Stockyard inventory prediction method, system, device and medium

Abstract

The present application provides a kind of yard inventory prediction method, system, equipment and medium, the method comprises: obtaining the loading and unloading condition characteristic parameters of the loading and unloading ship planned in target port area in future target time length, and the loading and unloading efficiency of ship is predicted based on loading and unloading condition characteristic parameters;Based on the land transportation condition characteristic sequence and actual land transportation inventory change sequence in past preset time length of target port area, land transportation condition characteristic sequence in future target time length, the target land transportation inventory change sequence of target port area in future target time length is predicted;Based on the loading and unloading efficiency of each operation ship and loading and unloading operation plan, the target water transportation inventory change sequence of target port area in future target time length is obtained;Based on the current yard inventory of target port area, target land transportation inventory change sequence and target water transportation inventory change sequence, the target yard inventory sequence of target port area in future target time length is obtained.The present application can improve the accuracy of port yard inventory prediction.

Description

Technical Field

[0001] This invention belongs to the field of port logistics technology, and in particular relates to a method, system, equipment and medium for predicting yard inventory. Background Technology

[0002] With the development of the automotive industry and international trade, roll-on / roll-off (Ro-Ro) shipping has become an important mode of waterway transportation for large cargo. In the operation of large multi-port Ro-Ro terminals, accurate forecasting of yard inventory is a crucial guarantee for scientifically scheduling ships, avoiding congestion, and reducing allocation costs.

[0003] Traditional yard inventory forecasting methods are mostly based on simple statistical models and rules of thumb, which cannot fully reflect the complexity and dynamism of yard inventory changes, resulting in poor forecast accuracy. Summary of the Invention

[0004] To address the shortcomings of the existing technology, this invention provides a method, system, equipment, and medium for predicting yard inventory, thereby improving the accuracy of port yard inventory prediction.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for predicting yard inventory, comprising: The loading and unloading condition characteristic parameters of the vessels planned to carry out loading and unloading operations in the target port area within the future target time period are obtained, and the loading and unloading operation efficiency of the vessels is predicted by a pre-trained loading and unloading efficiency prediction model based on the loading and unloading condition characteristic parameters. Based on the land transport condition feature sequence and actual land transport inventory change sequence of the target port area within a preset time period and within a preset time granularity in the past, and the land transport condition feature sequence within a future target time period and within the preset time granularity in the future, a pre-trained time series model is used to predict the target land transport inventory change sequence of the target port area within the future target time period and within the preset time granularity. Based on the loading and unloading efficiency and loading and unloading plan of the operating vessels, obtain the target water transport inventory change sequence of the target port area in the future target time period in the unit of the preset time granularity; Based on the current yard inventory of the target port area, the target land transport inventory change sequence, and the target water transport inventory change sequence, a target yard inventory sequence of the target port area within the future target time period is obtained by fusing the target time granularity.

[0006] Furthermore, the loading and unloading efficiency prediction model adopts a random forest model, and the training process of the random forest model includes: Obtain the first training set, the samples in the first training set include the loading and unloading condition characteristic parameters and actual loading and unloading operation efficiency of the corresponding historical operation vessels; Random sampling with replacement is performed on the first training set to generate multiple sample subsets; Multiple decision trees are constructed based on multiple sample subsets in a one-to-one correspondence to obtain the random forest model.

[0007] Furthermore, the process of constructing the decision tree includes: Determine whether the current node meets the preset splitting stop condition; When the current node does not meet the splitting stop condition, the principle is to minimize the total sum of squared residuals of the actual loading and unloading operation efficiency of the samples in the two child nodes after splitting. The loading and unloading condition feature parameters and the candidate splitting thresholds corresponding to each loading and unloading condition feature parameter are traversed. The optimal splitting feature and the corresponding optimal splitting threshold of the current node are determined from the loading and unloading condition feature parameters. The current node is then split into two child nodes according to the optimal splitting feature and the optimal splitting threshold. Each child node obtained from the split is taken as a new current node, and the process returns to the step of determining whether the current node meets the preset splitting stop condition.

[0008] Furthermore, the training process of the time series model includes: Obtain a second training set. The samples in the second training set include historical window data and corresponding prediction window data. The historical window data includes the land transport condition feature sequence of the target port area at multiple consecutive prior times and the corresponding actual land transport inventory change sequence. The prediction window data includes the land transport condition feature sequence and the corresponding actual land transport inventory change sequence at multiple consecutive subsequent times immediately following the multiple prior times. The interval between adjacent times is the preset time granularity. The land transport condition feature sequence and actual land transport inventory change sequence in the historical window data of each sample, as well as the land transport condition feature sequence in the prediction window data, are input together into the time series model to obtain multiple predicted land transport inventory change sequences corresponding to later times. The loss value is calculated based on the difference between the predicted land transport inventory change sequence and the corresponding multiple actual land transport inventory change sequences at later times, and the parameters of the time series model are updated based on the loss value through the backpropagation algorithm until the preset convergence condition is met.

[0009] Furthermore, the loading and unloading condition characteristic parameters include vessel type, type of cargo to be loaded or unloaded, planned number of personnel, work shifts, and weather conditions; and / or The land transport condition feature sequence includes land transport condition features at multiple consecutive times. The land transport condition features include the planned land transport inbound volume and planned land transport outbound volume for the corresponding time and date, as well as the weather information and schedule information for the corresponding time.

