Logistics prediction device

Through feature screening and genetic optimization, the BP neural network solves the shortcomings of traditional prediction systems in capturing nonlinear relationships, improves the accuracy and robustness of logistics freight volume prediction, and is suitable for complex logistics data models.

CN119027005APending Publication Date: 2024-11-26GUILIN UNIV OF AEROSPACE TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410916400.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-09
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Traditional logistics freight volume prediction systems are difficult to accurately capture complex nonlinear relationships, resulting in low prediction accuracy and affecting regional logistics planning decisions.

Method used

A BP neural network based on feature screening and genetic optimization is adopted to screen influencing factors through XGBoost and optimize the initial weight value and threshold of the BP neural network using genetic algorithm. The network is trained with the gradient descent method of additive momentum terms to build an efficient logistics prediction model.

Benefits of technology

It improves the accuracy and robustness of logistics freight volume forecasting, adapts to different data modes and complex relationships, and achieves more efficient logistics scheduling and planning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027005A_ABST
    Figure CN119027005A_ABST
Patent Text Reader

Abstract

The invention discloses a logistics prediction device. In order to improve the accuracy of traditional logistics freight prediction, the XGBoost is adopted to perform efficient feature screening, and the GA is combined to optimize the weight value and the threshold value of the BP neural network. According to the invention, the technical problem of low logistics prediction accuracy caused by highly complex non-linear and variable factors in the prior art is solved, and the logistics prediction technical effects of global adaptability, robustness and high precision are obtained. The method is suitable for the technical field of intelligent logistics.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of intelligent logistics, and in particular to a device for predicting regional logistics freight volume using a BP neural network based on feature screening and genetic optimization. Background Art

[0002] The application of artificial intelligence in logistics, supply chain and other fields is the trend of developing intelligent logistics technology in the current big data era. Existing technologies 1 to 4 are some representative applications in this field.

[0003] Prior art 1: CN110599077B, logistics prediction system and prediction method, Luojidi Co., Ltd., announcement date 2023.07.14;

[0004] Prior art 2: CN109934368B, method, device and system for predicting order quantity and order status, Cainiao Smart Logistics Holdings Co., Ltd., announcement date 2023.11.10;

[0005] Prior art 3: CN109829667B, a method and device for predicting the parcel volume at a logistics node, Cainiao Smart Logistics Holdings Co., Ltd., announcement date 2023.06.02;

[0006] Prior art 4: CN111191823B, a production logistics prediction method based on deep learning, Northwestern Polytechnical University, published on June 27, 2023.

[0007] The primary task in the early stages of regional logistics planning decisions is to forecast logistics freight volume, which can determine the development scale and capacity level of logistics facilities in the region, plan the overall layout of regional logistics, and divide the functions of regional logistics centers. It is one of the important bases for regional logistics planning.

[0008] Since regional logistics freight volume exhibits non-stationary characteristics and is affected by multiple factors, traditional forecasting systems often find it difficult to accurately capture these complex nonlinear relationships, resulting in low forecast accuracy and failure to achieve ideal forecasting results, which in turn affects subsequent decision-making and planning. Summary of the invention

[0009] In order to alleviate or partially alleviate the above technical problems, the solution of the present invention is as follows:

[0010] A logistics prediction device, the logistics prediction device includes a memory, a data reading and writing module, a preprocessing module, a feature screening module, a BP neural network module, and a network parameter adjustment module, and the network parameter adjustment module includes an initial network parameter optimization module and a training module; the data reading and writing module is used to read sample data from the memory; the preprocessing module is used to perform the following processing steps on the sample data: delete irrelevant data, duplicate data, smooth noise data, and process missing values ​​and abnormal values ​​to obtain operating data; the feature screening module is used to perform feature screening on the processed operating data; and, for the constructed BP neural network module, the initial weight value and threshold of the BP neural network are optimized through the initial network parameter optimization module in the network parameter adjustment module, and the optimized initial weight value and threshold are configured to the BP neural network, and the BP neural network in the BP neural network module is trained through the training module in the network parameter adjustment module to obtain a trained optimal BP neural network; in addition, based on the optimal BP neural network, the regional logistics freight volume is predicted and the prediction result is output.

