Order trajectory prediction method and device, equipment and storage medium

Through a hybrid model combining deep learning and recurrent neural network, the spatial and temporal characteristics of logistics trajectories are extracted, and the problem of insufficient logistics trajectory prediction accuracy in the existing technology is solved, achieving accurate prediction of logistics information and user experience improvement.

CN119067544APending Publication Date: 2024-12-03SHANGHAI DONGPU INFORMATION TECH CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411114843.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

When faced with emergencies, the existing logistics trajectory prediction methods are insufficient in the accuracy and cannot accurately predict the logistics trajectory, which affects the user experience and the credibility of the e-commerce platform.

Method used

Using a hybrid model based on a combination of deep learning and recurrent neural networks, spatial features are extracted through the CNN module, the RNN module captures time series features, and introduces an attention mechanism to allocate weights according to different features to accurately predict logistics trajectories.

Benefits of technology

It realizes accurate prediction of order trajectory, provides clearer and more accurate logistics information, improves user experience, and can respond to transportation delays and abnormal situations in a timely manner.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119067544A_ABST
    Figure CN119067544A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of logistics transportation, and discloses an order trajectory prediction method and device, equipment and a storage medium. The method comprises the following steps: collecting logistics track historical data; cleaning, screening and standardizing the logistics track historical data to obtain preprocessed logistics track historical data; constructing a hybrid model based on the combination of deep learning and a recurrent neural network; training the hybrid model by adopting preprocessed logistics track historical data to obtain a track prediction model; and inputting to-be-predicted preprocessed logistics trajectory data into the trajectory prediction model, and outputting a predicted order logistics trajectory. According to the invention, trajectory prediction is carried out by using the hybrid model based on the combination of deep learning and the recurrent neural network, an attention mechanism is introduced into the model, different weights are allocated according to the importance of different features, accurate prediction of order trajectories can be realized, and the prediction efficiency is improved. Therefore, the accurately predicted order trajectory result can be conveniently output to the e-commerce platform and the user, and the user experience is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of logistics and transportation technology, and in particular to an order trajectory prediction method, device, equipment and storage medium. Background Art

[0002] As an innovative function in the logistics service system of e-commerce platforms, the core value of "Mr. Check" is not limited to simply providing a time forecast of logistics track updates, but also to building an efficient and accurate logistics information interaction platform. This function can quickly and accurately predict and notify e-commerce platforms and users of the specific time point of the next logistics track update by real-time monitoring of each link in the logistics network, including sorting, transportation, and distribution, greatly improving the timeliness and predictability of logistics information.

[0003] However, faced with the ever-increasing volume of logistics and the complex and ever-changing transportation environment, the existing trajectory prediction methods that Miaomiaocha relies on have gradually exposed their limitations of insufficient accuracy. Traditional prediction models are often based on historical data for analysis, but their prediction capabilities are insufficient when faced with unexpected situations such as package loss, damage, bad weather affecting transportation speed, and other unforeseen factors. These unexpected events not only disrupt the original logistics plan, but may also cause the platform to be unable to inform users of logistics dynamics in a timely and accurate manner, thereby affecting user experience and the reputation of the e-commerce platform.

[0004] Therefore, the prior art still needs to be improved and developed. Summary of the invention

[0005] The main purpose of the present invention is to solve the problem that the existing technology is difficult to fully capture and respond to the dynamic change factors in the logistics network, resulting in the inability to accurately predict the logistics trajectory.

[0006] A first aspect of the present invention provides an order trajectory prediction method, comprising:

[0007] Collecting historical data of logistics trajectories, wherein the historical data of logistics trajectories include the starting position of the package, transportation route, transportation tools, weather conditions, traffic flow, processing efficiency of the distribution center, pickup time and receipt time; cleaning, screening and standardizing the historical data of logistics trajectories to obtain pre-processed historical data of logistics trajectories; constructing a hybrid model based on the combination of deep learning and recurrent neural network, wherein the hybrid model includes a CNN module for extracting spatial features, an RNN module for extracting spatial features, and an attention mechanism module for assigning weights to different features; training the hybrid model with the pre-processed historical data of logistics trajectories to obtain a trajectory prediction model; inputting the pre-processed logistics trajectory data to be predicted into the trajectory prediction model, and outputting a predicted order logistics trajectory.

[0008] Optionally, in a first implementation method of the first aspect of the present invention, the step of collecting logistics trajectory historical data includes: retrieving the starting location, transportation route, transportation tool, pickup time and receipt time of the package from the internal system of the logistics company through an API interface; retrieving the weather conditions of the corresponding location and time from the weather service system according to the transportation route, pickup time and receipt time, and retrieving the traffic flow of the corresponding location and time from the traffic data system; retrieving the time and throughput of package processing by each distribution center from the internal system of the logistics company through an API interface to obtain the processing efficiency of the distribution center.

[0009] Optionally, in a second implementation method of the first aspect of the present invention, the steps of cleaning, screening and standardizing the logistics trajectory historical data to obtain pre-processed logistics trajectory historical data include: using statistical methods to identify and process outliers in the logistics trajectory historical data to obtain screened data; deleting or merging duplicate record data in the screened data to ensure the uniqueness of the data set to obtain merged data; encoding the merged data and converting it into a unified format to obtain pre-processed logistics trajectory historical data.