[0010] Furthermore, the loading and unloading operation plan includes the operation start time, the planned loading volume, and the planned unloading volume; The steps for obtaining the target water transport inventory change sequence include: For each of the operating vessels in the target port area, based on the vessel's loading and unloading efficiency, operation start time, planned loading volume, and planned unloading volume, the change in waterway inventory caused by the vessel in each time period is calculated according to the preset time granularity to obtain the sequence of waterway inventory changes caused by the vessel in the future target time period, in units of the preset time granularity. The changes in waterway inventory caused by all the vessels operating in the target port area within the target future time period, in units of the preset time granularity, are summed to obtain the target waterway inventory change sequence.

[0011] Furthermore, the method also includes: Based on the target yard inventory sequence, obtain the yard utilization rate of the target port area at different times; The moment when the utilization rate of the storage yard exceeds a preset threshold is marked as a high-congestion-risk moment, and the corresponding congestion information is output.

[0012] Secondly, the present invention provides a yard inventory forecasting system, comprising: The efficiency prediction module is configured to obtain the loading and unloading condition characteristic parameters of the operating vessels that are planned to carry out loading and unloading operations in the target port area within a future target time period, and predict the loading and unloading operation efficiency of the operating vessels based on the loading and unloading condition characteristic parameters using a pre-trained loading and unloading efficiency prediction model. The land transport inventory change prediction module is configured to predict the target land transport inventory change sequence of the target port area in the future target period, in units of the preset time granularity, based on the land transport condition feature sequence and actual land transport inventory change sequence statistically analyzed by the preset time granularity within the past preset period of the target port area, and the land transport condition feature sequence statistically analyzed by the preset time granularity within the future target period. The water transport inventory change prediction module is configured to obtain the target water transport inventory change sequence of the target port area within the future target time period, with the preset time granularity, based on the loading and unloading operation efficiency and loading and unloading operation plan of the operating vessel. The inventory change fusion module is configured to fuse the current yard inventory of the target port area, the target land transport inventory change sequence, and the target water transport inventory change sequence to obtain the target yard inventory sequence of the target port area within the future target time period, with the preset time granularity as the unit.

[0013] Thirdly, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the yard inventory forecasting method as described above.

[0014] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the yard inventory prediction method as described above.

[0015] By adopting the above technical solution, the present invention has at least the following beneficial effects: This invention addresses two core scenarios affecting yard inventory: water transport and land transport. It employs targeted forecasting methods for each: a loading / unloading efficiency prediction model is used to estimate the loading / unloading efficiency of vessels, providing a reliable basis for predicting inventory changes caused by water transport; a time series model is used to fuse historical land transport data with future land transport condition characteristic sequences to predict inventory changes caused by land transport. Based on this, the inventory changes caused by water and land transport are integrated with the current yard inventory level to obtain the target yard inventory sequence, enabling dynamic estimation of future yard inventory. Based on this target yard inventory sequence, dispatchers can predict the yard load at various times in the port area, rationally allocate resources, and avoid yard congestion or idleness, thereby significantly improving the port area's yard operation efficiency and resource utilization, and enhancing the intelligence level and core competitiveness of terminal operations. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the yard inventory forecasting method of the present invention; Figure 2 This is a system architecture diagram of the port area yard inventory prediction system of the present invention; Figure 3 This is a hardware architecture diagram of the electronic device in this invention. Detailed Implementation

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the embodiments will be described in detail below with reference to the accompanying drawings. Obviously, the content described below are some examples or embodiments of the present invention. For those skilled in the art, without creative effort, the technical solutions or means disclosed in the present invention can be applied to other scenarios based on these technical contents.

[0018] It should be understood that the terms "system," "device," "unit," and / or "module" used in this invention are methods for distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0019] Unless otherwise specified, the technical terms used to describe components, elements, etc. in this invention are not specifically singular and may include plural. Generally speaking, terms such as "comprising" or "including" only indicate that explicitly identified steps, elements, or components are included, and these steps, elements, and components do not constitute an exclusive list. For example, the described method or apparatus may also include other steps or components.

[0020] This invention uses flowcharts to illustrate the operational steps performed by the apparatus or system of related embodiments. However, unless otherwise specified, the order in which these steps are described should not be construed as a limitation on the order of execution. Those skilled in the art can adjust the order of these steps based on the knowledge and information conveyed by the embodiments of this invention. Such adjustments include, but are not limited to, reversing the order of steps, merging multiple steps, and splitting a step.

[0021] Unlike container terminals and general cargo terminals, roll-on / roll-off terminals are specialized terminals that connect ships to land via ramps (shovels), allowing cargo (such as automobiles and construction machinery) to be driven directly in and out for loading and unloading operations. This mode of operation is called "roll-on / roll-off," and its core characteristic is that cargo is transported horizontally using its own power or with the help of tractor vehicles, without the need for lifting equipment.

[0022] As mentioned earlier, in the operation of large multi-port roll-on / roll-off terminals, accurate forecasting of yard inventory is crucial for the scientific scheduling of vessels, avoiding congestion, and reducing allocation costs. Traditional yard inventory forecasting methods are mostly based on simple statistical models and rules of thumb, which cannot fully reflect the complexity and dynamism of yard inventory changes, resulting in poor forecast accuracy.

[0023] After in-depth research, the inventors discovered that if the storage yards within the port area are considered as a single large storage yard, then changes in storage yard inventory can be attributed to changes in inventory caused by water and land transportation.