[0011] Furthermore, the BP neural network module is constructed by at least the following steps: normalizing the network input data; setting the input layer, hidden layer, output layer and the corresponding number of nodes to construct the BP neural network module, and satisfying Where Q is the number of neurons in the hidden layer, m is the number of neurons in the input layer, n is the number of neurons in the output layer, the positive integer α is an empirical number, and 1≤α≤10; in addition, the following BP neural network parameters are set: learning rate, number of hidden layers, momentum factor and activation function.

[0012] Furthermore, the initial network parameter optimization module is configured to perform the following sub-steps: initializing the population size and parameters; initializing the weight value and threshold of the BP neural network; optimizing the weight value and threshold of the BP neural network; judging whether the weight value and threshold have reached the optimum, and if not, repeating the step of optimizing the weight value and threshold of the BP neural network until the optimal initial weight value and threshold of the BP neural network are obtained.

[0013] Furthermore, the optimal initial weight value and threshold of the BP neural network obtained by the initial network parameter optimization module are configured to the BP neural network, and the training module trains the BP neural network according to the gradient descent method with momentum term added to obtain the optimal BP neural network.

[0014] Furthermore, the sample data includes at least: regional gross domestic product, output value of the primary industry, output value of the secondary industry, output value of the tertiary industry, total retail sales of consumer goods, total imports and exports, fixed asset investment, permanent population, per capita disposable income of urban residents, per capita disposable income of rural residents, per capita consumption expenditure of urban residents, per capita consumption expenditure of rural residents, highway mileage, railway mileage and freight volume.

[0015] Furthermore, the initialization population size and parameters specifically include: setting the population size to 500, the maximum number of evolutionary iterations to 1000, the mutation probability to between 0.20 and 0.99, the crossover probability to between 0.0001 and 0.8, and setting the fitness function: Where Y is the predicted output value, Y 0 is the actual data value, N is the number of iterations, and u is the chromosome number.

[0016] Furthermore, the optimization of the weight value and threshold of the BP neural network specifically includes: continuously iterating and optimizing the weight value and threshold of the BP neural network according to the basic operations of selective replication, crossover, and mutation. Among them, selective replication is to select chromosomes with higher fitness from the population for replication to generate the next generation population.

[0017] Furthermore, the feature screening module performs feature screening on the operating data based on XGBoost.

[0018] Furthermore, XGBoost is used to perform feature screening on the operating data to identify influencing factors that are highly correlated with regional logistics freight volume, and to filter out irrelevant factors.

[0019] The technical solution of the present invention has one or more of the following beneficial technical effects:

[0020] (1) It can efficiently screen out influencing factors that are highly correlated with regional logistics freight volume, avoid interference from too many irrelevant factors, improve the accuracy of prediction, and achieve more efficient and intelligent logistics scheduling effects.

[0021] (2) Compared with the traditional linear model, it is closer to reality and more suitable for dealing with the nonlinear trend of regional logistics freight volume changes and the impact of various variable factors on accuracy, and can accurately predict regional logistics freight volume.

[0022] (3) Compared with the BP neural network that has not been optimized in advance, the present invention has greater global adaptability and can adapt to different data patterns and complex relationships. It is not only more accurate but also more robust.

[0023] (4) Combining the feature screening module and network parameter adjustment module of XGBoost to process data can improve the performance and efficiency of the BP neural network module.

[0024] In addition, other beneficial effects of the present invention will be mentioned in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a structural diagram of the logistics prediction device of the present invention;

[0026] Figure 2 It is the operation flow chart of the logistics prediction device of the present invention;

[0027] Figure 3 It is a schematic diagram of the relationship between feature screening and BP neural network;

[0028] Figure 4 It is a flow chart of optimizing BP neural network of the present invention;

[0029] Figure 5 It is an accurate performance test chart of the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments.

[0031] In the description of the present invention, unless otherwise specified, "plurality" refers to two or more than two. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items.

[0032] In order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, words such as "first" and "second" are used to distinguish the same or similar items with substantially the same functions and effects. Those skilled in the art can understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit the difference.

