Warehousing transportation resource scheduling method and system based on goods quantity prediction

Abstract

Disclosed in the present invention are a warehousing transportation resource scheduling method and system based on goods quantity prediction, solving the problems of planning and scheduling of warehousing transportation resources in a warehousing environment. The technical solution comprises: collecting data to form a data set; preprocessing the collected data set to ensure the quality and accuracy of the data; constructing seasonal data in the preprocessed data set; introducing a bidirectional LSTM model, an input being a historical feature sequence, and an output being a predicted inbound quantity; using a deep learning framework to build a CTA-LSTM model used for predicting the inbound quantity; using a grey wolf optimizer to initialize parameters in the CTA-LSTM model, and using the grey wolf optimizer to perform parameter optimization to determine an optimal parameter configuration in the optimizer; using an adversarial mechanism to perform improvement and adversarial training on the built CTA-LSTM model so as to predict a warehouse goods quantity; building a warehousing simulation model; solving the built warehousing simulation model; and performing transportation resource scheduling on the basis of an obtained optimal solution.

Description

A method and system for scheduling warehousing and transportation resources based on cargo quantity prediction Invention Field

[0001] This invention relates to the application of deep learning technology and long short-term memory networks, as well as the scheduling of warehousing and transportation resources, specifically to a method and system for scheduling warehousing and transportation resources based on cargo quantity prediction. Background Technology

[0002] The development of LSTM (Long Short-Term Memory) builds upon the fields of neural networks and RNNs (Recurrent Neural Networks). Traditional neural networks have limitations when processing sequential data, unable to handle variable-length sequences and long-term dependencies. RNNs are neural network structures specifically designed for processing sequential data, featuring recurrent connections that allow information to propagate within the network and establish temporal dependencies. However, a major problem with traditional RNNs is the vanishing and exploding gradient phenomena. During backpropagation of gradients during training, gradient multiplication over multiple time steps can cause gradients to grow or decrease exponentially, making training difficult. LSTM introduces a mechanism called "gating units," including input, forget, and output gates, to control the flow and retention of information. These gating mechanisms, through the use of intermediate states called "cell states," allow LSTM to selectively remember and forget information, and better handle long-term dependencies. Bidirectional LSTM extends LSTM by introducing recurrent connections in both forward and backward directions, allowing information to propagate in both directions to better capture bidirectional dependencies in sequential data. In this way, the output at the current time step can simultaneously consider the inputs from previous and subsequent time steps, thus capturing the bidirectional dependencies in the sequence data more comprehensively. This enables LSTM to better handle sequence data in a wider range of tasks and domains.

[0003] In a modern warehouse environment, numerous warehousing and transportation resources are required. Cargo handling equipment includes forklifts, pallet lifts, conveyor belts, and AGVs. Warehouse facilities include racks, storage locations, and containers. These facilities are used to store and organize goods, ensuring their safety and efficient management. Loading and unloading equipment includes loading platforms, elevators, and conveyor belts. These devices are used for loading, unloading, and transferring goods. Human resources include warehouse workers and operators. The required warehousing and transportation resources vary depending on the volume of goods handled. Effective scheduling of warehousing and transportation resources is crucial for improving transportation efficiency, reducing costs, and optimizing supply chain processes.

[0004] Invention Overview

[0005] The following provides a brief overview of one or more aspects to offer a basic understanding of them. This overview is not an exhaustive summary of all conceived aspects, nor is it intended to identify key or decisive elements of all aspects, nor to define the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form to prepare for the more detailed descriptions that follow.

[0006] The purpose of this invention is to solve the above-mentioned problems and provide a method and system for scheduling warehousing and transportation resources based on cargo quantity prediction. This solves the problem of planning and scheduling warehousing and transportation resources in relation to the warehousing environment, improves resource utilization, reduces warehousing costs, avoids impacting warehousing and transportation efficiency, and achieves the rational use of resources.

[0007] The technical solution of this invention is as follows: This invention discloses a method for scheduling warehousing and transportation resources based on cargo quantity prediction, the method comprising:

[0008] Step 1: Collect data to form a dataset;

[0009] Step 2: Preprocess the collected dataset to ensure data quality and accuracy;

[0010] Step 3: Construct seasonal data from the preprocessed dataset;

[0011] Step 4: Introduce a bidirectional LSTM model. The input of the bidirectional LSTM model is the historical feature sequence, and the output is the predicted warehouse purchase quantity.