[0024] Inventory changes caused by water transport (referred to as water transport inventory changes) refer to the inflow (unloading) and outflow (loading) of inventory generated through ship loading and unloading operations. The pattern of these changes primarily depends on the efficiency of ship loading and unloading operations. This efficiency, in turn, is influenced by a combination of factors, including ship type (e.g., ro-ro ships, passenger ro-ro ships), cargo type (e.g., vehicles, machinery, general cargo), planned workforce, shift schedules, and weather conditions, exhibiting a strong non-linearity.

[0025] Inventory changes caused by land transportation (referred to as land transportation inventory changes) refer to the changes in inventory resulting from the transport of goods into or out of the port area via external transport vehicles such as flatbed trucks and external container trucks. The patterns of these changes exhibit a clear temporal correlation and are dynamically influenced by the planned land transportation inbound and outbound volumes for the day, as well as land transportation conditions such as weather and schedules.

[0026] Based on the above findings, the inventors proposed a divide-and-conquer forecasting approach: decomposing the overall change in yard inventory into two relatively independent sub-problems: "change in water transport inventory" and "change in land transport inventory." For each sub-problem, a forecasting scheme for inventory changes was designed, targeting the core influencing factors. Finally, the two forecasts were integrated with the current yard inventory level to obtain the yard inventory status at different times within a future period. Specifically: Firstly, regarding the prediction of changes in waterway inventory: considering that vessel loading and unloading efficiency is the core variable determining the amount of change in waterway inventory, a loading and unloading efficiency prediction model is first constructed, using vessel type, cargo type, number of personnel, shifts, weather, and other loading and unloading conditions as input features. This model uses machine learning methods (such as random forest) to train on historical loading and unloading operation data. After training, it can output the expected loading and unloading efficiency (e.g., "vehicles / hour") of each vessel in real time based on the input loading and unloading conditions. On this basis, combined with the loading and unloading operation plans of each vessel (i.e., the planned quantity of cargo to be loaded and unloaded and the start time of operation), the sequence of changes in yard inventory caused by waterway transport at different future times can be calculated.

[0027] Secondly, regarding the prediction of changes in land transport inventory: Land transport inventory changes exhibit a clear time-series dependence; therefore, a time series model is chosen to fit historical land transport data to capture its periodicity and trend patterns. Simultaneously, this model can dynamically incorporate planned land transport inbound and outbound volumes, weather conditions, and transport schedules for the corresponding dates, thereby outputting a more accurate sequence of land transport inventory changes.

[0028] Finally, a fusion calculation is performed: After obtaining the water transport inventory change sequence and the land transport inventory change sequence, the current stockyard inventory quantity is used as the initial value, and the water transport inventory change sequence and the land transport inventory change sequence are superimposed hourly according to a preset time granularity (e.g., hourly) to calculate the stockyard inventory status at different future times.

[0029] By adopting the above technical approach, the originally complex and coupled problem of yard inventory prediction can be transformed into several modelable and solvable sub-problems, providing solid data support for the scientific scheduling of ships, congestion early warning and scheduling cost optimization of ro-ro terminals, and has good application prospects and promotion value.

[0030] Based on this, the present invention provides a method, system, device, and medium for predicting stockyard inventory. The technical solutions provided by various embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0031] Example 1 This embodiment provides a method for forecasting yard inventory, such as... Figure 1 As shown, the method specifically includes the following steps: Step S1: Obtain the loading and unloading condition characteristic parameters of each operating vessel that is planned to carry out loading and unloading operations in the target port area within the future target time period, and based on these loading and unloading condition characteristic parameters, use a pre-trained loading and unloading efficiency prediction model to predict the loading and unloading efficiency of each operating vessel.

[0032] In this embodiment, the target port area refers to the terminal port area for which yard inventory forecasting is required. The yards within this port area are considered as a unified, large-scale yard; that is, the local inventory differences between different internal yard areas are no longer differentiated, but rather the dynamic changes in the total inventory of the port area's yards are viewed from a global perspective. The target port area can be an independent terminal operation area or one of the yards in a large multi-port roll-on / roll-off terminal system. The changes in yard inventory in the target port area are jointly determined by water transport (ship loading and unloading) and land transport (external vehicle transport). This embodiment uses this port area as the forecasting object to predict the inventory change trend of the port area's yards within a future target time period (e.g., 10 days).

[0033] In this embodiment, "operating vessels" refers to vessels that are planned to perform loading and unloading operations in the target port area within a future target time period (e.g., 10 days) and will affect the inventory in the target port yard. Specifically, it includes two categories: one is vessels currently docked in the target port area and performing or about to perform loading and unloading operations; the other is vessels scheduled to arrive in the target port area within the future target time period and perform loading and unloading operations. That is, regardless of whether a vessel is currently docked, as long as its loading and unloading operations will actually occur in the target port area within the future target time period, it falls within the scope of "operating vessels" as defined in this step.