[0033] refer to Figure 1 , which shows the logistics prediction device disclosed in the present invention, including a memory, a data reading and writing module, a preprocessing module, a feature screening module, a BP neural network module, and a network parameter adjustment module.

[0034] Logistics forecasting devices, for example, can be applied to cloud computing centers, servers, various node centers, or terminal devices, and serve as the core component of the entire logistics system to guide logistics resource planning and scheduling.

[0035] The memory is configured as a physical recording device for storing sample data, output results and other information, including non-volatile memory and volatile memory. In some embodiments, the two can be used in combination to store the aforementioned sample data, output results and other information. The data reading and writing modules, such as obtaining sample data, outputting prediction results, temporarily storing temporary or intermediate information, etc., can all rely on the data reading and writing modules, namely the so-called I / O system to control.

[0036] The preprocessing module performs various preprocessing operations on the original collected and unprocessed data, but generally does not involve making decisive judgments or influencing the result data.

[0037] The feature screening module performs an automated screening process on the preprocessed data to facilitate the subsequent BP neural network module to perform the reasoning steps more effectively and accurately.

[0038] The BP neural network module realizes the reasoning of input data by constructing a BP neural network with a specific network topology.

[0039] The network parameter adjustment module includes an initial network parameter optimization module and a training module. The initial network parameter optimization module uses a genetic algorithm to optimize the initial weight value (weight) and threshold (threshold) of the BP neural network, while the training module is used to iteratively adjust various network configuration parameters in the BP neural network module to achieve the best network inference results.

[0040] The aforementioned logistics prediction device may also include other modules not shown, which meet the needs of various industrial applications, such as a display module, a data acquisition module, a wireless or limited communication module,

[0041] The specific implementation of the above modules will be described in detail below.

[0042] refer to Figure 2 , which shows the specific operation of the aforementioned logistics prediction device. The operation between these components mainly consists of the following steps:

[0043] Step 1: Get sample data.

[0044] The sample data is read from the memory by reading and writing data. The sample data can be collected and acquired through various software and / or hardware tools such as wired network, wireless network, crawler, etc.

[0045] For example, the sample data includes at least: regional GDP, output value of the primary industry, output value of the secondary industry, output value of the tertiary industry, total retail sales of consumer goods, total imports and exports, fixed asset investment, permanent population, per capita disposable income of urban residents, per capita disposable income of rural residents, per capita consumption expenditure of urban residents, per capita consumption expenditure of rural residents, highway mileage, railway mileage, and freight volume.

[0046] Step 2: Preprocess the sample data to obtain running data, and perform feature screening on the preprocessed running data.

[0047] Preferably, the preprocessing includes: deleting irrelevant data, duplicate data, smoothing noisy data, and processing missing values ​​and outliers.

[0048] Then, feature screening is performed on the preprocessed running data based on XGBoost. Specifically, the feature importance score can be calculated first, and then the features can be sorted from high to low according to the importance score.

[0049] Figure 3 It is a schematic diagram of the relationship between feature screening and BP neural network. For example, the sample data includes various types of data such as regional GDP, primary industry value, secondary industry value, ..., railway mileage, etc. After extracting feature data through XGBoost, feature data such as regional GDP, total retail sales of consumer goods, ..., railway mileage, etc. are screened out, and these screened feature data are used as the input of the BP network.

[0050] Furthermore, the present invention also optimizes the initial weights (weight values) and thresholds (threshold values) of the neural network through a genetic algorithm.

[0051] Step 3: Construct the BP neural network topology.

[0052] This step specifically includes the following sub-steps:

[0053] S301, normalization processing of network input data: performing normalization processing on the data after feature screening in step 2:

[0054]

[0055] Where: x max is the maximum value of the sample, x min is the minimum value of the sample, and x is the data after feature filtering.

[0056] S302, setting the input layer, hidden layer, output layer and the corresponding number of nodes to construct a BP neural network module.

[0057] According to the data filtered out in step 2, for example, five variables such as regional GDP, total retail sales of consumer goods, total imports and exports, permanent population, and railway mileage are used as the input layer, and freight volume is used as the output layer.