[0012] Step 5: Use a deep learning framework to build an improved long short-term memory network prediction model based on the attention mechanism to predict the quantity of goods entering the warehouse;

[0013] Step 6: Initialize the parameters in the improved long short-term memory network prediction model based on the attention mechanism using the gray wolf algorithm, and use the gray wolf algorithm to fine-tune the parameters to determine the optimal parameter configuration in the algorithm.

[0014] Step 7: Employ an adversarial mechanism to improve and adversarially train the attention-based improved long short-term memory network prediction model built in the previous steps in order to predict the quantity of goods in the warehouse.

[0015] Step 8: Establish a warehouse simulation model;

[0016] Step 9: Solve the warehouse simulation model constructed in Step 8;

[0017] Step 10: Based on the optimal solution obtained in Step 9, schedule transportation resources.

[0018] According to an embodiment of the warehousing and transportation resource scheduling method based on cargo quantity prediction of the present invention, in step 2, the preprocessing includes:

[0019] The annual average method is used to fill in missing values, handle outliers, and delete duplicate values.

[0020] According to an embodiment of the warehousing and transportation resource scheduling method based on cargo quantity prediction of the present invention, in step 5, by calculating attention weights, the long short-term memory network prediction model can focus on the relative distance or relative offset between different time steps; a position encoder is introduced into the attention mechanism to add position information to the time steps in the input sequence.

[0021] According to an embodiment of the warehousing and transportation resource scheduling method based on cargo quantity prediction of the present invention, in step 6, the gray wolf algorithm is combined with bidirectional LSTM. During the training process, the search capability of the gray wolf algorithm is used to optimize the parameters and weights of LSTM to find the optimal solution or near-optimal solution more quickly. Step 6 further includes the processing of tracking and hunting prey.

[0022] According to an embodiment of the warehousing and transportation resource scheduling method based on cargo quantity prediction of the present invention, in step 7, an adversarial bidirectional LSTM is adopted. Given different attention mechanisms and even different training parameters, two network models that are adversarial to each other are trained, and the results are validated using a validation set. The two prediction results are combined to give the final result. The competition between the two models is utilized, in which one model attempts to better predict the cargo quantity, while the other model attempts to generate adversarial examples to interfere with the former's prediction, thereby improving the robustness of the former.

[0023] According to an embodiment of the warehousing and transportation resource scheduling method based on cargo quantity prediction of the present invention, in step 8, an objective function is established to evaluate the efficiency and performance of the warehousing system. The objective function is established based on indicators such as the resource utilization rate, throughput and operating cost of the warehousing system. Based on the objective function, the resource variables required according to the predicted cargo quantity are calculated. Each resource variable includes type and quantity. In the warehousing simulation model, the time cost and economic cost of warehousing and transportation resources are taken into account, and the comprehensive warehousing and transportation cost is minimized while meeting the time constraints.

[0024] According to an embodiment of the warehousing and transportation resource scheduling method based on cargo quantity prediction of the present invention, in step 9, the warehousing simulation model is solved using a linear programming method. The model is based on a multi-objective programming model, where objective 1 is the function for minimizing the economic cost of cargo transportation and objective 2 is the function for minimizing the time cost of cargo transportation volume. In the process of multi-objective optimization, an absolute optimal solution cannot be satisfied. While considering multiple objective functions, it is necessary to ensure that the values ​​of multiple objective functions are all acceptable. Therefore, a linear weighting method is used to process the multi-objective problem.

[0025] This invention also discloses a warehousing and transportation resource scheduling system based on cargo quantity prediction, the system comprising:

[0026] The data collection module collects data to form a dataset;

[0027] The dataset preprocessing module preprocesses the collected datasets to ensure data quality and accuracy.

[0028] The seasonal data construction module constructs seasonal data from the preprocessed dataset.

[0029] The bidirectional LSTM model module introduces a bidirectional LSTM model. The input of the bidirectional LSTM model is the historical feature sequence, and the output is the predicted warehouse purchase quantity.

[0030] The module for building an improved long short-term memory network prediction model based on attention mechanism uses a deep learning framework to build an improved long short-term memory network prediction model based on attention mechanism for predicting the quantity of goods entering the warehouse.

[0031] The Grey Wolf parameter tuning module uses the Grey Wolf algorithm to initialize the parameters in the improved long short-term memory network prediction model based on the attention mechanism, and uses the Grey Wolf algorithm to fine-tune the parameters to determine the optimal parameter configuration in the algorithm.

[0032] The adversarial training module uses an adversarial mechanism to improve and adversarially train the improved long short-term memory network prediction model based on the attention mechanism built in the previous steps, so as to predict the quantity of goods in the warehouse.

[0033] The warehouse simulation model creation module is used to create a warehouse simulation model.

