Big data processing method and system for smart logistics scheduling

Abstract

The present application provides a big data processing method and system for intelligent logistics scheduling, wherein the scheduling constraint information has route constraint conditions corresponding to the target route constraint level, and the pre-debugging knowledge representation can also represent the pre-debugging knowledge representation corresponding to the route constraint conditions. At the same time, the pre-debugging knowledge representation mining network is based on the mapping of the constraint feature space to the route feature space, and the process of route construction based on the constraint vector is refined into two links from the constraint vector to the route vector and then to the route construction, which can effectively alleviate the debugging complexity of the route representation vector construction network, increase the quality of logistics scheduling data construction, load the initial logistics scheduling data and the target route representation vector into the logistics scheduling data construction network for logistics scheduling data construction, obtain the target scheduling logistics scheduling data, and make the output target scheduling logistics scheduling data have accurate route scheduling results.

Description

Technical Field

[0001] The present application relates to the field of data processing, and more specifically, to a big data processing method and system applied to smart logistics scheduling. Background Art

[0002] With the rapid development of e-commerce and logistics industries, the efficiency and accuracy of logistics scheduling are crucial to the operation of enterprises and customer satisfaction. Traditional logistics scheduling methods usually rely on manual experience and simple rules, which are difficult to meet the increasingly complex logistics constraints. Although the existing logistics scheduling methods can perform basic scheduling according to preset routes and rules, they are often unable to adaptively optimize the details of logistics routes when faced with complex actual situations. For example, when factors such as road traffic conditions, weather changes, and customer needs change, traditional scheduling methods may not be able to adjust the route in time, resulting in low logistics efficiency. In addition, existing logistics scheduling methods can usually only provide relatively rough route planning and cannot take into account the details of the logistics process. For example, the inability to perform refined route planning based on factors such as the characteristics of the goods and the capacity of the transportation tools may lead to problems such as increased logistics costs and extended transportation time.

[0003] Therefore, there is a need for a big data processing method applied to smart logistics scheduling that can automatically optimize the details of logistics routes and generate new logistics routes, thereby improving logistics efficiency, reducing costs, and improving customer satisfaction. Summary of the invention

[0004] In view of this, the embodiments of the present application at least provide a big data processing method and system applied to smart logistics scheduling.

[0005] According to one aspect of an embodiment of the present application, a big data processing method for smart logistics scheduling is provided, the method comprising:

[0006] Acquire initial logistics scheduling data and scheduling constraint information having target logistics route data, wherein the scheduling constraint information is used to indicate the construction of logistics scheduling data using the target logistics route data as logistics route data, and the scheduling constraint information includes route constraint conditions corresponding to the target route constraint level;

[0007] Loading the scheduling constraint information into a pre-debugging knowledge representation mining network to perform pre-debugging knowledge representation mining to obtain a pre-debugging knowledge representation corresponding to the scheduling constraint information;

[0008] Loading the pre-commissioning knowledge representation into a route representation vector construction network, adjusting the route according to the pre-commissioning knowledge representation, and constructing a target route representation vector corresponding to the target logistics route data;

[0009] Loading the initial logistics scheduling data and the target route characterization vector into the logistics scheduling data construction network to construct the logistics scheduling data, and obtaining the target scheduling logistics scheduling data;

[0010] Among them, the route characterization vector construction network is based on route constraint condition samples corresponding to multiple preset route constraint levels and a multi-level route line adjustment network corresponding to the multiple preset route constraint levels, and the preset route characterization vector construction network is taught to perform route line adjustment and debugging of the multiple preset route constraint levels; the target route constraint level is one or more preset route constraint levels among the multiple preset route constraint levels.