[0058] And according to the formula Set the number of nodes, where Q is the number of neurons in the hidden layer, m is the number of neurons in the input layer, n is the number of neurons in the output layer, and the positive integer α is an empirical number (1≤α≤10). At the same time, set other network configuration parameters of the BP neural network, for example, these parameters include learning rate, number of hidden layers, momentum factor, and activation function.

[0059] Step 4: Optimize the initial weight value and threshold of the BP neural network through the initial network parameter optimization module, and configure the optimal initial weight value and threshold to the BP neural network. After training the BP neural network through the training module, the trained optimal BP neural network is obtained.

[0060] refer to Figure 4 , which shows the flowchart of the present invention for optimizing the BP neural network. Specifically, the initial network parameter optimization module performs the following sub-steps:

[0061] Sub-step S401: Initialize population size and parameters: For example, set the population size to 500, the maximum number of evolutionary iterations to 1000, the mutation probability to between 0.20 and 0.99, the crossover probability to between 0.0001 and 0.8, and set the fitness function: Where Y is the predicted output value, Y 0 is the actual data value, N is the number of iterations, and u is the chromosome number.

[0062] Sub-step S402: Initialize the weight value and threshold value of the BP neural network: Encode the weight value and threshold value representing the initial BP neural network through binary numbers.

[0063] Furthermore, after the weight values ​​and thresholds of the BP neural network are initialized and these initial values ​​are binary-coded, the error obtained by training the BP neural network is used as the fitness value, and then the process proceeds to sub-step S403.

[0064] Sub-step S403: Optimizing the weight value and threshold value of the BP neural network: According to the basic operations of copy selection, crossover, and mutation, the weight value and threshold value of the BP neural network are continuously optimized iteratively.

[0065] Sub-step S404: Determine whether the weight value and threshold value are optimal. If not, repeat sub-step S403 until the optimal initial weight value and threshold value of the BP neural network are obtained.

[0066] For example, when the fitness value satisfies the termination condition, it is considered that the optimal initial weight value and threshold of the BP neural network are obtained.

[0067] Then, execute sub-step S405:

[0068] The optimal initial weight value and threshold of the BP neural network obtained by the initial network parameter optimization module are configured to the BP neural network. The training module trains the BP neural network according to the gradient descent method with momentum term added to obtain the optimal BP neural network.

[0069] Furthermore, in this step, the weight value and threshold value of the BP neural network are updated by calculating the reasoning error of the BP neural network and using the gradient descent method. When the end condition is met, the training of the weight value and threshold value of the BP neural network is completed to obtain the optimal BP neural network.

[0070] Step 5: Use the optimal BP neural network to predict the regional logistics freight volume and output the prediction results.

[0071] Furthermore, the prediction results output by the BP neural network are written into the memory through the aforementioned data reading and writing module. Preferably, the prediction results stored in the memory are also transmitted to other modules or devices, such as cloud computing centers, servers, and various node centers.

[0072] For example, the present invention uses mean absolute error (MAE), mean square error (MSE), mean relative error (MAPE), root mean square relative error (RMSE) and coefficient of determination (R 2 ) These five error analysis methods analyze and evaluate the accuracy of the prediction model to verify the accuracy of the prediction model. These error analysis formulas are:

[0073]

[0074] Where X is the number of samples, y is i , and are the arithmetic means of the actual value, predicted value and true value of sample i respectively.

[0075] Figure 5 It is the result of the present invention's solution performing feature screening on the running data after preprocessing the data and then testing it. Through analysis, it is found that the present invention has faster training speed and higher accuracy.

[0076] Table 1 shows the mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), mean absolute percentage error (MAPE) and determination coefficient (R 2) Quantitative indicators for evaluating the quality of standard solutions.

[0077] Table 1 Evaluation criteria of test models under different conditions

[0078]

[0079] It can be seen from Table 1 that the solution proposed in the present invention can well predict the actual freight volume based on the relevant data of regional logistics freight volume. The experimental results verify that the present invention has high accuracy and practical value.

[0080] In the existing technology, it is generally believed that XGBoost, GA (genetic algorithm) and BP neural network have their own parameters and configuration requirements. These technologies will significantly increase the complexity of the system and require precise tuning and highly specialized knowledge and a lot of programming and debugging work. Not only that, each technology has its own parameters and optimization process. How to effectively integrate these models and ensure overall performance is also a huge challenge.