[0034] In this embodiment, the loading and unloading condition characteristic parameters include, but are not limited to, the vessel type, the type of cargo to be loaded or unloaded, the planned number of personnel, the shift schedule, and the weather conditions. These data can be sourced from the terminal operating system, the Automatic Identification System (AIS), and the meteorological service system. For example, vessel type refers to the type of vessel being operated, including roll-on / roll-off (Ro-Ro) ships and passenger Ro-Ro ships, represented by coded values; cargo type refers to the type of cargo to be loaded or unloaded, including vehicles, machinery, and general cargo, represented by coded values; planned number of personnel refers to the number of personnel planned to be assigned to the vessel being operated, represented by numerical values; the shift schedule can be the shift corresponding to the start time of the operation, for example, shift one corresponds to 10 PM to 6 AM the next day, shift two corresponds to 6 AM to 2 PM, and shift three corresponds to 2 PM to 10 PM, represented by coded values; the weather conditions can be the weather forecast for the target port area on the day of the operation start time, such as using a combination of features [T, RH, R, v, V]. vis Characterization, where T represents temperature, RH represents relative humidity, R represents rainfall, v represents wind speed, and V represents wind speed. vis This represents visibility. These multidimensional features collectively constitute the basic vector of the corresponding vessel loading and unloading conditions, providing data support for accurate prediction of subsequent loading and unloading efficiency.

[0035] In some implementations, the loading and unloading efficiency prediction model preferably adopts the random forest model. Random forest is an ensemble learning algorithm based on decision trees. By constructing multiple independent decision trees and averaging or voting on the prediction results of each decision tree, it can effectively reduce the risk of overfitting of a single decision tree and exhibit good robustness and generalization ability in handling high-dimensional features and complex nonlinear relationships.

[0036] In some implementations, the specific training process of the random forest model is as follows: First, the first training set is obtained. Each sample in the first training set corresponds to a complete historical loading and unloading operation of a vessel. The sample includes the loading and unloading condition characteristic parameters corresponding to that operation and the actual loading and unloading efficiency. Among them, the loading and unloading condition characteristic parameters include, but are not limited to, the aforementioned vessel type, the type of cargo to be loaded or unloaded, the planned number of personnel, the shift, and the weather conditions. The actual loading and unloading efficiency is recorded, for example, in units of vehicles per hour. This historical data can be extracted from the terminal operating system, the Automatic Identification System (AIS), and the meteorological service system, and after preprocessing such as cleaning and standardization, a structured training sample set is formed.

[0037] Then, random sampling with replacement is performed on the first training set to generate multiple sample subsets. Specifically, a certain number of samples are randomly selected from the first training set using sampling with replacement, meaning that samples are returned after each round of selection, allowing samples from the same training set to be selected multiple times. By repeating the above sampling process multiple times, multiple sample subsets with the same or similar size as the original training set but different sample distributions can be obtained.

[0038] Finally, multiple decision trees are constructed based on a one-to-one correspondence between multiple sample subsets to obtain the random forest model. Specifically, a decision tree is trained independently for each sample subset. During the growth of the decision trees, instead of selecting the optimal splitting feature from all features at each node split, the best splitting method is sought from a randomly selected feature subset. This strategy further enhances the differentiation between the decision trees. Once all decision trees are constructed, they collectively constitute the random forest model. For a new input sample, the model outputs the prediction efficiency value of each decision tree, and then takes the arithmetic mean of the prediction results of all decision trees as the final prediction efficiency.

[0039] The random forest model obtained through the above training method can effectively capture the complex mapping relationship between loading and unloading condition feature parameters and loading and unloading operation efficiency, thus providing a reliable foundation for subsequent prediction of water transport inventory change sequences.

[0040] In some implementations, the decision tree construction process is as follows: First, determine whether the current node meets the preset splitting stopping conditions. These conditions can be preset according to actual application needs, such as the number of samples in the current node being less than a preset minimum sample number threshold, all samples in the current node having the same actual loading and unloading efficiency value (i.e., zero residual), the current decision tree depth having reached a preset maximum depth limit, or continued splitting no longer significantly reducing the prediction error.

[0041] If the current node meets any of the above splitting stopping conditions, then the node is marked as a leaf node, and splitting will stop. The average of the actual loading and unloading efficiency of all samples within that node will be used as the predicted output value of that leaf node. It should be understood that the root node contains all samples in the corresponding sample subset.

[0042] If the current node does not meet the splitting stopping condition, then the optimal splitting method and the optimal splitting threshold need to be selected for that node. Specifically, the candidate features and their candidate thresholds are searched by minimizing the total sum of squared residuals of the actual loading and unloading efficiency of the samples in the two child nodes after splitting. The sum of squared residuals is calculated as follows: for any child node, the sum of squared differences between the actual loading and unloading efficiency of each sample in that child node and the mean of the actual loading and unloading efficiency of all samples in that child node is calculated. The total cost after splitting is obtained by adding the sum of squared residuals of the two child nodes respectively.

[0043] For example, suppose the current node contains N samples, and the actual loading and unloading efficiency of the i-th sample is y. i Assuming that under a certain candidate feature and corresponding candidate splitting threshold, the current node is divided into left child nodes L (containing N). L (N samples) and right child node R (containing N samples) and right child node R (containing N samples) R (If there are 10 samples), the formula for calculating the total cost (Cost) after splitting is as follows: Cost=

[0044] in, This represents the mean of the actual loading and unloading efficiency of all samples within the left child node. This represents the mean of the actual loading and unloading efficiency of all samples within the right child node. This cost function reflects the sum of the dispersion of the samples within each of the two child nodes around its mean. A smaller cost indicates a more concentrated efficiency of the samples within the two child nodes after the split, resulting in a better splitting effect.

[0045] In this embodiment, each feature in the loading and unloading condition feature parameters and the multiple candidate splitting thresholds corresponding to that feature are traversed, the total cost under each splitting method is calculated, and the feature that minimizes the total cost and the corresponding splitting threshold are selected as the optimal splitting feature and optimal splitting threshold of the current node, respectively.