[0011] According to an example of an embodiment of the present application, the step of loading the pre-debugging knowledge representation into a route representation vector construction network, adjusting the route according to the pre-debugging knowledge representation, and constructing a target route representation vector corresponding to the target logistics route data includes:

[0012] Loading the pre-debugging knowledge representation and the preset enhanced knowledge representation into the route representation vector construction network, and guiding the preset enhanced knowledge representation to perform the adjustment operation of the target route constraint level according to the pre-debugging knowledge representation to obtain the target route representation vector;

[0013] The pre-debugging knowledge representation mining network includes a first information embedding mapping component, a second information embedding mapping component and a multimodal representation vector mapping network, and the loading of the scheduling constraint information into the pre-debugging knowledge representation mining network for pre-debugging knowledge representation mining, and obtaining the pre-debugging knowledge representation corresponding to the scheduling constraint information includes:

[0014] The scheduling constraint information is loaded into the first information embedding mapping component to perform constraint condition feature mining to obtain a first constraint condition representation vector, wherein the first information embedding mapping component is an information embedding mapping component using multi-modal alignment pre-debugging;

[0015] The scheduling constraint information is loaded into the second information embedding mapping component for constraint condition feature mining to obtain a second constraint condition representation vector, wherein the second information embedding mapping component is an information embedding mapping component that adopts constraint condition feature mining pre-debugging;

[0016] The first constraint condition representation vector and the second constraint condition representation vector are loaded into the multimodal representation vector mapping network to perform multimodal representation vector mapping to obtain the pre-debugging knowledge representation.

[0017] According to an example of an embodiment of the present application, the logistics scheduling data construction network includes: a scheduling data feature extraction component, a representation vector integration component and a data restoration mapping component, the initial logistics scheduling data and the target route representation vector are loaded into the logistics scheduling data construction network to construct the logistics scheduling data, and obtaining the target scheduling logistics scheduling data includes:

[0018] Loading the initial logistics scheduling data into the scheduling data feature extraction component to perform data representation vector mining to obtain the initial logistics scheduling data representation vector;

[0019] Loading the initial logistics scheduling data representation vector and the target route representation vector into the representation vector integration component to perform a representation vector integration operation to obtain an integrated logistics scheduling data representation vector;

[0020] The integrated logistics scheduling data representation vector is loaded into the data restoration mapping component for data restoration mapping to obtain the target scheduling logistics scheduling data.

[0021] According to an example of an embodiment of the present application, the method further includes:

[0022] Cloning the network internal variables of the preset route representation vector construction network to obtain a first route representation vector construction network and a second route representation vector construction network, wherein the first route representation vector construction network and the second route representation vector construction network are at the same level;

[0023] Acquire a first pre-debugging knowledge representation sample and an enhanced knowledge representation sample corresponding to a first scheduling constraint information sample having the route constraint condition sample;

[0024] When adjusting the first route representation vector construction network and the second route representation vector construction network to construct a route representation vector according to the first pre-debugging knowledge representation sample, the enhanced knowledge representation sample and the preset subsidiary enhancement information corresponding to the enhanced knowledge representation sample, adding the route constraint condition sample to the first route representation vector construction network according to the multi-level route line adjustment network, so that the first route representation vector construction network is adjusted and debugged to obtain a first debugging network;

[0025] Merging the network internal variables of the first debugging network and the network internal variables of the second route representation vector construction network in order to obtain the target network internal variables;

[0026] According to the target network internal variables, the network internal variables of the preset route representation vector construction network are optimized to obtain the route representation vector construction network.

[0027] According to an example of an embodiment of the present application, the first route representation vector construction network includes: a first embedding mapping component, the second route representation vector construction network includes: a second embedding mapping component and a restoration mapping component, and when the first route representation vector construction network and the second route representation vector construction network are adjusted to perform route representation vector construction according to the first pre-debugging knowledge representation sample, the enhanced knowledge representation sample, and the preset subsidiary enhancement information corresponding to the enhanced knowledge representation sample, the route constraint condition sample is added to the first route representation vector construction network according to the multi-level route line adjustment network, so that the first route representation vector construction network performs route line adjustment debugging, and obtaining the first debugging network includes:

[0028] Loading the route constraint condition samples corresponding to the plurality of preset route constraint levels into the multi-level route line adjustment network to perform route constraint condition representation vector mining to obtain route constraint condition representation vector samples;