[0010] Optionally, in a third implementation of the first aspect of the present invention, a hybrid model based on the combination of deep learning and recurrent neural network is constructed, including the steps of: constructing a hybrid model using a CNN module, an RNN module and an attention mechanism module; the composition structure of the CNN module includes: the input end is connected to the first convolutional layer, the first convolutional layer includes 64 convolutional kernels, each convolutional kernel size is 3*3, the step size is 2, the activation function is Relu, and the output end of the first convolutional layer is connected to the input end of the second convolutional layer; the second convolutional layer includes 128 convolutional kernels, each convolutional kernel size is 3*3, the step size is 2, the activation function is Relu, and the output end of the second convolutional layer is connected to the input end of the third convolutional layer; the third convolutional layer includes four branches, the output ends of the four branches are connected, and then connected to the input end of the fourth convolutional layer end; the fourth convolution layer includes 256 convolution kernels, each convolution kernel size is 3*3, the step size is (2,1), the activation function is Relu, and the output end of the fourth convolution layer is connected to the input end of the fifth convolution layer; the fifth convolution layer includes 512 convolution kernels, each convolution kernel size is 3*3, the step size is (2,1), the activation function is Relu, and the output end of the fifth convolution layer is connected to the input end of the sixth convolution layer; the sixth convolution layer includes 512 convolution kernels, each convolution kernel size is 2*1, the step size is 1, the activation function is Relu, and the output end of the sixth convolution layer is connected to the input end of the seventh convolution layer; the seventh convolution layer includes a Squeeze module, and the Squeeze module performs a Squeeze operation on the input features, removes the first dimension, and obtains the output features of the CNN module.

[0011] Optionally, in a fourth implementation method of the first aspect of the present invention, the four branches of the third convolutional layer are: the first branch is a convolution branch, including 128 convolution kernels, each convolution kernel size is 1*1, and the activation function is Relu; the second branch is a depth-separable convolution branch, including 128 convolution kernels, each convolution kernel size is 3*3, the step size is 1, the dilation rate is 1, and the activation function is Relu; the third branch is a depth-separable convolution branch, including 128 convolution kernels, each convolution kernel size is 3*3, the step size is 1, the dilation rate is 3, and the activation function is Relu; the fourth branch is a depth-separable convolution branch, including 128 convolution kernels, each convolution kernel size is 3*3, the step size is 1, the dilation rate is 5, and the activation function is Relu.

[0012] Optionally, in a fifth implementation method of the first aspect of the present invention, the pre-processed logistics trajectory historical data is used to train the hybrid model, and the step of obtaining the trajectory prediction model includes: dividing the pre-processed logistics trajectory historical data into a training set and a validation set in a ratio of 7:3; setting training parameters, and using the training set to train the hybrid model. During the training process, the weights and biases of the hybrid model are updated through a stochastic gradient descent algorithm to minimize the loss function; using the validation set to verify the hybrid model during the training process, and adjusting the structure or training parameters of the hybrid model according to the verification results to obtain the trajectory prediction model.

[0013] Optionally, in a sixth implementation method of the first aspect of the present invention, the pre-processed logistics trajectory data to be predicted is input into the trajectory prediction model, and a predicted order logistics trajectory is output, including the steps of: inputting the pre-processed logistics trajectory data to be predicted into the trajectory prediction model to obtain predicted path planning position points and their corresponding timestamps; connecting the path planning position points according to the timestamps and performing smoothing to output the predicted order logistics trajectory.

[0014] The second aspect of the present invention provides an order trajectory prediction device, including: a collection module, used to collect logistics trajectory historical data, the logistics trajectory historical data includes the starting position of the package, transportation route, transportation tools, weather conditions, traffic flow, distribution center processing efficiency, pickup time and receipt time; a processing module, used to clean, screen and standardize the logistics trajectory historical data to obtain pre-processed logistics trajectory historical data; a model building module, used to build a hybrid model based on deep learning and recurrent neural network, the hybrid model includes a CNN module for extracting spatial features, an RNN module for extracting spatial features, and an attention mechanism module for assigning weights to different features; a training module, used to train the hybrid model using the pre-processed logistics trajectory historical data to obtain a trajectory prediction model; a prediction module, used to input the pre-processed logistics trajectory data to be predicted into the trajectory prediction model, and output a predicted order logistics trajectory.

[0015] Optionally, in a first implementation method of the second aspect of the present invention, the collection module includes: a first retrieval unit, used to retrieve the starting location, transportation route, transportation tool, pickup time and receipt time of the package from the internal system of the logistics company through an API interface; a second retrieval unit, used to retrieve the weather conditions of the corresponding location and time from the weather service system according to the transportation route, pickup time and receipt time, and retrieve the traffic flow of the corresponding location and time from the traffic data system; a third retrieval unit, used to retrieve the time and throughput of package processing by each distribution center from the internal system of the logistics company through an API interface to obtain the processing efficiency of the distribution center.

[0016] Optionally, in a second implementation method of the second aspect of the present invention, the processing module includes: a screening unit, used to identify and process abnormal values ​​in the logistics trajectory historical data using statistical methods to obtain screened data; a merging unit, used to delete or merge duplicate record data in the screened data to ensure the uniqueness of the data set to obtain merged data; a conversion unit, used to encode the merged data and convert it into a unified format to obtain pre-processed logistics trajectory historical data.

[0017] Optionally, in a third implementation of the second aspect of the present invention, the model construction module includes: a construction unit, which uses a CNN module, an RNN module and an attention mechanism module to construct a hybrid model; the composition structure of the CNN module includes: the input end is connected to the first convolutional layer, the first convolutional layer includes 64 convolutional kernels, each convolutional kernel size is 3*3, the step size is 2, the activation function is Relu, and the output end of the first convolutional layer is connected to the input end of the second convolutional layer; the second convolutional layer includes 128 convolutional kernels, each convolutional kernel size is 3*3, the step size is 2, the activation function is Relu, and the output end of the second convolutional layer is connected to the input end of the third convolutional layer; the third convolutional layer includes four branches, the output ends of the four branches are connected, and then connected to the input end of the fourth convolutional layer; the fourth The convolution layer includes 256 convolution kernels, each of which has a size of 3*3, a step size of (2,1), and an activation function of Relu. The output of the fourth convolution layer is connected to the input of the fifth convolution layer; the fifth convolution layer includes 512 convolution kernels, each of which has a size of 3*3, a step size of (2,1), and an activation function of Relu. The output of the fifth convolution layer is connected to the input of the sixth convolution layer; the sixth convolution layer includes 512 convolution kernels, each of which has a size of 2*1, a step size of 1, and an activation function of Relu. The output of the sixth convolution layer is connected to the input of the seventh convolution layer; the seventh convolution layer includes a Squeeze module, which performs a Squeeze operation on the input features, removes the first dimension, and obtains the output features of the CNN module.

