Capacity Scheduling Based On Supply And Demand Forecasting In The Steel Logistics

The widespread adoption of digital technology in online freight platforms has mitigated the problem of information asymmetry between shippers and drivers,which has propelled the digital transformation of industries such as steel.However,the regional nature of road freight has resulted in significant disparities in the demand for transport resources(i.e.the number of transport vehicles)among shippers in various transport flows.Moreover,when drivers accept freight transport tasks,they factor in various considerations such as the direction and type of goods,as well as their familiarity with certain transport routes based on the types of goods that the truck can carry.As a consequence,the distribution of transport capacity across different transport flows is highly uneven,with insufficient transport capacity leading to severe backlog of goods and exorbitant freight costs in some flows.Hence,there is an urgent need to develop an appropriate scheduling mechanism that can achieve a balance between the supply and demand of transport capacity for all transport flows on the freight platform.In recent years,there has been significant academic and industry research focused on capacity scheduling issues related to online transportation platforms.These platforms rely on demand forecasting to anticipate the need for ride-sharing cars,shared bicycles,and other vehicles in various regions.Capacity is then allocated between regions in advance to minimize passenger wait times.In contrast,steel logistics primarily relies on trucks,such as heavy-duty and super-heavy-duty trucks,to transport goods due to their heavy weight and large volume.Compared to transportation platforms that focus on medium and short-distance transportation,the steel online freight platform faces more limited capacity resources due to multiple long stops during long-distance transportation.This necessitates the prediction of capacity supply based on factors such as vehicle stopping behavior.Additionally,when predicting the actual transportation quantity of goods,enterprise production capacity must be taken into account.Therefore,scheduling capacity for different transportation flows requires careful consideration of the driver’s preferences for transportation flow and cargo type to determine the optimal scheduling strategy.This thesis proposes a capacity scheduling method based on supply and demand forecasting with truck driver preferences.The contributions of this thesis are significant,as it addresses the unique challenges faced by the steel logistics industry and provides a practical solution to optimize capacity scheduling:(1)Self-attention based Capacity Supply Prediction:To meet the demand for steel transportation,it is necessary to predict the availability of idle vehicles(such as vehicle ID)for carrying different flows of goods in the future,including vehicles without transportation tasks and vehicles that can return to the steel plant to complete transportation tasks in the future(referred to as future capacity).To solve the problem of future capacity prediction,it is necessary to evaluate the possibility of each truck returning to the steel plant during the future period(referred to as capacity reachability).The unpredictable stopping time and missing return trajectories of trucks during long-distance transportation pose a serious challenge to accurate capacity reachability prediction.To address these challenges,this thesis analyzes multiple datasets,such as waybills,vehicles,and trajectories,and extracts features such as truck stopping behavior,transportation destination,and environment to effectively model vehicle stopping and driving behavior.Self-attention mechanisms are introduced to obtain the weights of different features that affect the time required for vehicles to return to the steel plant.Based on this,a capacity supply prediction method based on self-attention mechanism is proposed,including three parts: capacity candidate set generation based on historical flow similarity,capacity reachability prediction based on self-attention mechanism,and capacity transport flow prediction based on long short-term memory network model.Finally,extensive experiments are conducted on real datasets.The experimental results show that the capacity prediction method proposed in this thesis has an accuracy rate of up to 83.5%,and can provide strong decision support for tasks such as capacity scheduling optimization in steel logistics.(2)Multi-graph Convolutional Networks based Capacity Demand Forecasting:Given the special requirements of different types of goods for truck models,the capacity demand prediction for steel logistics aims to predict the demand for different types of trucks for different goods in different transportation directions in the future.Considering the upstream and downstream relationships of different types of goods in the supply chain production,i.e.,the production of one product depends on the production of another product,the demand for different types of goods will affect each other when predicting the capacity demand for each transportation direction.Therefore,in addition to considering factors such as the geographical location,time,and weather of the transportation directions,the actual output sequence of different types of goods in each transportation direction should also be included in the impact on demand.The correlation between different types of goods in different transportation directions can be well represented by graphs.To model the comprehensive impact of multiple factors on the capacity demand for each transportation direction,i.e.,to extract the connection relationships in multiple related graphs,this thesis introduces a multi-graph convolutional network suitable for non-Euclidean structured data to model the spatial correlation between transportation directions,and models the time correlation of demand quantity in the time dimension through gated recurrent units.This thesis verifies in a real dataset that the capacity demand prediction model based on multi-graph convolutional network has better prediction performance than the state-of-the-art demand prediction model,reducing the error to 18.5%.(3)Capacity Scheduling for Maximizing Supply-Demand Balance:This thesis extracts the preferences of drivers for cargo types and transportation directions when selecting transportation tasks in steel logistics based on multi-source data such as waybills and vehicles to ensure the successful implementation of transportation capacity scheduling.The optimization objective of capacity scheduling is to balance the supply and demand of transportation capacity in accordance with the drivers’ transportation preferences and the platform’s overall transportation direction.Considering the advantages of σ-constraint in reducing the solution space range and accelerating the solution speed,this thesis uses σ-constraint to transform the multi-objective optimization problem of transportation capacity scheduling into a single-objective optimization problem,and uses integer programming to solve it,obtaining a transportation capacity scheduling strategy that satisfies both the supply and demand balance of transportation capacity and driver satisfaction.The real transportation data set of a steel logistics company in Shandong Province is used for evaluation.The experimental results show that the scheduling strategy proposed in this thesis can meet an average of 87% of demand,and the average waiting time for vehicles to pick up orders at the steel plant is reduced from the original average of 2.5 hours to within 50 minutes.In summary,this thesis proposes a capacity scheduling method that meets the transportation preferences of drivers to address the issue of uneven transportation capacity in the various transportation flows of the steel logistics industry.Firstly,a self-attention mechanism is used to predict the future capacity supply of each flow based on the modeling of vehicle stopping behavior and the prediction of the capacity accessibility of each flow.Secondly,a multi-graph convolutional network model is used to predict the future capacity demand of each flow by considering factors such as cargo capacity,time,space,and weather.Finally,the available capacity for the current and future periods is scheduled in advance based on the differences in capacity supply and demand of each flow in order to achieve a balance between the preferences of drivers for transportation flows and cargo types and the overall supply and demand balance of the freight platform,thereby improving driver satisfaction with capacity scheduling.