[0034] The warehouse simulation model solving module solves the warehouse simulation model constructed by the warehouse simulation model building module.

[0035] The transportation resource scheduling module is based on the optimal solution obtained from the warehouse simulation model solution module.

[0036] Compared with existing technologies, this invention offers the following advantages: It achieves significant progress in the fields of cargo quantity prediction and resource scheduling optimization by utilizing historical and real-time data to construct an improved Long Short-Term Memory (CTA-LSTM) prediction model based on an attention mechanism. This model not only improves prediction accuracy but also provides strong support for subsequent warehousing and transportation resource scheduling.

[0037] In terms of model construction, this invention employs an improved Long Short-Term Memory network (CTA-LSTM) based on an attention mechanism. This network structure fully leverages the advantages of LSTM in processing sequential data, while introducing an attention mechanism that allows the model to automatically focus on important historical data points and ignore irrelevant information during prediction. This design not only improves prediction accuracy but also makes the model more flexible and robust.

[0038] Based on the predicted cargo quantity, this invention further considers factors such as the availability, cost, and efficiency of warehousing and transportation resources. By establishing a warehousing simulation model, different resource scheduling schemes are simulated and compared, and the model is solved to find the optimal resource scheduling plan.

[0039] Finally, this invention allocates goods to suitable equipment and optimizes transportation routes and sequences based on a resource scheduling plan. This not only improves transportation efficiency and reduces transportation costs, but also ensures timely delivery of goods and enhances customer satisfaction.

[0040] In summary, this invention achieves accurate prediction of cargo quantities over a future period by constructing an improved long short-term memory network prediction model based on an attention mechanism, and formulates an optimized resource scheduling plan based on this prediction. This not only improves the efficiency and cost-effectiveness of warehousing and transportation, but also provides strong support for the sustainable development of enterprises. Attached Figure Description

[0041] The above-described features and advantages of the present invention will be better understood after reading the following detailed description of embodiments of the present disclosure in conjunction with the accompanying drawings. In the drawings, components are not necessarily drawn to scale, and components having similar related characteristics or features may have the same or similar reference numerals.

[0042] Figure 1 shows a flowchart of an embodiment of the warehousing and transportation resource scheduling method based on cargo quantity prediction of the present invention.

[0043] Figure 2 shows a schematic diagram of an embodiment of the warehousing and transportation resource scheduling system based on cargo quantity prediction of the present invention.

[0044] Detailed description of the invention

[0045] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should be noted that the aspects described below with reference to the accompanying drawings and specific embodiments are merely exemplary and should not be construed as limiting the scope of protection of the present invention in any way.

[0046] Figure 1 illustrates the flowchart of an embodiment of the warehousing and transportation resource scheduling method based on cargo quantity prediction of the present invention. Referring to Figure 1, the implementation steps of this embodiment are detailed below.

[0047] Step 1: Collect data to form a dataset.

[0048] The data collection timeframe should cover the past few years, including warehouse inbound and outbound quantities. For LSTM models, time-series data is required, including timestamps and corresponding inbound and outbound volumes.

[0049] Step 2: Preprocess the collected dataset to ensure data quality and accuracy.

[0050] Missing values ​​are filled in using the annual mean method.

[0051] The annual mean method is a data processing technique primarily used to handle missing and outlier values. When dealing with missing values, it first calculates the average value of the data for a specific year or period, and then uses this average to fill in the missing data points. This method assumes that the trend or level of data change is relatively stable or consistent within the same year or time period, therefore the average value can be used to approximate the missing data.

[0052] In addition, the annual mean method can also be used to handle outliers. Outliers are data points that deviate significantly from the overall data distribution. By calculating the annual or long-term mean, a reasonable threshold can be set. Data points exceeding this threshold are considered outliers and then handled accordingly, such as replacing or deleting them with the mean.

[0053] For duplicate values, the annual mean method typically recommends direct deletion. Duplicate values ​​can distort the true distribution and statistical properties of the data, so deleting them is an important step in ensuring the accuracy of data analysis.

[0054] In step 2, the specific processing procedure of the annual mean method is as follows.

[0055] For missing data for a particular day, calculate the average purchase data using purchase data from the same day in different years before and after. Assuming there are data from year m+1, the missing data would be the purchase data for the same day in the remaining m years. i The average value of (where i is a natural number from 1 to m) is calculated using the following formula:

[0056] After supplementing missing values ​​using the methods described above, a weighted moving average (WMA) method is used to reduce noise and fluctuations in the data. This smoothing effect is achieved by calculating the average value of data points within a continuous time window. The moving average is calculated by assigning different weights to data points at different time points, based on the importance of the data points or the length of the time span. The formula is as follows:

[0057] Among them, w i x is the weighting coefficient. i For each data point, n represents the number of time points.