[0081] Due to the overall complexity of the system, the high consumption of computing resources, and the time-consuming and labor-intensive difficulty of tuning multiple technologies at the same time, current research mainly focuses on the optimization and application of a single technology or simple combination (such as GA-BP, XGBoost-BP).

[0082] However, the technical solution of the present invention creatively combines XGBoost, GA and BP neural networks. Compared with the traditional solution of only using GA-BP, XGBoost-BP or BP, it achieves high accuracy and low error, and can more accurately predict the technical effect of regional logistics freight volume.

[0083] Although the present invention has been described in extensive detail, one skilled in the art will appreciate that the present invention may be practiced without some of the specific details.

[0084] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A logistics prediction device, characterized in that: The logistics prediction device includes a memory, a data reading and writing module, a preprocessing module, a feature screening module, a BP neural network module, and a network parameter adjustment module, and the network parameter adjustment module includes an initial network parameter optimization module and a training module; The data reading and writing module is used to read sample data from the memory; The preprocessing module is used to perform the following processing steps on the sample data: delete irrelevant data, duplicate data, smooth noise data, and process missing values ​​and abnormal values ​​to obtain operation data; The feature screening module is used to perform feature screening on the processed operation data, specifically, to perform feature screening on the operation data according to XGBoost; The initial network parameter optimization module is configured to perform the following sub-steps: Initialize population size and parameters; Initialize the weight value and threshold of the BP neural network; Optimizing the weight value and threshold of the BP neural network, specifically including: continuously iteratively optimizing the weight value and threshold of the BP neural network according to the basic operations of selecting replication, crossover, and mutation; Determine whether the weight value and threshold value are optimal. If not, repeat the steps of optimizing the weight value and threshold value of the BP neural network until the optimal initial weight value and threshold value of the BP neural network are obtained. For the constructed BP neural network module, the initial weight value and threshold value of the BP neural network are optimized by the initial network parameter optimization module in the network parameter adjustment module, and the optimal initial weight value and threshold value are configured to the BP neural network, and the BP neural network in the BP neural network module is trained by the training module in the network parameter adjustment module to obtain the trained optimal BP neural network; and, Based on the optimal BP neural network, the regional logistics freight volume is predicted and the prediction result is output.

2. The logistics prediction device according to claim 1, characterized in that: The BP neural network module is constructed by at least the following steps: Normalization of network input data; Set the input layer, hidden layer, output layer and the corresponding number of nodes to build a BP neural network module and satisfy Where Q is the number of neurons in the hidden layer, m is the number of neurons in the input layer, n is the number of neurons in the output layer, and the positive integer α is an empirical number, and 1≤α≤10; In addition, the following BP neural network parameters are set: learning rate, number of hidden layers, momentum factor and activation function.

3. The logistics prediction device according to claim 2, characterized in that: The optimal initial weight value and threshold of the BP neural network obtained by the initial network parameter optimization module are configured to the BP neural network. The training module trains the BP neural network according to the gradient descent method with momentum term added to obtain the optimal BP neural network.

4. The logistics prediction device according to claim 3, characterized in that: The initialization population size and parameters specifically include: Set the population size to 500, the maximum number of evolutionary iterations to 1000, the mutation probability to between 0.20 and 0.99, the crossover probability to between 0.0001 and 0.8, and set the fitness function: Where Y is the predicted output value, Y0 is the actual data value, N is the number of iterations, and u is the chromosome number.

5. The logistics prediction device according to claim 4, characterized in that: The sample data include at least: regional GDP, output value of the primary industry, output value of the secondary industry, output value of the tertiary industry, total retail sales of consumer goods, total import and export volume, fixed asset investment, permanent population, per capita disposable income of urban residents, per capita disposable income of rural residents, per capita consumption expenditure of urban residents, per capita consumption expenditure of rural residents, highway mileage, railway mileage and freight volume.

6. The logistics prediction device according to claim 5, characterized in that: XGBoost is used to perform feature screening on the operating data, filter out the influencing factors that are highly correlated with the regional logistics freight volume, and filter out irrelevant factors.