[0029] Loading the enhanced knowledge representation sample, the first pre-debugging knowledge representation sample and the route constraint condition representation vector sample into the first embedding mapping component, and embedding and mapping the enhanced knowledge representation sample according to the first pre-debugging knowledge representation sample and the route constraint condition representation vector sample to obtain a first embedding mapping representation vector;

[0030] Loading the enhanced knowledge representation sample and the first pre-debugging knowledge representation sample into the second embedding mapping component, and performing embedding mapping on the enhanced knowledge representation sample according to the first pre-debugging knowledge representation sample to obtain a second embedding mapping representation vector;

[0031] Loading the first embedded mapping representation vector, the second embedded mapping representation vector and the first pre-debugging knowledge representation sample into the restoration mapping component, and performing restoration mapping on the first embedded mapping representation vector and the second embedded mapping representation vector according to the first pre-debugging knowledge representation sample to obtain a route representation vector sample;

[0032] According to a first enhancement error between a first enhancement information sample and the preset subsidiary enhancement information, the first embedding mapping component is debugged to obtain the first debugging network; the first enhancement information sample is the enhancement information between the enhancement knowledge representation sample and the route representation vector sample.

[0033] According to an example of an embodiment of the present application, the first embedding mapping component includes a first pooling operator of a cascaded multi-feature granularity, the route constraint representation vector sample includes the route constraint representation vector of the multi-feature granularity, the first embedding mapping representation vector includes the first pooling representation vector of the multi-feature granularity, the enhanced knowledge representation sample, the first pre-debugging knowledge representation sample and the route constraint representation vector sample are loaded into the first embedding mapping component, and the enhanced knowledge representation sample is embedded and mapped according to the first pre-debugging knowledge representation sample and the route constraint representation vector sample. The step of obtaining the first embedded mapping representation vector includes: based on the first pooling operator of each feature granularity in the multiple feature granularities, extracting the pooling representation vector of the execution representation vector of the first pooling operator of each feature granularity according to the first pre-debugging knowledge representation sample, and obtaining the first pooling representation vector of each feature granularity, wherein the execution representation vector of the first pooling operator of the first feature granularity is the enhanced knowledge representation sample, and the execution representation vectors of the first pooling operators from the second feature granularity to the last feature granularity are the first pooling representation vector of the previous feature granularity itself and the route constraint representation vector of the previous feature granularity itself;

[0034] The second embedding mapping component includes a cascaded second pooling operator of the multiple feature granularities, the second embedding mapping representation vector includes the second pooling representation vector of the multiple feature granularities, the loading of the enhanced knowledge representation sample and the first pre-debugging knowledge representation sample into the second embedding mapping component, embedding and mapping the enhanced knowledge representation sample according to the first pre-debugging knowledge representation sample, and obtaining the second embedding mapping representation vector includes: based on the second pooling operator of each feature granularity, according to the first pre-debugging knowledge representation sample, performing pooling representation vector extraction on the execution representation vector of the second pooling operator of each feature granularity, and obtaining the second pooling representation vector of each feature granularity, wherein the execution representation vector of the second pooling operator of the first feature granularity is the enhanced knowledge representation sample, and the execution representation vector of the second pooling operator from the second feature granularity to the last feature granularity is the second pooling representation vector of the previous feature granularity itself;

[0035] The restoration mapping component includes the cascaded anti-pooling operators of the multiple feature granularities, the first embedded mapping representation vector, the second embedded mapping representation vector and the first pre-debugging knowledge representation sample are loaded into the restoration mapping component, and the first embedded mapping representation vector and the second embedded mapping representation vector are restored and mapped according to the first pre-debugging knowledge representation sample, and the route representation vector sample is obtained, including: based on the anti-pooling operator of each feature granularity, according to the first pre-debugging knowledge representation sample, the execution representation vector of the anti-pooling operator of each feature granularity is extracted by anti-pooling representation vector, and the anti-pooling representation vector of each feature granularity is obtained, wherein , the execution characterization vector of the reverse pooling operator of the last feature granularity connected to the second pooling operator of the last feature granularity is the second pooling characterization vector of the last feature granularity, the first pooling characterization vector of the same feature granularity as itself, and the route constraint characterization vector of the same feature granularity as itself, and the execution characterization vector of the reverse pooling operator from the first feature granularity to the last second feature granularity includes the reverse pooling characterization vector of the next feature granularity itself, the first pooling characterization vector of the same feature granularity as itself, and the route constraint characterization vector of the same feature granularity as itself; the reverse pooling characterization vector output by the reverse pooling operator of the first feature granularity is used as the route characterization vector sample.