[0046] Once the optimal splitting feature and optimal splitting threshold are determined, the current node can be split into two child nodes according to these optimal splitting features and thresholds. Specifically, the samples in the current node are compared with the optimal splitting threshold based on their values ​​on the optimal splitting feature. Samples with values ​​less than or equal to the threshold are assigned to the left child node, and samples with values ​​greater than the threshold are assigned to the right child node.

[0047] After the split is complete, each resulting child node is treated as the new current node. The process is then repeated recursively, returning to the step of determining whether the current node meets the preset split stopping condition, until all branch nodes meet the split stopping condition and become leaf nodes. In this way, a complete decision tree is constructed.

[0048] Once all decision trees in the random forest have been constructed as described above, input the loading and unloading condition feature parameters of the vessels operating in the target port area. Each decision tree can independently predict the loading and unloading efficiency of the vessels. The random forest model takes the average of the prediction results of all decision trees as the final predicted value of the loading and unloading efficiency.

[0049] It should be understood that the loading and unloading efficiency prediction model in this embodiment is not limited to the random forest model. It can also use models with nonlinear fitting capabilities, such as gradient boosting trees, support vector regression, or shallow neural networks, which can predict the loading and unloading operation efficiency based on the characteristic parameters of loading and unloading conditions.

[0050] Step S2: Based on the land transport condition feature sequence and actual land transport inventory change sequence of the target port area within the past preset time period and the land transport condition feature sequence within the future target time period and the preset time granularity, a pre-trained time series model is used to predict the target land transport inventory change sequence of the target port area within the future target time period and the preset time granularity.

[0051] In this embodiment, the target port area's storage yard is considered as a unified, large storage yard. The inventory change caused by land transportation refers to the change in storage yard inventory resulting from the transport of goods such as automobiles into or out of the port area via external transport vehicles such as flatbed trucks and external trucks. This change exhibits clear time-series characteristics and is dynamically influenced by factors such as the planned land transportation inbound and outbound volumes for the day, weather conditions, and transport schedules. To accurately predict the land transportation inventory change sequence within the target timeframe, this step employs a pre-trained time-series model for prediction.

[0052] Specifically, firstly, the system retrieves from the terminal operating system the land transport condition feature sequence and the actual land transport inventory change sequence for the target port area within a preset time period (e.g., 1 hour). The preset time period can be set according to forecasted demand, for example, the past 14 days (336 hours). The land transport condition feature sequence includes land transport condition features for multiple consecutive time points. These features include the planned land transport inbound and outbound volumes for the corresponding date, as well as the corresponding weather and schedule information. The planned land transport inbound and outbound volumes can be obtained from the land transport plan input by the customer. For times within the next 24 hours, the weather information for each time point can be the weather forecast information for that time point. For times 24 hours after the next 24 hours, the weather information for each time point can be the weather forecast information for that date. The weather information can also use the aforementioned combined features [T, RH, R, v, V]. vis The shift information reflects the work shift (such as day shift or night shift) that the current time belongs to.

[0053] In this embodiment, the time series model can be, for example, an LSTM model with an encoder-decoder architecture, and its specific training process is as follows: First, a second training set is obtained. The samples in the second training set include historical window data and corresponding prediction window data. The historical window data includes the land transport condition feature sequence of the target port area at multiple consecutive prior times (e.g., 336 hours) and the corresponding actual land transport inventory change sequence. The prediction window data includes the land transport condition feature sequence of multiple consecutive subsequent times (e.g., 240 hours) immediately following the multiple prior times and the corresponding actual land transport inventory change sequence. The interval between adjacent times is the aforementioned preset time granularity (e.g., 1 hour).

[0054] Then, the land transport condition feature sequence and the actual land transport inventory change sequence in the historical window data of each sample, as well as the land transport condition feature sequence in the prediction window data, are input into the time series model to obtain multiple predicted land transport inventory change sequences corresponding to later time periods.

[0055] Finally, the loss value is calculated based on the differences between the predicted land transport inventory change sequence and the corresponding actual land transport inventory change sequences at multiple later time points. The parameters of the time series model are then updated using the backpropagation algorithm based on the loss value until a preset convergence condition is met. The convergence condition can be: the loss value decreases below a preset threshold, the loss value no longer decreases significantly after multiple consecutive training rounds, or the preset maximum number of training rounds is reached.

[0056] After training, the time series model can be used for actual prediction. By inputting the land transport condition feature sequence and actual land transport inventory change sequence of the target port area within a preset time period (e.g., 336 hours) and the land transport condition feature sequence (mainly derived from land transport plans and weather forecasts) within a preset time period (i.e., 240 hours) into the trained time series model, the predicted value of land transport inventory change within the preset time period can be obtained, thus forming the target land transport inventory change sequence.

[0057] It should be noted that the LSTM model of the encoder-decoder architecture is only a preferred implementation. Those skilled in the art can also use other time series prediction models, such as Transformer-based models, gated recurrent units (GRUs), etc. As long as the above prediction function can be achieved, they all fall within the protection scope of this invention.

[0058] Step S3: Based on the loading and unloading efficiency and loading and unloading plan of each operating vessel, obtain the target water transport inventory change sequence of the target port area within a future target time period in a preset time granularity.