[0058] Furthermore, outliers with extreme quantities of goods exceeding the normal range are not calculated and are directly deleted.

[0059] Step 3: Construct seasonal data from the preprocessed dataset.

[0060] In step 3, a typical monthly time series is decomposed into seasonal data as follows:

[0061] M i =T i +S i +U i

[0062] Where M i Represents time series data, T i Indicates trend term, S i Indicates seasonal items, U i This indicates the irregular term, and W represents the weighting coefficient of the trend term.

[0063] S i +U i =M i -WT i

[0064] S i +U i The estimate is S i This data serves as seasonal data. Seasonal factors can influence warehouse inbound and outbound volumes, which may increase during holiday seasons or promotional events. Extracting useful temporal features from historical inventory data and adding seasonal data to the dataset makes it more closely reflect reality.

[0065] Step 4: Introduce a bidirectional LSTM model. The input to the bidirectional LSTM model is the historical feature sequence, and the output is the predicted warehouse arrival quantity.

[0066] The input sequence of a bidirectional LSTM model is fed into two independent LSTM layers: a forward LSTM and a backward LSTM. The forward LSTM processes the sequence chronologically from front to back, while the backward LSTM processes it from back to front. Each LSTM layer has its own hidden state and memory unit, and their outputs are combined in the final layer. The forward LSTM learns patterns from the past to the future, while the backward LSTM learns patterns from the future to the past. By combining information from both directions, the bidirectional LSTM model can simultaneously capture historical and future contextual information, better understanding dependencies in the sequence. By integrating information from both directions for prediction, the accuracy of predictions is improved, leading to better predictions of future states or generating more accurate sequence predictions.

[0067] Step 5: Construct an improved Long Short-Term Memory (CTA-LSTM) prediction model based on an attention mechanism using a deep learning framework to predict warehouse incoming quantities. Calculate attention weights so the model can focus on the relative distance or offset between different time steps. The relative positions between different time steps focus on the periodicity of time, such as the week, month, or even the relative dates of different years.

[0068] In attention mechanisms, the calculation of attention weights allows the predictive model to focus on the relative distances or offsets between different time steps. The relative positions between different time steps emphasize the periodicity of time, such as the weekday, month, or even the relative dates of different years. This increases the weight of data points that are Mondays or the same day of the same month. Let X1, X2…X… k Represents the input sequence; h1, h2…h k This represents the state value of the hidden layer corresponding to the input sequence; X represents i With the corresponding hidden layer h i Similarity, α i h represents the contribution of the hidden layer states of the historical input to the current input. i ′ represents the hidden state value of the last node in the final output.

[0069] The formula for calculating attention weights is as follows:

[0070] Among them, S ji Fitting using an RNN network:

[0071] S ji =γtanh(WX j +Uh i +b)

[0072] U represents the hidden layer state value h output at time i. iThe weight, W represents the weight of sample X. j The corresponding weights, where b is a user-defined constant. Y represents the weights applied to S during the softmax process. ji The weight.

[0073] Finally, the contribution values ​​are weighted and summed to obtain the feature vector h. i ′ is the hidden vector of the last node.

[0074] Introducing a position encoder into the attention mechanism adds positional information to the time steps of the input sequence, highlighting key influencing factors and thus helping the model make more accurate judgments. The tangent function is used to encode information at different positions. The output of the position encoder has the same dimension as the input sequence and can be directly added to it.

[0075] Step 6: Initialize the parameters of the attention-based improved long short-term memory network (CTA-LSTM) prediction model using the Grey Wolf algorithm, and then fine-tune the parameters to determine the optimal parameter configuration. Combining the Grey Wolf algorithm with bidirectional LSTM leverages its search capability during training to optimize the LSTM parameters and weights, finding the optimal solution or near-optimal solution more quickly. This reduces training time and resource consumption, improving the performance and effectiveness of the bidirectional LSTM model in sequence data modeling and prediction tasks.

[0076] Gray wolves employ a pyramidal social hierarchy. At the top are the alpha wolves, the leaders of the pack; the second level consists of beta wolves, their advisors; the third level comprises delta wolves, who follow the decisions and commands of the alpha and beta wolves; and at the bottom are omega wolves, responsible for maintaining balance within the pack. This hierarchical structure plays a crucial role in group hunting. Alpha wolves dominate the hunting process, playing a leading role in three main areas: tracking prey, pursuing prey until it stops moving, and attacking the prey. In the gray wolf algorithm, the first three optimal solutions are defined as α, β, and δ, guiding other wolves towards the optimal target. The remaining solutions are defined as ω, updating their positions according to α, β, or δ.