[0036] According to an example of an embodiment of the present application, the multi-level route regulation network includes route regulation components corresponding to the plurality of preset route constraint levels, and the route constraint condition samples corresponding to the plurality of preset route constraint levels are loaded into the multi-level route regulation network to perform route constraint condition representation vector mining, and obtaining the route constraint condition representation vector samples includes:

[0037] Loading the route constraint condition samples corresponding to each preset route constraint level in the plurality of preset route constraint levels into the route adjustment component corresponding to the preset route constraint level to perform route constraint condition representation vector mining, and obtaining the target route constraint condition representation vector corresponding to each preset route constraint level;

[0038] The target route constraint condition representation vectors corresponding to the plurality of preset route constraint levels are eccentrically adjusted and integrated to obtain the route constraint condition representation vector sample.

[0039] According to an example of an embodiment of the present application, the route adjustment component includes a coordinate inverse combination operator and a cascaded multi-feature granularity residual representation vector extraction operator, the target route constraint condition representation vector includes the multi-feature granularity residual route constraint condition representation vector, and the route constraint condition samples corresponding to each preset route constraint level in the multiple preset route constraint levels are loaded into the route adjustment component of the corresponding preset route constraint level to perform route constraint condition representation vector mining, and obtaining the target route constraint condition representation vector corresponding to each preset route constraint level includes:

[0040] Loading the route constraint condition samples corresponding to each preset route constraint level into the coordinate inverse combination operator corresponding to the preset route constraint level to perform coordinate inverse combination processing, so as to obtain the initial constraint condition representation vector corresponding to each preset route constraint level;

[0041] According to the residual representation vector extraction operator of the multi-feature granularity corresponding to each preset route constraint level, the residual representation vector of the multi-feature granularity is extracted for the initial constraint condition representation vector to obtain the residual route constraint condition representation vector of the multi-feature granularity corresponding to each preset route constraint level;

[0042] The route constraint condition representation vector example includes the route constraint condition representation vector of the multi-feature granularity, and the eccentric adjustment and integration of the target route constraint condition representation vectors corresponding to the multiple preset route constraint levels to obtain the route constraint condition representation vector example includes:

[0043] The residual route constraint condition representation vectors of each characteristic granularity corresponding to the plurality of preset route constraint levels are respectively eccentrically adjusted and integrated to obtain the route constraint condition representation vector of each characteristic granularity.

[0044] According to an example of an embodiment of the present application, the method further includes:

[0045] Obtaining an initial logistics scheduling data sample having a logistics route data sample, a second scheduling constraint information sample having the route constraint condition sample, and a logistics scheduling data representation vector sample corresponding to the scheduling logistics scheduling data sample matching the second scheduling constraint information sample, wherein the second scheduling constraint information sample is used to indicate the construction of logistics scheduling data with the logistics route data sample as the logistics route data;

[0046] According to the second scheduling constraint information sample and the logistics scheduling data representation vector sample, pre-debugging knowledge representation mining debugging is performed on a preset pre-debugging knowledge representation mining network to obtain the pre-debugging knowledge representation mining network;

[0047] Determining a second pre-debugging knowledge representation example corresponding to the second scheduling constraint information example according to the pre-debugging knowledge representation mining network;

[0048] Based on the route constraint condition sample, the second pre-debugging knowledge representation sample, the initial logistics scheduling data sample and the scheduled logistics scheduling data sample, the preset route characterization vector construction network, the multi-level route line adjustment network and the preset neural network are collaboratively debugged to obtain the route characterization vector construction network, the debugged multi-level route line adjustment network and the logistics scheduling data construction network.