[0018] Optionally, in a fourth implementation method of the second aspect of the present invention, the training module includes: a division unit, used to divide the pre-processed logistics trajectory historical data into a training set and a verification set in a ratio of 7:3; a training unit, used to set training parameters, and use the training set to train the hybrid model. During the training process, the weights and biases of the hybrid model are updated through a stochastic gradient descent algorithm to minimize the loss function; a verification unit, used to verify the hybrid model in the training process using the verification set, and adjust the structure or training parameters of the hybrid model according to the verification results to obtain the trajectory prediction model.

[0019] Optionally, in the fifth implementation method of the second aspect of the present invention, the prediction module includes: an input unit, used to input the pre-processed logistics trajectory data to be predicted into the trajectory prediction model to obtain predicted path planning position points and their corresponding timestamps; an output unit, used to connect the path planning position points according to the timestamps and then perform smoothing to output the predicted order logistics trajectory.

[0020] The third aspect of the present invention provides an order trajectory prediction device, comprising: a memory and at least one processor, wherein the memory stores computer-readable instructions, and the memory and the at least one processor are interconnected through lines; the at least one processor calls the computer-readable instructions in the memory so that the order trajectory prediction device executes the various steps of the order trajectory prediction method as described above.

[0021] A fourth aspect of the present invention provides a computer-readable storage medium, in which computer-readable storage medium is stored computer-readable instructions, which, when executed on a computer, enable the computer to execute the various steps of the order trajectory prediction method described above.

[0022] Beneficial effects: In the technical solution of the present invention, a hybrid model based on deep learning combined with recurrent neural network (RNN) is used for trajectory prediction. CNN is used to extract spatial features in the data, and RNN is used to capture time series features. The attention mechanism is introduced into the model to assign different weights according to the importance of different features to improve the prediction accuracy. The present invention can achieve accurate prediction of order trajectories, thereby facilitating the output of accurately predicted order trajectory results to e-commerce platforms and users in an intuitive and clear manner, including the estimated update time of the trajectory, possible delays, and detailed route information. The method of the present invention can provide e-commerce platforms and users with clearer and more accurate logistics information, thereby improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A first flow chart of the order trajectory prediction method provided by an embodiment of the present invention;

[0024] Figure 2 A second flow chart of the order trajectory prediction method provided by an embodiment of the present invention;

[0025] Figure 3 A third flow chart of the order trajectory prediction method provided by an embodiment of the present invention;

[0026] Figure 4 A fifth flow chart of the order trajectory prediction method provided by an embodiment of the present invention;

[0027] Figure 5 A sixth flow chart of the order trajectory prediction method provided by an embodiment of the present invention;

[0028] Figure 6 A schematic diagram of the structure of an order trajectory prediction device provided by an embodiment of the present invention;

[0029] Figure 7 Another structural schematic diagram of the order trajectory prediction device provided by an embodiment of the present invention;

[0030] Figure 8 A schematic diagram of the structure of an order trajectory prediction device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The embodiment of the present invention provides an order trajectory prediction method, device, equipment and storage medium, collects logistics trajectory historical data, the logistics trajectory historical data includes the starting position of the package, transportation route, transportation tool, weather conditions, traffic flow, distribution center processing efficiency, pickup time and receipt time; cleans, screens and standardizes the logistics trajectory historical data to obtain pre-processed logistics trajectory historical data; constructs a hybrid model based on deep learning and recurrent neural network, the hybrid model includes a CNN module for extracting spatial features, an RNN module for extracting spatial features, and an attention mechanism module for assigning weights to different features; uses the pre-processed logistics trajectory historical data to train the hybrid model to obtain a trajectory prediction model; inputs the pre-processed logistics trajectory data to be predicted into the trajectory prediction model, and outputs the predicted order logistics trajectory. An order trajectory prediction method provided by the present invention can realize the accurate prediction of order trajectory, so as to facilitate the output of the accurately predicted order trajectory results to the e-commerce platform and users in an intuitive and clear manner, including the estimated update time of the trajectory, possible delays and detailed route information. The method of the present invention can provide clearer and more accurate logistics information for e-commerce platforms and users, and improve user experience.

[0032] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0033] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , the first embodiment of the order trajectory prediction method in the embodiment of the present invention includes:

[0034] S100. Collecting historical data of logistics track, wherein the historical data of logistics track includes the starting location of the package, transportation route, transportation tool, weather conditions, traffic flow, distribution center processing efficiency, pickup time and receipt time.

[0035] Specifically, logistics trajectory historical data refers to a data set that records the moving paths of goods or transportation tools (such as trucks, express vehicles, etc.) during logistics transportation. Logistics trajectory historical data has the characteristics of large-scale, multidimensional, spatiotemporal and complex. The amount of data generated by logistics vehicles every day reaches GB or even TB level, and continues to grow; each trajectory data contains multidimensional information such as location, time, speed, direction, etc.; the trajectory data forms a curve on the geographic space and time axis, reflecting the spatiotemporal changes of the moving object; due to human factors (such as driving habits), geographical factors (such as road conditions), weather factors, etc., the trajectory data is complex and uncertain. In order to achieve subsequent accurate prediction of order trajectories, the logistics trajectory historical data collected in this embodiment mainly includes the starting location of the package, transportation path, transportation tools, weather conditions, traffic flow, distribution center processing efficiency, pickup time and receipt time.

[0036] S200: Clean, filter and standardize the logistics trajectory historical data to obtain pre-processed logistics trajectory historical data.

[0037] Specifically, there may be missing values, abnormal values ​​(such as unreasonable geographical location, timestamp errors, etc.), duplicate values ​​and other problems in the logistics trajectory historical data. If these problem data are used directly for analysis, they will seriously affect the accuracy and reliability of the results. Therefore, cleaning, screening and standardizing the logistics trajectory historical data are key steps to ensure data quality and improve the accuracy and efficiency of data analysis. Logistics trajectory historical data usually contains multiple types of variables (such as time, distance, speed, etc.). These variables differ in dimensions, units, value ranges, etc. Direct comparison or analysis may be misleading. This embodiment performs standardization on them, which can convert variables of different dimensions into dimensionless values ​​for easy comparison and analysis. The standardized data has a unified value range and distribution characteristics, so that data from different sources or at different time points can be compared. At the same time, in machine learning or data mining, standardization can improve the operating efficiency and accuracy of the algorithm and reduce numerical problems in the calculation process.