[0059] In this embodiment, each operating vessel is configured with a loading and unloading operation plan, which includes, but is not limited to, the planned unloading volume Qdis, the planned loading volume Qload, and the operation start time tstart. Specifically, for operating vessels scheduled to arrive at port, the operation start time tstart is their planned berthing time; for operating vessels that have already arrived at port, the operation start time tstart is the current time, and the planned unloading volume Qdis and planned loading volume Qload represent the remaining unloading volume and remaining loading volume at the current time, respectively.

[0060] In this step, firstly, for each operating vessel in the target port area, based on the vessel's loading and unloading efficiency, start time, planned loading volume, and planned unloading volume, the change in waterway inventory caused by the operating vessel in each time period is calculated according to a preset time granularity to obtain the waterway inventory change sequence caused by the operating vessel in the future target duration, in units of the preset time granularity; then, the waterway inventory changes caused by all operating vessels in the future target duration, divided into each time moment according to the preset time granularity, are summed up to obtain the aforementioned target waterway inventory change sequence.

[0061] Specifically, this step starts from the vessel's operation start time tstart and recursively calculates the inventory change caused by the vessel in each time period according to a preset time granularity Δt (e.g., 1 hour). Simultaneously, the planned unloading volume Qdis and planned loading volume Qload are updated to the remaining unloading and loading volumes at the end of the corresponding time period, until all operations are completed or the preset target duration is exceeded. This yields the water transport inventory change sequence of the operating vessel within the future target duration, using the preset time granularity. For multiple vessels, the water transport inventory changes of each vessel at the same time are summed to obtain the aforementioned target water transport inventory change sequence.

[0062] For each operating vessel, taking the example of unloading before loading, the calculation process of the water transport inventory change sequence for that vessel is explained in detail below: First, let η be the predicted loading and unloading efficiency of the corresponding vessel in step S1 (assuming that the unloading and loading efficiencies are the same; if the loading and unloading condition parameters are different, such as different planned personnel or different types of cargo, different loading and unloading condition parameters can be substituted for unloading and loading in step S1 to obtain the unloading efficiency ηdis and the loading efficiency ηload respectively). At the initial time t=tstart, the operation stage = unloading stage, the remaining unloading volume Rdis=Qdis, and the remaining loading volume Rload=Qload.

[0063] For the current time period, if it belongs to the unloading phase, then calculate the required operation time tneed = Rdis / η for the remaining unloading volume; if tneed ≥ Δt, it means that the entire time period is used for unloading operations, and the completed unloading volume Δ = η Δt represents the change in waterway inventory during that period, which is equal to +Δ. Then, Rdis is updated to Rdis. Δt, the operation phase remains unchanged; if tneed < Δt, it means that the current remaining unloading volume (completed volume Rdis) is completed within the first tneed time period, and the remaining time Δt tneed begins loading, with a loading quantity Δload = η (Δt tneed) (If the loading and unloading efficiency is different, η is replaced by ηload), but it must not exceed the remaining loading capacity Rload. The change in waterway inventory during this period = +Rdis Δload, then update Rdis=0, Rload=Rload Δload, the operation phase switches to the loading phase.

[0064] If the current period is the loading phase, calculate the time required to complete the remaining loading: tneed = Rload / η; if tneed ≥ Δt, it means that the entire period is used for loading, then the completed loading quantity Δ = η. Δt, the change in waterway inventory during this period = Δ, then update Rload=Rload Δ, the stage remains unchanged; if tneed < Δt, all remaining loading is completed within the first tneed time period, and there is no operation in the remaining time, the change in waterway inventory during this period = Rload is calculated, then Rload is updated to 0, and the calculation process ends.

[0065] The changes in waterway inventory for each time period obtained by the above recursive calculation (positive values ​​indicate net inflows and negative values ​​indicate net outflows) are taken as the changes in waterway inventory at the end of the corresponding time period. After arranging them in chronological order, the sequence of waterway inventory changes for the operating vessel in the future target time period with the preset time granularity is obtained.

[0066] It should be understood that, for situations where loading and unloading operations are performed simultaneously, the above recursive logic can be used as a reference to calculate the loading outflow and unloading inflow in parallel for each time period based on the actual loading and unloading efficiency, thereby obtaining the change in water transport inventory for each time period.

[0067] Step S4: Based on the current yard inventory of the target port area, the target land transport inventory change sequence, and the target water transport inventory change sequence, a target yard inventory sequence of the target port area within a future target time period is obtained by merging them into a preset time granularity.

[0068] Specifically, first, let the current time (i.e., the prediction start time) be t, and the current inventory in the target port area at time t (i.e., the inventory in the yard at the prediction start time) be S(t). This inventory can be obtained in real time from the terminal production management system. Assume the preset time granularity is Δt (e.g., 1 hour), and the future target duration includes T time granularities (e.g., 240 hours), corresponding to times t+1, t+2, ..., t+T.

[0069] Meanwhile, assume that the target land transport inventory change sequence predicted in step S2 is denoted as {ΔStruck(t+1), ΔStruck(t+2), ..., ΔStruck(t+T)}, where ΔStruck(t+k) represents the change in yard inventory caused by land transport at time k. Also assume that the target water transport inventory change sequence predicted in step S3 is denoted as {ΔSship(t+1), ΔSship(t+2), ..., ΔSship(t+T)}, where ΔSship(t+k) represents the change in yard inventory caused by water transport at time k.