[0049] According to another aspect of an embodiment of the present application, a big data processing system is provided, including: a processor; and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the method described above.

[0050] The beneficial effects of this application include at least:

[0051] The big data processing method and system for intelligent logistics scheduling provided in the present application are based on obtaining initial logistics scheduling data and scheduling constraint information with target logistics route data, the scheduling constraint information is used to indicate the construction of logistics scheduling data with target logistics route data as logistics route data, the scheduling constraint information has route constraint conditions corresponding to the target route constraint level, the target route constraint level is, for example, one or more preset route constraint levels among multiple preset route constraint levels, and then the scheduling constraint information is loaded into the pre-debugging knowledge representation mining network for pre-debugging knowledge representation mining to obtain the pre-debugging knowledge representation corresponding to the scheduling constraint information, because the scheduling constraint information has the route constraint conditions corresponding to the target route constraint level, the pre-debugging knowledge representation can also represent the pre-debugging knowledge representation corresponding to the route constraint conditions, and at the same time, the pre-debugging knowledge representation mining network is based on the mapping of constraint feature space to route feature space, and the process of route construction based on constraint vector is refined into two links from constraint vector to route vector and then to route construction, which can effectively alleviate the debugging complexity of the route representation vector construction network and increase the quality of logistics scheduling data construction, and then The pre-debugging knowledge representation is loaded into the route representation vector construction network, and the route line adjustment is performed according to the pre-debugging knowledge representation to construct the target route representation vector corresponding to the target logistics route data, wherein the route representation vector construction network is a multi-level route line adjustment network corresponding to the route constraint condition samples corresponding to multiple preset route constraint levels and the preset route constraint levels, and the preset route representation vector construction network is taught to perform route line adjustment and debugging at multiple preset route constraint levels, and the preset route representation vector construction network is guided to perform route line adjustment and debugging at multiple preset route constraint levels based on the multi-level route line adjustment network, that is, the preset route representation vector construction network is optimized in detail, so that the route representation vector construction network obtained after the detail optimization can quickly complete the detailed route construction adjustment and the re-calling of the existing route construction performance, so as to increase the accuracy of the route integrity of the target route representation vector, and finally the initial logistics scheduling data and the target route representation vector are loaded into the logistics scheduling data construction network to construct the logistics scheduling data, and the target scheduling logistics scheduling data is obtained, so that the output target scheduling logistics scheduling data has an accurate route scheduling result.

[0052] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 It is a schematic diagram of the architecture of an application scenario provided by this application;

[0055] Figure 2 It is a flowchart of a big data processing method applied to smart logistics scheduling provided by this application;

[0056] Figure 3 It is a structural schematic diagram of a big data processing device provided in an embodiment of the present application;

[0057] Figure 4 It is a structural diagram of a big data processing system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0059] In order to facilitate a clearer understanding of the present application, the application scenarios of the big data processing method for smart logistics scheduling of the present application are first introduced, such as Figure 1 As shown, the application scenario includes a big data processing system 10 and a terminal cluster, and the terminal cluster may include one or more terminals, and the number of terminals is not limited here. Figure 1 As shown, the terminal cluster may specifically include terminal 1, terminal 2, ..., terminal n; it can be understood that terminal 1, terminal 2, terminal 3, ..., terminal n can all be connected to the big data processing system 10 through a network connection, so that each terminal can interact with the big data processing system 10 through a network connection.

[0060] It is understandable that the big data processing system 10 may refer to a device that performs logistics route optimization and adjustment, and the big data processing system 10 may also be used to store logistics big data. The big data processing system 10 may be a server or other electronic device, wherein the server may be an independent physical server, or a server cluster or distributed system composed of at least two physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms. The terminal may specifically refer to a vehicle-mounted terminal, a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a screen speaker, a smart watch, etc., but is not limited thereto. Each terminal and server may be directly or indirectly connected by wired or wireless communication, and the number of terminals and servers may be one or at least two, which is not limited in this application.