[0038] S300, constructing a hybrid model based on the combination of deep learning and recurrent neural network, wherein the hybrid model includes a CNN module for extracting spatial features, an RNN module for extracting spatial features, and an attention mechanism module for assigning weights to different features.

[0039] Specifically, the hybrid model based on the combination of deep learning and recurrent neural network (RNN) is a powerful data processing and prediction tool that combines the feature extraction ability of deep learning with the sequence processing ability of recurrent neural network to process complex time series data and data with spatial structure. In the hybrid model, the deep learning part (such as convolutional neural network CNN, graph convolution network GCN, etc.) is usually used to extract the feature representation of the data, while the recurrent neural network (RNN) or its variants (such as long short-term memory network LSTM, gated recurrent unit GRU, etc.) is used to process the temporal dependencies in these feature representations. This combination enables the model to capture both the spatial structure and temporal dynamics of the data. As an example, Graph Convolutional LSTM (graph convolutional long short-term memory network) is a typical hybrid model that combines graph convolutional network (GCN) and LSTM to handle time series prediction problems with graph structured data. GCN is used to extract the feature representation of each node in the graph, while LSTM is used to capture the changing trend of these feature representations over time.

[0040] S400: Using the pre-processed logistics trajectory historical data to train the hybrid model to obtain a trajectory prediction model.

[0041] S500, input the pre-processed logistics trajectory data to be predicted into the trajectory prediction model, and output the predicted order logistics trajectory.

[0042] Specifically, since logistics trajectory data is usually affected by many factors (such as traffic conditions, weather conditions, geographical location, etc.), these factors make the prediction of logistics trajectories complicated. The hybrid model based on deep learning and recurrent neural networks constructed by the present invention has strong adaptability and robustness, can handle complex and changing environmental factors, and give relatively accurate prediction results. Based on accurate prediction results, customers can understand the transportation status and estimated arrival time of the package in real time, enhance their trust and satisfaction with logistics services; at the same time, when encountering delays or abnormal situations, logistics companies can notify customers in advance based on the prediction results and provide solutions to reduce customer dissatisfaction and complaints.

[0043] See also Figure 2 , a second embodiment of the order trajectory prediction method in the embodiment of the present invention includes:

[0044] S110, retrieve the starting location, transportation route, transportation tool, pickup time and receipt time of the package from the internal system of the logistics company through the API interface;

[0045] S120, according to the transportation route, the pickup time and the receipt time, retrieve the weather conditions at the corresponding location and time from the weather service system, and retrieve the traffic flow at the corresponding location and time from the traffic data system;

[0046] S130. Obtain the time and throughput of parcel processing at each distribution center from the internal system of the logistics company through an API interface to obtain the processing efficiency of the distribution center.

[0047] Specifically, first, you need to apply for or obtain access to the API interface from the logistics company. These interfaces usually allow developers to obtain detailed information about the package through HTTP requests, including the starting location, transportation route, transportation tool, pickup time, and receipt time, etc.; through the written program or script, use the API interface to send a request to the logistics company system, and parse the returned data in JSON or XML format to extract the required information. According to the transportation route, pickup time, and receipt time of the package, the weather conditions at the corresponding location and time are retrieved through the API interface of the weather service system (such as OpenWeatherMap, Hefeng Weather, etc.). This helps to analyze the impact of weather on logistics transportation; similarly, according to the transportation route and time, the traffic flow information at the corresponding location and time is retrieved through the API interface of the traffic data system (such as Gaode Map, Baidu Map, etc.), which helps to evaluate the impact of traffic conditions on logistics transportation. Again, the API interface of the internal system of the logistics company is used to retrieve the time (such as arrival time, processing time, departure time, etc.) and throughput data of each distribution center for processing packages. According to the time and throughput of the distribution center for processing packages, the processing efficiency can be calculated, such as the number of packages processed per hour and the average processing time.

[0048] This embodiment obtains data in real time through the API interface, which can ensure the accuracy and timeliness of the information and provide timely support for logistics decisions; this embodiment not only obtains basic information on logistics transportation, but also combines external factors such as weather and traffic, as well as the internal processing of the distribution center, to provide a more comprehensive perspective for logistics analysis; based on the acquired data, this embodiment can optimize the logistics transportation plan, such as selecting a better transportation route, adjusting transportation tools, and responding to weather and traffic conditions in advance, thereby improving the efficiency and reliability of logistics transportation.

[0049] See also Figure 3 , a third embodiment of the order trajectory prediction method in the embodiment of the present invention includes:

[0050] S210, using a statistical method to identify and process abnormal values ​​in the logistics trajectory historical data to obtain screening data;

[0051] S220, deleting or merging duplicate record data in the screened data to ensure the uniqueness of the data set and obtain merged data;

[0052] S230, encoding the combined data and converting it into a unified format to obtain pre-processed logistics trajectory historical data.

[0053] Specifically, we first need to select appropriate statistical methods to identify outliers in the data. Common methods include methods based on standard deviation (such as Z-score, IQR, etc.), distribution-based methods (such as assuming that the data follows a normal distribution, the statistical characteristics of the normal distribution can be used to identify outliers), etc. After identifying outliers, we can choose to delete these values, replace them with the mean, median or other reasonable values, or mark them as special values ​​for subsequent analysis according to the actual situation. Next, by comparing the records in the data set, find those records that are exactly the same in all or key fields. This usually requires the use of database queries or data processing functions in programming languages; for the identified duplicate records, we can choose to delete some of them (usually retain one as a representative) or merge them into one record (if applicable, such as merging multiple identical transportation events into a more detailed record). For categorical variables in the data set (such as city name, type of transportation, etc.), encoding is required to ensure that all fields in the data set follow a unified format and type for subsequent data analysis and processing. The benefits of this embodiment are: by identifying and processing outliers and deleting or merging duplicate records, the quality of the data set can be significantly improved, which helps to reduce deviations and errors in data analysis and improve the reliability of the results; unified data format and encoding can simplify subsequent data processing and analysis, which makes data easier to be processed by different software tools and systems, reduces the complexity of data integration and conversion, and when data is presented in a unified format, they are easier to be processed by visualization tools. This helps analysts understand the patterns and trends in the data more intuitively, so as to make more informed decisions. The preprocessed data set can be used as input for advanced analysis technologies such as machine learning and data mining. These technologies can discover hidden patterns and relationships in data, thereby providing enterprises with deeper insights and competitive advantages.

