules constraints, type loading constraints and flow direction ship type constraints.

7. The method for planning the demand for steel product ships in water based on parameter self-learning according to claim 6 is characterized in that: The step of obtaining the extended constraint condition comprises: Obtain candidate historical ship lot transportation details data to generate historical constraint mining sample data, including ship lot number, departure port, destination port, shipping type, ship type and ship lot operation income; Based on the historical constraint conditions, sample data is mined, an association rule algorithm is used to analyze the rules of historical ship batch destination ports, and a port allocation rule set is output as a port allocation rule constraint condition of the water transport ship demand planning model; Based on the historical constraints, sample data is mined, a clustering algorithm is used to cluster the distribution of species in the port, and a set of loading and unloading rules is output as the type loading and unloading constraints of the water transport ship demand planning model; The transport ship types of the ships flowing to the destination are clustered, and the main ship types of the flow are output as the flow ship type constraints.

8. The method for planning the demand for steel product ships in water based on parameter self-learning according to claim 1 is characterized in that: The steps of realizing autonomous optimization of the operating parameters of the water transport ship demand planning model include: Obtaining the review results and implementation benefits of the demand planning scheme output by the water transport ship demand planning model as a sample data set for autonomous parameter optimization; Set the fine-tuning threshold of operation parameters and the frequency of high-frequency operation parameters to generate a fine-tuning parameter set; Based on the sample data set and the fine-tuning parameter set, a parameter optimization algorithm is used to optimize the parameters and output the adjusted operating parameters; Conduct an actual benefit evaluation on the adjusted operating parameters and decide whether to update the operating parameters based on the results of the benefit evaluation. The actual benefit evaluation indicators include the approval rate of ship demand, ship operating efficiency, and average factory cycle.

9. The method for planning the demand for steel product water transport vessels based on parameter self-learning according to claim 1, characterized in that: The parameter optimization algorithm includes one of an Epsilon-Greedy algorithm, a UCB algorithm and a Thompson sampling algorithm.

10. A steel product water transport ship demand planning system based on parameter self-learning, characterized in that: include: Demand pool and capacity pool construction module: used to obtain the basic demand data of steel products and the basic capacity data of water-based ships, and to respectively construct the demand pool and capacity pool for ship use; Planning module: used to construct a water transport vessel demand planning model based on the demand pool and capacity pool, solve it, output decision variables, and use the decision variables as a demand planning scheme, wherein the water transport vessel demand planning model includes an objective function and constraints, and the constraints are autonomously expanded through machine learning technology. The operating parameters of the water transport vessel demand planning model are self-learned through a parameter optimization algorithm to achieve autonomous parameter optimization.

Images