[0077] The refinement process in step 6 includes the following steps.

[0078] Step 6-1: Track the prey.

[0079] The formulas for updating the distance between an individual and its prey and the position of the gray wolf are as follows:

[0080] In the formula: Indicates the distance between an individual and its prey. Represents the coefficient vector. t represents the position of the gray wolf, and t represents the number of iterations. Represents the prey's position vector. A random number between [0, 1] The convergence factor decreases linearly from 2 to 0 with the number of iterations.

[0081] Step 6-2: Hunting the prey.

[0082] The mathematical model for the hunting process is shown in the following formula. Formula 1 represents the mathematical model describing the position of an individual gray wolf following its prey. Formula 2 defines the position of individual ω in the wolf pack as it moves towards a higher-ranking wolf. Formula 3 defines the final position of wolf ω.

[0083] In the formula Let α, B, δ be the distances to other individuals. The positions of α, B, and δ are given. It is a random vector. This indicates the current position of the Grey Wolves.

[0084] From the fourth formula in step 6-1, we can obtain: A decrease in value will cause The fluctuation of value, i.e. It is a random vector that varies within the interval [-α, α]. When the value is within the range, the gray wolf's next position can be anywhere between its current position and the prey's position. for The absolute value of.

[0085] when The gray wolf attack indicates that the system is trapped in a local optimum.

[0086] when The gray wolves then spread out to search, expanding their search area.

[0087] By optimizing the Grey Wolf algorithm, the optimal solution or near-optimal solution can be found more quickly, thereby reducing training time and resource consumption. This can improve the performance and effectiveness of bidirectional LSTM models in sequence data modeling and prediction tasks.

[0088] Step 7: Employ an adversarial mechanism to improve and adversarially train the attention-based improved long short-term memory network (CTA-LSTM) prediction model built in the previous steps in order to predict the quantity of goods in the warehouse.

[0089] An adversarial bidirectional LSTM is employed. Given different attention mechanisms and even different training parameters, two adversarial network models are trained and validated using a validation set. The final result is then derived by combining the two predictions. The method leverages the competition between the two models: one model attempts to better predict the quantity of goods, while the other model tries to generate adversarial examples to interfere with the former's predictions, thereby improving the robustness of the former.

[0090] Two LSTM models, LSTM model A and LSTM model B, are trained. Both models have the same architecture and parameter initialization. In each training iteration, LSTM model A is trained first, aiming to improve its performance on the cargo quantity prediction task. Then, LSTM model B is used as an adversarial example generator to generate adversarial examples against LSTM model A. These adversarial examples are samples with minor modifications to the original cargo quantity sequence, used to interfere with model A's predictions. Next, LSTM model B is trained using the generated adversarial examples and their corresponding labels (e.g., the original cargo quantities), aiming to enable it to generate more challenging adversarial examples. Adversarial training is then repeated, allowing the two LSTM models A and B to compete and gradually improve their ability to predict and generate adversarial examples. After adversarial training is complete, the predictive performance of LSTM model A is evaluated using test set data, and the degree to which the adversarial examples generated by LSTM model B interfere with LSTM model A is assessed.

[0091] Step 8: Establish a warehouse simulation model.

[0092] An objective function is established to evaluate the efficiency and performance of the warehousing system. This objective function is based on indicators such as the resource utilization rate, throughput, and operating costs of the warehousing system. Based on this objective function, the resource variables required to predict the volume of goods are calculated. Each resource variable includes type, quantity, etc.

[0093] This invention considers the time and economic costs of warehousing and transportation resources when establishing a warehousing simulation model. It ensures the minimum overall warehousing and transportation cost while meeting time constraints. minf1(x) represents the objective function for minimizing economic cost, and minf2(x) represents the objective function for minimizing time cost. The constructed warehousing simulation model is as follows:

[0094] In the formula: i = 1, 2, 3…nj = 1, 2, 3…m,

[0095] j represents the type of goods being transported, Y j Let x be a variable representing warehousing and transportation resources. ij Let Y be the variable representing warehousing and transportation resources. i The carrying rate, A j For the quantity of goods j that need to be transported, B ijC is the unit transportation cost of warehousing and transportation resources for transporting goods. ij For warehousing and transportation resources Y j The idle cost of transporting goods of class j. j V ij For Y i Total transport capacity under full load conditions, Y j For the rate of transporting goods of class j, t ij The time required for warehousing and transportation resource Yi to complete the transportation of goods of type j, Y ij =1 indicates selection Y i Transport will proceed, otherwise Y i =0. Y j k is the material coefficient for the h-th type of goods; ij For goods of category j, the storage and transportation resources Y i Transportation volume time cost coefficient.