[0070] In this step, starting with the current actual inventory S(t), the inventory changes caused by land and water transportation are accumulated moment by moment to recursively generate the target yard inventory sequence for each future moment. Specifically, for the k-th future moment, the yard inventory S(t+k) is calculated using the following formula: S k=1,2,...,T Meanwhile, physical constraints must be considered during the calculation process. Due to prediction errors or extreme conditions (such as port throughput exceeding current inventory), the calculated yard inventory may be negative at some point. Negative inventory does not exist in the actual physical scenario; therefore, when the calculated yard inventory S(t+k) < 0 at a certain moment, it is forcibly set to zero. This approach ensures the physical rationality of the inventory sequence and avoids cumulative bias in subsequent recursions, because negative inventory means there is actually no more stock to ship, and subsequent calculations should continue based on zero inventory.

[0071] Arranging the future yard inventory levels in chronological order yields the target yard inventory sequence {S(t+1), S(t+2), ..., S(t+T)}. This sequence visually illustrates the trend of target port yard inventory changes over time within the target future period, including the timing of inventory peaks, inventory troughs, and periods of potential congestion risk.

[0072] In some implementations, early warning information can be output based on the prediction results. For example, firstly, based on the target yard inventory sequence, the yard utilization rate of the target port area at different times can be obtained. Then, the time when the yard utilization rate exceeds a preset threshold is marked as a high-congestion risk time, and the corresponding congestion information is output to prompt dispatchers to take measures such as diversion or accelerated port clearance in advance.

[0073] In addition, the target yard inventory sequence can be visualized in chart form to assist operational decision-making.

[0074] This embodiment organically integrates the prediction results from both water and land transportation dimensions, ultimately outputting a target yard inventory sequence that can be directly used for production scheduling decisions. This provides a quantitative basis for scientifically formulating work plans, avoiding yard congestion, and reducing cross-port transfer costs.

[0075] Example 2 This embodiment provides a yard inventory forecasting system, such as Figure 2 As shown, it includes: an efficiency prediction module 11, a land transport inventory change prediction module 12, a water transport inventory change prediction module 13, and an inventory change fusion module 14, wherein: The efficiency prediction module 11 is configured to perform the aforementioned step S1, namely, to obtain the loading and unloading condition characteristic parameters of the operating vessels that are planned to carry out loading and unloading operations in the target port area within the future target time period, and to predict the loading and unloading operation efficiency of the operating vessels based on the loading and unloading condition characteristic parameters using a pre-trained loading and unloading efficiency prediction model.

[0076] The land transport inventory change prediction module 12 is configured to perform the aforementioned step S2, that is, based on the land transport condition feature sequence and actual land transport inventory change sequence statistically analyzed by preset time granularity within the past preset time period of the target port area, and the land transport condition feature sequence statistically analyzed by preset time granularity within the future target time period, the target land transport inventory change sequence of the target port area within the future target time period is predicted by a pre-trained time series model.

[0077] The water transport inventory change prediction module 13 is configured to perform the aforementioned step S3, that is, based on the loading and unloading efficiency and loading and unloading plan of the operating vessels, to obtain the target water transport inventory change sequence of the target port area in the future target time period in a preset time granularity.

[0078] The inventory change fusion module 14 is configured to perform the aforementioned step S4, that is, based on the current yard inventory of the target port area, the target land transport inventory change sequence, and the target water transport inventory change sequence, to obtain the target yard inventory sequence of the target port area within a future target time period in terms of a preset time granularity.

[0079] Example 3 This embodiment provides an electronic device, which can be represented in the form of a computing device (e.g., a server device), including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein when the processor executes the computer program, it can implement the steps of the yard inventory forecasting method provided in Embodiment 1.

[0080] Figure 3 A schematic diagram of the hardware structure of this embodiment is shown, as follows: Figure 3 As shown, the electronic device 30 specifically includes: At least one processor 31, at least one memory 32, and a bus 33 for connecting different system components (including the processor 31 and the memory 32), wherein: Bus 33 includes a data bus, an address bus, and a control bus.

[0081] The memory 32 includes volatile memory, such as random access memory (RAM) 321 and / or cache memory 322, and may further include read-only memory (ROM) 323.

[0082] The memory 32 also includes a program / utility 325 having a set (at least one) of program modules 324, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0083] The processor 31 executes various functional applications and data processing by running computer programs stored in the memory 32, such as the steps of the yard inventory forecasting method provided in Embodiment 1 of the present invention.

[0084] Electronic device 30 can further communicate with one or more external devices 34 (e.g., keyboard, pointing device, etc.). This communication can be performed via input / output (I / O) interface 35. Furthermore, electronic device 30 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public network, such as the Internet) via network adapter 36. Network adapter 36 communicates with other modules of electronic device 30 via bus 33. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0085] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0086] Example 4 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the yard inventory prediction method provided in Embodiment 1.

[0087] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0088] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of the present invention is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but all such changes and modifications fall within the scope of protection of the present invention.

Claims

1. A method for predicting yard inventory, characterized in that, include: The loading and unloading condition characteristic parameters of the vessels that are planned to carry out loading and unloading operations in the target port area within the future target time period are obtained, and the loading and unloading operation efficiency of the vessels is predicted by a pre-trained loading and unloading efficiency prediction model based on the loading and unloading condition characteristic parameters. Based on the land transport condition feature sequence and actual land transport inventory change sequence of the target port area within a preset time period and within a preset time granularity in the past, and the land transport condition feature sequence within a future target time period and within the preset time granularity in the future, a pre-trained time series model is used to predict the target land transport inventory change sequence of the target port area within the future target time period and within the preset time granularity. Based on the loading and unloading efficiency and loading and unloading plan of the operating vessels, obtain the target water transport inventory change sequence of the target port area in the future target time period in the unit of the preset time granularity; Based on the current yard inventory of the target port area, the target land transport inventory change sequence, and the target water transport inventory change sequence, a target yard inventory sequence of the target port area within the future target time period is obtained by fusing the target time granularity.