[0054] A fourth embodiment of the order trajectory prediction method in the embodiment of the present invention includes: constructing a hybrid model using a CNN module, an RNN module, and an attention mechanism module;

[0055] The composition structure of the CNN module includes: the input end is connected to the first convolution layer, the first convolution layer includes 64 convolution kernels, each convolution kernel size is 3*3, the step size is 2, the activation function is Relu, and the output end of the first convolution layer is connected to the input end of the second convolution layer; the second convolution layer includes 128 convolution kernels, each convolution kernel size is 3*3, the step size is 2, the activation function is Relu, and the output end of the second convolution layer is connected to the input end of the third convolution layer; the third convolution layer includes four branches, the output ends of the four branches are connected, and then connected to the input end of the fourth convolution layer, wherein the first branch is a convolution branch, including 128 convolution kernels, each convolution kernel size is 1*1, and the activation function is Relu; the second The first branch is a depth-separable convolution branch, including 128 convolution kernels, each of which is 3*3 in size, with a step size of 1, a dilation rate of 1, and an activation function of Relu; the third branch is a depth-separable convolution branch, including 128 convolution kernels, each of which is 3*3 in size, with a step size of 1, a dilation rate of 3, and an activation function of Relu; the fourth branch is a depth-separable convolution branch, including 128 convolution kernels, each of which is 3*3 in size, with a step size of 1, a dilation rate of 5, and an activation function of Relu; in this embodiment, the pre-processed logistics trajectory historical data is added to the convolution module through the input end, and when it reaches the third convolution layer, the output features of the four branches are spliced ​​and input into the fourth convolution layer.

[0056] In this embodiment, the CNN module also includes: the fourth convolution layer includes 256 convolution kernels, each convolution kernel size is 3*3, the step size is (2,1), the activation function is Relu, and the output end of the fourth convolution layer is connected to the input end of the fifth convolution layer; the fifth convolution layer includes 512 convolution kernels, each convolution kernel size is 3*3, the step size is (2,1), the activation function is Relu, and the output end of the fifth convolution layer is connected to the input end of the sixth convolution layer; the sixth convolution layer includes 512 convolution kernels, each convolution kernel size is 2*1, the step size is 1, the activation function is Relu, and the output end of the sixth convolution layer is connected to the input end of the seventh convolution layer; the seventh convolution layer includes a Squeeze module, and the Squeeze module performs a Squeeze operation on the input features, removes the first dimension, and obtains the output features of the CNN module.

[0057] The specific structure of the RNN module and the attention mechanism module is as follows: the output features of the CNN module are used as the input features of the RNN module and the attention mechanism module, and are input into the RNN module and the attention mechanism module. The first layer includes two branches. The first branch includes a Position Embedding layer. The output end of the Position Embedding layer is respectively connected to the adder A and the first Multi-head Attention layer. The output end of the first Multi-head Attention layer is connected to the adder A. The output end of the adder A is connected to the first Layer Normalization layer. The output end of the first Layer Normalization layer is respectively connected to the adder B and the first Position-wise Feed-Forward layer. The output end of the first Position-wise Feed-Forward layer is connected to the adder B. The output end of the adder B is connected to the second Layer Normalization layer. The output end of the second Layer Normalization layer is connected to the adder C.

[0058] The second branch includes a first bidirectional LSTM layer, and the output end of the first bidirectional LSTM layer is connected to an adder C; the second layer includes two branches, the first branch includes a second Multi-head Attention layer connected to the output end of the second Layer Normalization layer, the output end of the second Multi-head Attention layer and the output end of the second LayerNormalization layer are connected to an adder D, the output end of the adder D is connected to a third Layer Normalization layer, the output end of the third Layer Normalization is respectively connected to a fourth Position-wise Feed-Forward layer and an adder E, the output end of the fourth Position-wise Feed-Forward layer is connected to adder E, the output end of adder E is connected to a fourth Layer Normalization layer, and the output end of the fourth Layer Normalization layer is connected to an adder F; the second branch includes a second bidirectional LSTM layer, the output end of adder C is connected to the second bidirectional LSTM layer, and the output end of the second bidirectional LSTM layer is connected to adder F; the third layer includes a fifth Layer Normalization layer, the output end of adder F is connected to the fifth Layer Normalization layer, and the fifth Layer The output of the Normalization layer is connected to the third bidirectional LSTM layer, and the output of the third bidirectional LSTM layer is connected to the fully connected layer Dense. The number of neurons in the fully connected layer is the number of characters plus 1, and finally the feature matrix is ​​output.

[0059] The preferred LSTM in this embodiment solves the gradient vanishing and gradient exploding problems of the traditional RNN, and can capture long-term dependencies, which is particularly important for predicting the order logistics trajectory of long distances or complex paths. At the same time, this embodiment can automatically learn the complex features in the data through the multi-layer nonlinear transformation in the deep learning model. After combining with RNN, the hybrid model can more comprehensively utilize the information in the order logistics data, thereby improving the accuracy of the prediction; because the deep learning model is adaptive, the hybrid model can automatically adjust its internal parameters and structure as the data increases and changes to adapt to the new logistics environment and transportation conditions. This dynamic adaptability makes the model more flexible and effective in practical applications; further, the prediction results of the hybrid model can be visualized, and the predicted order logistics trajectory can be compared and analyzed with the actual trajectory. This visual analysis helps users intuitively understand the prediction effect of the model and discover potential problems and improvement points; based on the prediction results of the hybrid model, enterprises can formulate more scientific and reasonable logistics plans and scheduling plans. At the same time, the model can also provide real-time logistics monitoring and early warning functions to help enterprises promptly discover and respond to potential logistics risks.