[0096] Equations 1 and 2 are the objective functions, where Equation 1 is the economic cost function, including transportation costs and idle costs, and is an integer programming model from 0 to 1; Equation 2 is the cargo volume transportation time cost function, where transportation volume time is expressed as the cargo volume multiplied by the transportation time, and then multiplied by the cargo material coefficient; Equation 3 indicates that the total transportation volume of the selected transportation resources should not exceed the full-load transportation capacity; Equation 4 indicates that the total volume of transported goods should not be less than the total volume of goods received from the warehouse; Equation 5 indicates whether to select Y. i Transportation resources are used to transport goods of class j. If Y is selected... ij If Y is not selected, the value is 1. ij It is 0.

[0097] Step 9: Solve the warehouse simulation model constructed in Step 8.

[0098] The warehouse simulation model was solved using linear programming.

[0099] This embodiment constructs a multi-objective programming model. Objective 1 is the function that minimizes the economic cost of freight transportation, and objective 2 is the function that minimizes the time cost of freight transportation. In the process of solving multi-objective optimization problems, an absolutely optimal solution cannot be satisfied. While considering multiple objective functions, it is necessary to ensure that the values ​​of all objective functions are acceptable. A linear weighting method is used to process the multi-objective problem, assigning weight coefficients w1 and w2 to the multi-objective programming problem according to their importance, where w... i ≥0, i=1,2,3,...p, and Then use

[0100] The problem is then solved again using this as the new objective function. There are a total of p weight coefficients, where p is the objective function f. iThe number of (x) is 2 in this invention.

[0101] Obtain the optimal solution x (i) x * =x (i) As a solution to a multi-objective programming problem.

[0102] When solving the warehouse resource scheduling model, weights w1 and w2 are assigned to the two objective functions, where 0 ≤ w1 ≤ 1, 0 ≤ w2 ≤ 1, and w1 + w2 = 1. The new objective function is F(x) = w1·minf1(x) + w2·minf2(x). To determine the weight coefficients, the original weight values ​​are first obtained, and the average weight is calculated. The offset ω is then calculated using the original weights and the average weight. * Then, through the formula, we get

[0103] The weight coefficients ω1 and ω2 are calculated to be 0.4 and 0.6 respectively, so F(X) = 0.4minf1(x) + 0.6minf2(x).

[0104] Step 10: Based on the optimal solution obtained in Step 9, schedule transportation resources.

[0105] The optimal solution obtained in step 9 should be adjusted appropriately to ensure its operability in practical applications. It is necessary to ensure that transportation vehicles, manpower, equipment, etc., are available in actual scheduling and do not exceed their limits.

[0106] In practical scheduling, besides obtaining the optimal solution, the following aspects need attention: Ensuring the optimal solution is feasible and satisfies all constraints. Sometimes, numerical solutions may be obtained during calculation, but these solutions may not be feasible in practice. Therefore, feasibility verification must be performed before applying the optimal solution. Conduct sensitivity analysis on the optimal solution to understand its robustness under parameter changes. If changes in certain parameters or conditions cause significant changes in the optimal solution, it is necessary to consider how to handle these changes. Consider the feasibility and operability of the optimal solution in actual operation. Sometimes, the optimal solution may require making some impractical or difficult-to-implement decisions, so it needs to be reasonably adjusted to ensure its operability in practical applications. The optimal solution may involve the use of specific resources, such as transportation vehicles, manpower, and equipment. It is necessary to ensure that these resources are available in actual scheduling and do not exceed their limits. In practical scheduling, timeliness and real-time performance are often important factors. Sometimes the optimal solution is derived based on data and conditions for a specific time period, and in practical applications, it needs to be dynamically adjusted according to real-time data and conditions. Consider potential risks and uncertainties. The optimal solution may be derived based on assumptions and predictions, and may face uncertainties in practical applications. Risk assessment is necessary, and measures should be taken to mitigate potential risks. In actual scheduling, human factors must also be considered, such as employee work efficiency, traffic conditions, and customer demand. These factors may affect the actual execution effect of the optimal solution. Continuous monitoring and adjustments are also required during the scheduling process to address potential changes and unforeseen circumstances.

[0107] Figure 2 illustrates the principle of an embodiment of the warehousing and transportation resource scheduling system based on cargo quantity prediction of the present invention. Referring to Figure 2, the system of this embodiment includes: a data collection module, a dataset preprocessing module, a seasonal data construction module, a bidirectional LSTM model module, an improved long short-term memory network prediction model construction module based on attention mechanism, a gray wolf parameter tuning module, an adversarial training module, a warehousing simulation model establishment module, a warehousing simulation model solving module, and a transportation resource scheduling module.