2. The yard inventory forecasting method according to claim 1, characterized in that, The loading and unloading efficiency prediction model adopts a random forest model, and the training process of the random forest model includes: Obtain the first training set, the samples in the first training set include the loading and unloading condition characteristic parameters and actual loading and unloading operation efficiency of the corresponding historical operation vessels; Random sampling with replacement is performed on the first training set to generate multiple sample subsets; Multiple decision trees are constructed based on multiple sample subsets in a one-to-one correspondence to obtain the random forest model.

3. The yard inventory forecasting method according to claim 2, characterized in that, The process of constructing the decision tree includes: Determine whether the current node meets the preset splitting stop condition; When the current node does not meet the splitting stop condition, the principle is to minimize the total sum of squared residuals of the actual loading and unloading operation efficiency of the samples in the two child nodes after splitting. The loading and unloading condition feature parameters and the candidate splitting thresholds corresponding to each loading and unloading condition feature parameter are traversed. The optimal splitting feature and the corresponding optimal splitting threshold of the current node are determined from the loading and unloading condition feature parameters. The current node is then split into two child nodes according to the optimal splitting feature and the optimal splitting threshold. Each child node obtained from the split is taken as a new current node, and the process returns to the step of determining whether the current node meets the preset splitting stop condition.

4. The yard inventory forecasting method according to claim 1, characterized in that, The training process of the time series model includes: Obtain a second training set. The samples in the second training set include historical window data and corresponding prediction window data. The historical window data includes the land transport condition feature sequence of the target port area at multiple consecutive prior times and the corresponding actual land transport inventory change sequence. The prediction window data includes the land transport condition feature sequence and the corresponding actual land transport inventory change sequence at multiple consecutive subsequent times immediately following the multiple prior times. The interval between adjacent times is the preset time granularity. The land transport condition feature sequence and actual land transport inventory change sequence in the historical window data of each sample, as well as the land transport condition feature sequence in the prediction window data, are input together into the time series model to obtain multiple predicted land transport inventory change sequences corresponding to later times. The loss value is calculated based on the difference between the predicted land transport inventory change sequence and the corresponding multiple actual land transport inventory change sequences at later times, and the parameters of the time series model are updated based on the loss value through the backpropagation algorithm until the preset convergence condition is met.

5. The yard inventory forecasting method according to claim 1, characterized in that, The loading and unloading condition characteristic parameters include vessel type, type of cargo to be loaded or unloaded, planned number of personnel, work shifts, and weather conditions; and / or The land transport condition feature sequence includes land transport condition features at multiple consecutive times. The land transport condition features include the planned land transport inbound volume and planned land transport outbound volume for the corresponding time and date, as well as the weather information and schedule information for the corresponding time.

6. The yard inventory forecasting method according to claim 1, characterized in that, The loading and unloading operation plan includes the start time of the operation, the planned loading volume, and the planned unloading volume; The steps for obtaining the target water transport inventory change sequence include: For each of the operating vessels in the target port area, based on the vessel's loading and unloading efficiency, operation start time, planned loading volume, and planned unloading volume, the change in waterway inventory caused by the vessel in each time period is calculated according to the preset time granularity to obtain the sequence of waterway inventory changes caused by the vessel in the future target time period, in units of the preset time granularity. The changes in waterway inventory caused by all the vessels operating in the target port area within the target future time period, in units of the preset time granularity, are summed to obtain the target waterway inventory change sequence.

7. The yard inventory forecasting method according to any one of claims 1-6, characterized in that, The method further includes: Based on the target yard inventory sequence, obtain the yard utilization rate of the target port area at different times; The moment when the utilization rate of the storage yard exceeds a preset threshold is marked as a high-congestion-risk moment, and the corresponding congestion information is output.

8. A yard inventory forecasting system, characterized in that, include: The efficiency prediction module is configured to obtain the loading and unloading condition characteristic parameters of the operating vessels that are planned to carry out loading and unloading operations in the target port area within a future target time period, and predict the loading and unloading operation efficiency of the operating vessels based on the loading and unloading condition characteristic parameters using a pre-trained loading and unloading efficiency prediction model. The land transport inventory change prediction module is configured to predict the target land transport inventory change sequence of the target port area in the future target period, in units of the preset time granularity, based on the land transport condition feature sequence and actual land transport inventory change sequence statistically analyzed by the preset time granularity within the past preset period of the target port area, and the land transport condition feature sequence statistically analyzed by the preset time granularity within the future target period. The water transport inventory change prediction module is configured to obtain the target water transport inventory change sequence of the target port area within the future target time period, with the preset time granularity, based on the loading and unloading operation efficiency and loading and unloading operation plan of the operating vessel. The inventory change fusion module is configured to fuse the current yard inventory of the target port area, the target land transport inventory change sequence, and the target water transport inventory change sequence to obtain the target yard inventory sequence of the target port area within the future target time period, with the preset time granularity as the unit.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the computer program, it implements the yard inventory forecasting method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the yard inventory forecasting method as described in any one of claims 1 to 7.

Images