[0060] See also Figure 4 The fifth embodiment of order trajectory prediction in the embodiment of the present invention includes:

[0061] S310, dividing the pre-processed logistics trajectory historical data into a training set and a validation set in a ratio of 7:3;

[0062] S320, setting training parameters, using the training set to train the hybrid model, and during the training process, updating the weights and biases of the hybrid model by a stochastic gradient descent algorithm to minimize the loss function;

[0063] S330, using the validation set to validate the hybrid model in the training process, adjusting the structure or training parameters of the hybrid model according to the validation result, to obtain the trajectory prediction model.

[0064] Specifically, the preprocessed logistics trajectory historical data is divided into a training set and a validation set in a ratio of 7:3. This can usually be achieved by random sampling or stratified sampling to ensure that the data distribution of the training set and the validation set is as consistent as possible. Then, according to the specific needs of the hybrid model, appropriate training parameters are set, such as learning rate, number of iterations, batch size, etc. The setting of these parameters has an important impact on the training effect of the model. The hybrid model is trained using the training set. During the training process, the weights and biases of the model are updated by the stochastic gradient descent (SGD) algorithm to minimize the loss function. The loss function is used to measure the difference between the model's predicted value and the actual value. By minimizing the loss function, the model can better fit the training data. Then the hybrid model in the training process is verified using the validation set. The role of the validation set is to evaluate the generalization ability of the model without touching the test set. According to the verification results, the structure or training parameters of the model can be adjusted to optimize the performance of the model. Repeat the above training and verification process until the performance of the model on the validation set is satisfactory. In this process, the model can be continuously adjusted to improve its generalization ability, thereby obtaining a trajectory prediction model.

[0065] This embodiment divides the preprocessed logistics trajectory historical data into a training set and a validation set in a ratio of 7:3, and uses the training set to train the hybrid model and uses the validation set to verify and adjust the model parameters. This has significant benefits in improving model performance, optimizing model structure, saving computing resources, and improving development efficiency.

[0066] See also Figure 5 The sixth embodiment of order trajectory prediction in the embodiment of the present invention includes:

[0067] S410, inputting the pre-processed logistics trajectory data to be predicted into the trajectory prediction model to obtain the predicted path planning position point and its corresponding timestamp;

[0068] S420, connecting the path planning location points according to timestamps and performing smoothing processing, and outputting a predicted order logistics trajectory.

[0069] In this embodiment, the trajectory prediction model uses deep learning and recurrent neural networks to reason based on the input data, and outputs the predicted path planning position points and their corresponding timestamps; the predicted path planning position points are connected in the order of timestamps to form a preliminary predicted trajectory, and the preliminary predicted trajectory is smoothed to eliminate the trajectory unevenness caused by prediction error or data noise. Smoothing can use a variety of algorithms, such as Kalman filtering, sliding average, curve fitting, etc., and the smoothed predicted trajectory is output as the final result. This embodiment can make full use of the time series characteristics and spatial characteristics in the logistics trajectory data by using a hybrid model of deep learning and recurrent neural networks to improve the accuracy of the prediction. Smoothing can eliminate noise and fluctuations in the predicted trajectory, making the predicted trajectory smoother and continuous, and more in line with the actual logistics and transportation situation. The predicted order logistics trajectory can provide important decision support for logistics companies, such as path planning, vehicle scheduling, inventory management, etc. By predicting logistics trajectories in advance, companies can arrange transportation resources more reasonably and improve transportation efficiency. At the same time, accurate order logistics trajectory predictions can help improve customer experience. Customers can understand the transportation status and estimated arrival time of their orders in real time, reducing waiting and anxiety. During the logistics transportation process, if an emergency occurs (such as traffic congestion, vehicle failure, etc.), companies can adjust and optimize based on real-time updated logistics trajectory information to ensure that orders can be delivered on time. By combining advanced technologies such as deep learning with logistics trajectory predictions, the intelligent development of the logistics industry can be promoted.

[0070] The order trajectory prediction method in the embodiment of the present invention is described above. The order trajectory prediction device in the embodiment of the present invention is described below. Figure 6 , an embodiment of the order trajectory prediction device in the embodiment of the present invention includes:

[0071] The collection module 10 is used to collect the historical data of the logistics track, which includes the starting location of the package, the transportation route, the transportation tool, the weather conditions, the traffic flow, the processing efficiency of the distribution center, the pickup time and the receipt time;

[0072] A processing module 20 is used to clean, filter and standardize the logistics trajectory historical data to obtain pre-processed logistics trajectory historical data;

[0073] A model building module 30, used to build a hybrid model based on a combination of deep learning and recurrent neural network, wherein the hybrid model includes a CNN module for extracting spatial features, an RNN module for extracting spatial features, and an attention mechanism module for assigning weights to different features;

[0074] A training module 40, for training the hybrid model using the pre-processed logistics trajectory historical data to obtain a trajectory prediction model;

[0075] The prediction module 50 is used to input the pre-processed logistics trajectory data to be predicted into the trajectory prediction model and output the predicted order logistics trajectory.

[0076] In this embodiment, a hybrid model based on deep learning combined with a recurrent neural network (RNN) is used for trajectory prediction. CNN is used to extract spatial features from the data, and RNN is used to capture time series features. The attention mechanism is introduced into the model, and different weights are assigned according to the importance of different features to improve the prediction accuracy. The present invention can achieve accurate prediction of order trajectories, thereby facilitating the output of accurately predicted order trajectory results to e-commerce platforms and users in an intuitive and clear manner, including the estimated update time of the trajectory, possible delays, and detailed route information. The method of the present invention can provide e-commerce platforms and users with clearer and more accurate logistics information, and improve user experience.