[0108] The functions implemented by each module are as follows.

[0109] The data collection module collects data to form a dataset. The specific implementation of this module is the same as step 1 of the method embodiment shown in Figure 1, and will not be repeated here.

[0110] The dataset preprocessing module preprocesses the collected dataset to ensure data quality and accuracy. The specific implementation of this module is the same as step 2 of the method embodiment shown in Figure 1, and will not be repeated here.

[0111] The seasonal data construction module constructs seasonal data from the preprocessed dataset. The specific implementation of this module is the same as step 3 of the method embodiment shown in Figure 1, and will not be repeated here.

[0112] The bidirectional LSTM model module introduces a bidirectional LSTM model. The input of the bidirectional LSTM model is the historical feature sequence, and the output is the predicted warehouse receiving quantity. The specific implementation of this module is the same as step 4 of the method embodiment shown in Figure 1, and will not be repeated here.

[0113] The module for constructing an improved long short-term memory network prediction model based on an attention mechanism uses a deep learning framework to build such a model for predicting warehouse inventory quantities. The specific implementation of this module is the same as step 5 of the method embodiment shown in Figure 1, and will not be repeated here.

[0114] The Grey Wolf parameter tuning module uses the Grey Wolf algorithm to initialize the parameters in the improved long short-term memory network prediction model based on the attention mechanism, and then uses the Grey Wolf algorithm to fine-tune the parameters to determine the optimal parameter configuration in the algorithm. The specific implementation of this module is the same as step 6 of the method embodiment shown in Figure 1, and will not be repeated here.

[0115] The adversarial training module employs an adversarial mechanism to improve and adversarially train the attention-based improved long short-term memory network prediction model built in the previous steps, in order to predict the quantity of goods in the warehouse. The specific implementation of this module is the same as step 7 of the method embodiment shown in Figure 1, and will not be repeated here.

[0116] The warehouse simulation model building module establishes a warehouse simulation model. The specific implementation of this module is the same as step 8 of the method embodiment shown in Figure 1, and will not be repeated here.

[0117] The warehouse simulation model solving module solves the warehouse simulation model constructed by the warehouse simulation model building module. The specific implementation of this module is the same as step 9 of the method embodiment shown in Figure 1, and will not be repeated here.

[0118] The transportation resource scheduling module is based on the optimal solution obtained from the warehouse simulation model solution module. The specific implementation of this module is the same as step 10 of the method embodiment shown in Figure 1, and will not be repeated here.

[0119] Although the methods described above are illustrated and depicted as a series of actions for the sake of simplicity, it should be understood and appreciated that these methods are not limited by the order of the actions, as some actions may occur in a different order and / or concurrently with other actions from the illustrations and descriptions herein or not illustrated and described herein but which may be understood by those skilled in the art, according to one or more embodiments.

[0120] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps are described above in a generalized manner in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Those skilled in the art may implement the described functionality in different ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the invention.

[0121] The various illustrative logic blocks, modules, and circuits described in conjunction with the embodiments disclosed herein can be implemented or performed using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. The general-purpose processor may be a microprocessor, but in alternatives, it may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors cooperating with a DSP core, or any other such configuration.

[0122] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor such that the processor can read and write information to / from the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0123] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functionality may be stored or transmitted as one or more instructions or code on or through a computer-readable medium. A computer-readable medium includes both computer storage media and communication media, encompassing any medium that facilitates the transfer of a computer program from one location to another. A storage medium may be any available medium accessible to a computer. By way of example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and is accessible to a computer. Any connection is also legitimately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of a medium. As used in this article, disk and disc include compact discs (CDs), laser discs, optical discs, digital multi-purpose discs (DVDs), floppy disks, and Blu-ray discs. Disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of these should also be included within the scope of computer-readable media.