[0061] For further information, see Figure 2 , is a flow chart of a big data processing method for smart logistics scheduling provided in an embodiment of the present application. Figure 2 As shown, this method can be Figure 1 The big data processing method applied to smart logistics scheduling may include the following steps:

[0062] Step S110, obtaining initial logistics scheduling data and scheduling constraint information having target logistics route data, wherein the scheduling constraint information is used to indicate the construction of logistics scheduling data having the target logistics route data as logistics route data, and the scheduling constraint information contains route constraint conditions corresponding to the target route constraint level; the target route constraint level is one or more preset route constraint levels among multiple preset route constraint levels.

[0063] In this application, the initial logistics scheduling data includes data associated with logistics scheduling, specifically, it may include the category of transport vehicles, the attributes of goods, transportation prices, inventory data (such as warehouse location, inventory quantity) and logistics transportation routes. In the embodiment of the present application, the logistics scheduling data is represented as a matrix data for subsequent neural network processing. Specifically, each variable can be represented as a vector and then spliced ​​into a matrix data. The following is a specific example. Assuming that there are 6 transportation tasks, the category of the transport vehicle, the attributes of the goods, the transportation price, the warehouse location, the inventory quantity and the logistics transportation route are recorded respectively, for example as follows:

[0064] truck Fragile 100 A 50 A->B->C truck General cargo 80 B 70 B->A->C truck Dangerous Goods 120 C 60 C->B->A train General cargo 60 A 80 A->C train Fragile 80 B 80 B->C train Dangerous Goods 100 C 90 C->A

[0065] Among them, the first column represents the category of transport vehicles, the second column represents the attributes of goods, the third column represents the transportation price, the fourth column represents the warehouse location, the fifth column represents the inventory quantity, and the sixth column represents the logistics transportation route. It can be seen that the variables such as the category of transport vehicles, the attributes of goods, the warehouse location, and the logistics transportation route are non-numerical data, which cannot be recognized and calculated by the neural network. For these discrete data, they are converted into numerical data based on encoding, embedding, etc. For example, one-hot encoding is used to represent the attributes of goods and transportation types. For each cargo attribute, a binary vector of length 3 is created, where the i-th bit indicates whether the attribute is the i-th type of goods (1 for yes, 0 for no). For example, for the fragile attribute, the vector is [1,0,0], indicating that it is a fragile item. For each transportation type, a binary vector of length 2 is created, where the i-th bit indicates whether the attribute is the i-th type of transportation (1 for yes, 0 for no). For example, for the train attribute, the vector is [1,0], indicating that it is a train. For the warehouse location, it can be represented as a categorical variable, represented by a number or a string, for example, 1 represents Beijing and 2 represents Shanghai; for the logistics transportation route, it can be represented by two-dimensional coordinates (latitude and longitude), which can include the departure point, transit point and destination. The number of transit points is, for example, multiple. For example, an example is (39.9042°N, 116.4074°E)->(31.2304°N, 121.4737°E), represented as (39.9042, 116.4074, 31.2304, 121.4737). Then, after the encoding conversion, the new matrix obtained is, for example:

[0066]

[0067]

[0068] The target logistics route data is, for example, one of the logistics route data in the initial logistics scheduling data, such as (39.9042, 116.4074, 31.2304, 121.4737, 31.2304, 121.473) in the above example. In actual situations, the transit points involved are often more, forming a vector containing more elements. In the present application, the scheduling constraint information can be used to indicate the construction of logistics scheduling data with the target logistics route data as the logistics route data. The scheduling constraint information can be based on the existing route of the target logistics route data. The route constraint conditions to be constructed corresponding to the target route constraint level. The scheduling constraint information can be uploaded by the terminal. The target route constraint level is, for example, one or more preset route constraint levels among multiple preset route constraint levels. Specifically, the multiple preset route constraint levels include: transportation route toll constraints, transportation route type constraints, transportation route environmental constraints, transportation route smoothness constraints, etc. The scheduling constraint information can be represented by text or by a numerical label. The route constraint represents the specific constraint content under the constraint level of the corresponding target route. For example, the route constraint level contains two, namely, the transport route type constraint and the transport route smoothness constraint. When the target route constraint level is the transport route type constraint, the route constraint is, for example, to avoid taking the expressway. When the target route constraint level is the transport route smoothness constraint, the route constraint is, for example, to avoid traffic jams. For example, when expressed in text, the possible example is "avoid taking the expressway, avoid traffic jams". When expressed in numerical labels, the possible example is (01,03), where "01" means avoid taking the expressway and "03" means avoid traffic jams.