[0077] See also Figure 7 Another embodiment of the order trajectory prediction device in the embodiment of the present invention includes:

[0078] The collection module 10 is used to collect the historical data of the logistics track, which includes the starting location of the package, the transportation route, the transportation tool, the weather conditions, the traffic flow, the processing efficiency of the distribution center, the pickup time and the receipt time;

[0079] A processing module 20 is used to clean, filter and standardize the logistics trajectory historical data to obtain pre-processed logistics trajectory historical data;

[0080] A model building module 30, used to build a hybrid model based on a combination of deep learning and recurrent neural network, wherein the hybrid model includes a CNN module for extracting spatial features, an RNN module for extracting spatial features, and an attention mechanism module for assigning weights to different features;

[0081] A training module 40, for training the hybrid model using the pre-processed logistics trajectory historical data to obtain a trajectory prediction model;

[0082] The prediction module 50 is used to input the pre-processed logistics trajectory data to be predicted into the trajectory prediction model and output the predicted order logistics trajectory.

[0083] In this embodiment, the collection module 10 includes:

[0084] The first retrieval unit 11 is used to retrieve the starting location, transportation route, transportation tool, pickup time and receipt time of the package from the internal system of the logistics company through the API interface;

[0085] The second retrieving unit 12 is used to retrieve the weather conditions at the corresponding place and time from the weather service system according to the transportation route, the pickup time and the receipt time, and retrieve the traffic flow at the corresponding place and time from the traffic data system;

[0086] The third retrieval unit 13 is used to retrieve the time and throughput of parcel processing of each distribution center from the internal system of the logistics company through an API interface to obtain the processing efficiency of the distribution center.

[0087] In this embodiment, the processing module 20 includes:

[0088] A screening unit 21, used to identify and process abnormal values ​​in the logistics trajectory historical data by using a statistical method to obtain screening data;

[0089] A merging unit 22, used to delete or merge duplicate record data in the screened data to ensure the uniqueness of the data set and obtain merged data;

[0090] The conversion unit 23 is used to encode the combined data and convert it into a unified format to obtain pre-processed logistics trajectory historical data.

[0091] In this embodiment, the model building module 30 includes:

[0092] The construction unit 31 uses a CNN module, an RNN module and an attention mechanism module to construct a hybrid model; the composition structure of the CNN module includes: the input end is connected to the first convolution layer, the first convolution layer includes 64 convolution kernels, each convolution kernel size is 3*3, the step size is 2, the activation function is Relu, and the output end of the first convolution layer is connected to the input end of the second convolution layer; the second convolution layer includes 128 convolution kernels, each convolution kernel size is 3*3, the step size is 2, the activation function is Relu, and the output end of the second convolution layer is connected to the input end of the third convolution layer; the third convolution layer includes four branches, the output ends of the four branches are connected, and then connected to the input end of the fourth convolution layer; the fourth convolution layer includes 256 convolution kernels, each convolution kernel size is 3*3, the step size is 2, the activation function is Relu, and the output end of the second convolution layer is connected to the input end of the third convolution layer; The convolution kernel size is 3*3, the step size is (2,1), the activation function is Relu, and the output end of the fourth convolution layer is connected to the input end of the fifth convolution layer; the fifth convolution layer includes 512 convolution kernels, each convolution kernel size is 3*3, the step size is (2,1), the activation function is Relu, and the output end of the fifth convolution layer is connected to the input end of the sixth convolution layer; the sixth convolution layer includes 512 convolution kernels, each convolution kernel size is 2*1, the step size is 1, the activation function is Relu, and the output end of the sixth convolution layer is connected to the input end of the seventh convolution layer; the seventh convolution layer includes a Squeeze module, and the Squeeze module performs a Squeeze operation on the input features, removes the first dimension, and obtains the output features of the CNN module.

[0093] In this embodiment, the training module 40 includes:

[0094] A division unit 41 is used to divide the pre-processed logistics trajectory historical data into a training set and a verification set in a ratio of 7:3;

[0095] A training unit 42, used to set training parameters, train the hybrid model using the training set, and during the training process, update the weights and biases of the hybrid model by a stochastic gradient descent algorithm to minimize the loss function;

[0096] The verification unit 43 is used to verify the hybrid model in the training process using the verification set, and adjust the structure or training parameters of the hybrid model according to the verification result to obtain the trajectory prediction model.

[0097] In this embodiment, the prediction module 50 includes:

[0098] An input unit 51 is used to input the pre-processed logistics trajectory data to be predicted into the trajectory prediction model to obtain the predicted path planning position point and its corresponding time stamp;

[0099] The output unit 52 is used to connect the path planning position points according to the timestamps and then perform smoothing processing to output the predicted order logistics trajectory.

[0100] above Figure 6 and Figure 7 The order trajectory prediction device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The order trajectory prediction device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0101] Figure 8 1 is a schematic diagram of the structure of an order trajectory prediction device provided by an embodiment of the present invention. The order trajectory prediction device 100 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 11 (for example, one or more processors) and a memory 12, and one or more storage media 13 (for example, one or more massive storage devices) storing application programs 133 or data 132. Among them, the memory 12 and the storage medium 13 can be short-term storage or permanent storage. The program stored in the storage medium 13 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the order trajectory prediction device 100. Furthermore, the processor 11 can be configured to communicate with the storage medium 13 to execute a series of instruction operations in the storage medium 13 on the order trajectory prediction device 100.

[0102] The order trajectory prediction device 100 may also include one or more power supplies 14, one or more wired or wireless network interfaces 15, one or more input and output interfaces 16, and / or one or more operating systems 131, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that Figure 8 The device structure shown does not constitute a limitation of the order trajectory prediction device 100, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0103] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the order trajectory prediction method.

[0104] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0105] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0106] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting order trajectories, characterized in that: The order trajectory prediction method includes: Collect historical logistics trajectory data, including the package's starting location, transportation route, transportation vehicle, weather conditions, traffic flow, distribution center processing efficiency, pickup time, and receipt time; Cleaning, screening and standardizing the logistics trajectory historical data to obtain pre-processed logistics trajectory historical data; Constructing a hybrid model based on the combination of deep learning and recurrent neural networks, the hybrid model includes a CNN module for extracting spatial features, an RNN module for extracting spatial features, and an attention mechanism module for assigning weights to different features; The hybrid model is trained using the pre-processed logistics trajectory historical data to obtain a trajectory prediction model; The pre-processed logistics trajectory data to be predicted is input into the trajectory prediction model, and the predicted order logistics trajectory is output.