[0124] The prior description of this disclosure is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not intended to be limited to the examples and designs described herein, but should be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for scheduling warehousing and transportation resources based on cargo quantity prediction, characterized in that, The methods include: Step 1: Collect data to form a dataset; Step 2: Preprocess the collected dataset to ensure data quality and accuracy; Step 3: Construct seasonal data from the preprocessed dataset; Step 4: Introduce a bidirectional LSTM model. The input of the bidirectional LSTM model is the historical feature sequence, and the output is the predicted warehouse inbound quantity. Step 5: Use a deep learning framework to build an improved long short-term memory network prediction model based on the attention mechanism to predict the quantity of goods entering the warehouse; Step 6: Initialize the parameters in the improved long short-term memory network prediction model based on the attention mechanism using the gray wolf algorithm, and use the gray wolf algorithm to fine-tune the parameters to determine the optimal parameter configuration in the algorithm. Step 7: Employ an adversarial mechanism to improve and adversarially train the attention-based improved long short-term memory network prediction model built in the previous steps in order to predict the quantity of goods in the warehouse. Step 8: Establish a warehouse simulation model; Step 9: Solve the warehouse simulation model constructed in Step 8; Step 10: Schedule transportation resources based on the optimal solution obtained in Step 9.

2. The warehousing and transportation resource scheduling method based on cargo quantity prediction according to claim 1, characterized in that, In step 2, the preprocessing includes: The annual average method is used to fill in missing values, handle outliers, and delete duplicate values.

3. The warehousing and transportation resource scheduling method based on cargo quantity prediction according to claim 1, characterized in that, In step 5, by calculating attention weights, the long short-term memory network prediction model can focus on the relative distance or relative offset between different time steps; a position encoder is introduced into the attention mechanism to add position information to the time steps in the input sequence.

4. The warehousing and transportation resource scheduling method based on cargo quantity prediction according to claim 1, characterized in that, In step 6, the Grey Wolf algorithm is combined with bidirectional LSTM. During the training process, the search capability of the Grey Wolf algorithm is used to optimize the parameters and weights of the LSTM, so as to find the optimal solution or close to the optimal solution more quickly. Step 6 further includes the processing of tracking and hunting prey.

5. The warehousing and transportation resource scheduling method based on cargo quantity prediction according to claim 1, characterized in that, In step 7, an adversarial bidirectional LSTM is used. Given different attention mechanisms and even different training parameters, two adversarial network models are trained and validated using a validation set. The two prediction results are combined to give the final result. The competition between the two models is utilized, with one model trying to better predict the quantity of goods, while the other model tries to generate adversarial examples to interfere with the former's prediction, thereby improving the robustness of the former.

6. The warehousing and transportation resource scheduling method based on cargo quantity prediction according to claim 1, characterized in that, In step 8, an objective function is established to evaluate the efficiency and performance of the warehousing system. This objective function is based on indicators such as the resource utilization rate, throughput, and operating costs of the warehousing system. Based on this objective function, the resource variables required to predict the cargo volume are calculated. Each resource variable includes type and quantity. When establishing the warehousing simulation model, the time cost and economic cost of warehousing and transportation resources are taken into account, and the overall warehousing and transportation cost is minimized while meeting the time constraints.

7. The warehousing and transportation resource scheduling method based on cargo quantity prediction according to claim 1, characterized in that, In step 9, the warehousing simulation model is solved using linear programming. The model is based on a multi-objective programming model. Objective 1 is the function that minimizes the economic cost of cargo transportation, and objective 2 is the function that minimizes the time cost of cargo transportation. In the process of solving multi-objective optimization, an absolute optimal solution cannot be satisfied. While considering multiple objective functions, it is necessary to ensure that the values ​​of multiple objective functions are acceptable. Therefore, a linear weighting method is used to handle the multi-objective problem.

8. A warehousing and transportation resource scheduling system based on cargo quantity prediction, characterized in that, The system includes: The data collection module collects data to form a dataset; The dataset preprocessing module preprocesses the collected datasets to ensure data quality and accuracy. The seasonal data construction module constructs seasonal data from the preprocessed dataset; The bidirectional LSTM model module introduces a bidirectional LSTM model. The input of the bidirectional LSTM model is the historical feature sequence, and the output is the predicted warehouse purchase quantity. The module for building an improved long short-term memory network prediction model based on attention mechanism uses a deep learning framework to build an improved long short-term memory network prediction model based on attention mechanism for predicting the quantity of goods entering the warehouse. The Grey Wolf parameter tuning module uses the Grey Wolf algorithm to initialize the parameters in the improved long short-term memory network prediction model based on the attention mechanism, and uses the Grey Wolf algorithm to fine-tune the parameters to determine the optimal parameter configuration in the algorithm. The adversarial training module uses an adversarial mechanism to improve and adversarially train the improved long short-term memory network prediction model based on the attention mechanism built in the previous steps, so as to predict the quantity of goods in the warehouse. The warehouse simulation model creation module is used to create a warehouse simulation model. The warehouse simulation model solving module solves the warehouse simulation model constructed by the warehouse simulation model building module. The transportation resource scheduling module is based on the optimal solution obtained from the warehouse simulation model solution module.

Images