2. The order trajectory prediction method according to claim 1, characterized in that: The steps to collect historical data of logistics trajectory include: Through the API interface, the starting location, transportation route, transportation tool, pickup time and receipt time of the package are retrieved from the logistics company's internal system; Based on the transport route, pickup time, and receipt time, retrieve the weather conditions at the corresponding location and time from the weather service system, and retrieve the traffic flow at the corresponding location and time from the traffic data system; Through the API interface, the time and throughput of parcel processing at each distribution center are retrieved from the internal system of the logistics company to obtain the processing efficiency of the distribution center.

3. The order trajectory prediction method according to claim 1, characterized in that: The steps of cleaning, screening and standardizing the logistics trajectory historical data to obtain pre-processed logistics trajectory historical data include: Using statistical methods to identify and process outliers in the logistics trajectory historical data to obtain filtered data; Deleting or merging duplicate records in the screened data to ensure the uniqueness of the data set and obtain merged data; The combined data is encoded and converted into a unified format to obtain pre-processed logistics trajectory historical data.

4. The order trajectory prediction method according to claim 1, characterized in that: Building a hybrid model based on deep learning and recurrent neural networks includes the following steps: A hybrid model is constructed using CNN module, RNN module and attention mechanism module; The CNN module comprises: an input end connected to a first convolutional layer, wherein the first convolutional layer includes 64 convolution kernels, each of which is 3*3 in size, has a stride of 2, and uses a ReLU activation function; and an output end of the first convolutional layer is connected to an input end of a second convolutional layer; The second convolutional layer includes 128 convolution kernels, each of which is 3*3 in size, with a stride of 2 and an activation function of ReLU. The output of the second convolutional layer is connected to the input of the third convolutional layer. The third convolutional layer includes four branches, the output ends of the four branches are connected, and then connected to the input end of the fourth convolutional layer; The fourth convolutional layer includes 256 convolution kernels, each of which has a size of 3*3, a step size of (2, 1), an activation function of ReLU, and the output end of the fourth convolutional layer is connected to the input end of the fifth convolutional layer; The fifth convolutional layer includes 512 convolution kernels, each of which has a size of 3*3, a step size of (2, 1), an activation function of ReLU, and the output end of the fifth convolutional layer is connected to the input end of the sixth convolutional layer; The sixth convolutional layer includes 512 convolution kernels, each of which has a size of 2*1, a step size of 1, and an activation function of ReLU. The output end of the sixth convolutional layer is connected to the input end of the seventh convolutional layer; The seventh convolutional layer includes a Squeeze module, which performs a Squeeze operation on the input features, removes the first dimension, and obtains the output features of the CNN module.

5. The order trajectory prediction method according to claim 4, characterized in that: The four branches of the third convolutional layer are: The first branch is the convolution branch, which includes 128 convolution kernels, each of which is 1*1 in size and uses Relu as the activation function. The second branch is the depth-wise separable convolution branch, which includes 128 convolution kernels, each of which is 3*3 in size, with a step size of 1, a dilation rate of 1, and an activation function of ReLU. The third branch is the depth-wise separable convolution branch, which includes 128 convolution kernels, each of which is 3*3 in size, with a step size of 1, a dilation rate of 3, and an activation function of ReLU. The fourth branch is the depth-wise separable convolution branch, which includes 128 convolution kernels, each of which is 3*3 in size, with a step size of 1, a dilation rate of 5, and an activation function of ReLU.

6. The order trajectory prediction method according to claim 1, characterized in that: The steps of training the hybrid model using the pre-processed logistics trajectory historical data to obtain a trajectory prediction model include: The pre-processed logistics trajectory historical data is divided into a training set and a validation set in a ratio of 7:3; Setting training parameters, training the hybrid model using the training set, and updating weights and biases of the hybrid model using a stochastic gradient descent algorithm during the training process to minimize a loss function; The hybrid model in the training process is verified using the verification set, and the structure or training parameters of the hybrid model are adjusted according to the verification result to obtain the trajectory prediction model.

7. The order trajectory prediction method according to claim 1, characterized in that: Inputting the pre-processed logistics trajectory data to be predicted into the trajectory prediction model and outputting the predicted order logistics trajectory includes the following steps: Inputting the pre-processed logistics trajectory data to be predicted into the trajectory prediction model to obtain the predicted path planning location points and their corresponding timestamps; The path planning location points are connected according to the timestamps and then smoothed to output the predicted order logistics trajectory.

8. An order trajectory prediction device, characterized in that: include: A collection module is used to collect historical logistics trajectory data, including the package's starting location, transportation route, transportation vehicle, weather conditions, traffic flow, distribution center processing efficiency, pickup time, and receipt time; a processing module, configured to clean, filter, and standardize the logistics trajectory historical data to obtain pre-processed logistics trajectory historical data; A model building module, used to build a hybrid model based on the combination of deep learning and recurrent neural networks, wherein the hybrid model includes a CNN module for extracting spatial features, an RNN module for extracting spatial features, and an attention mechanism module for assigning weights to different features; A training module, configured to train the hybrid model using the pre-processed logistics trajectory historical data to obtain a trajectory prediction model; The prediction module is used to input the pre-processed logistics trajectory data to be predicted into the trajectory prediction model and output the predicted order logistics trajectory.

9. An order trajectory prediction device, characterized in that: comprising a memory and at least one processor, wherein the memory has computer-readable instructions stored therein; The at least one processor calls the computer-readable instructions in the memory to execute the various steps of the order trajectory prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium having computer-readable instructions stored thereon, characterized in that: When the computer-readable instructions are executed by a processor, the various steps of the order trajectory prediction method according to any one of claims 1 to 7 are implemented.

Citation Information

Cited By

  • Receiving information real-time synchronization and abnormity early warning method and system

    CN119863178A