[0069] Step S120 , loading the scheduling constraint information into a pre-debugging knowledge representation mining network to perform pre-debugging knowledge representation mining to obtain the pre-debugging knowledge representation corresponding to the scheduling constraint information.

[0070] In this application, the pre-debugging knowledge representation mining network can be used to perform pre-debugging knowledge representation mining on the scheduling constraint information. The pre-debugging knowledge representation here is the prior knowledge obtained in advance, also known as expert knowledge, pre-trained features, etc. The pre-debugging knowledge representation is the pre-debugging knowledge representation of the route vector space obtained after the scheduling constraint information is extracted from the constraint vector and the vector space is mapped. Specifically, because the scheduling constraint information has the route constraint conditions corresponding to the target route constraint level, the pre-debugging knowledge representation can also represent the pre-debugging knowledge representation corresponding to the route constraint conditions.

[0071] In a feasible design, the pre-debugging knowledge is represented as a feature vector in the route vector space. In a feasible design, the pre-debugging knowledge representation mining network is a deep learning network, including a first information embedding mapping component (i.e., a network layer for embedding processing), a second information embedding mapping component, and a multimodal representation vector mapping network. The scheduling constraint information is loaded into the pre-debugging knowledge representation mining network for pre-debugging knowledge representation mining, and the pre-debugging knowledge representation corresponding to the scheduling constraint information is obtained, for example, including:

[0072] Step S101, loading the scheduling constraint information into the first information embedding mapping component to perform constraint condition feature mining to obtain a first constraint condition representation vector, the first information embedding mapping component is an information embedding mapping component that adopts multi-modal alignment pre-debugging.

[0073] In a feasible design, the first information embedding mapping component is used to extract constraint vectors from the scheduling constraint information, and the first constraint representation vector can be a constraint representation vector obtained after extracting the constraint vector from the scheduling constraint information and aligned with the vector in the route vector space.

[0074] In a feasible design, the first constraint characterization vector is a feature vector of a text space or a numerical space, depending on the specific form of the scheduling constraint information. The first information embedding mapping component is, for example, an information embedding mapping component in a pre-debugging (i.e., pre-training) multimodal alignment network. Here, the pre-debugging multimodal alignment network is, for example, obtained by pre-debugging the vector alignment of the preset information embedding mapping component and the preset scheduling data feature extraction component (i.e., the network layer for encoding scheduling data) in the preset multimodal alignment network. As a feasible design, for example, the preset information embedding mapping component and the preset scheduling data feature extraction component are vector-aligned and debugged based on metric learning. For example, the network structure of the first information embedding mapping component here is, for example, a transformer network.

[0075] Step S102, loading the scheduling constraint information into the second information embedding mapping component to perform constraint condition feature mining to obtain a second constraint condition representation vector, the second information embedding mapping component is an information embedding mapping component that uses constraint condition feature mining pre-debugging.

[0076] In a feasible design, the second information embedding mapping component can be used to extract the constraint vector of the scheduling constraint information. Correspondingly, the second constraint characterization vector is, for example, a constraint characterization vector obtained after the constraint vector of the scheduling constraint information is extracted. In a feasible design, the second constraint characterization vector is also, for example, a feature vector of a text space or a numerical space. The second information embedding mapping component is, for example, an information embedding mapping component that only uses constraint feature mining for pre-debugging. For example, the network structure of the second information embedding mapping component here is, for example, a converter network.

[0077] Step S103: Load the first constraint condition representation vector and the second constraint condition representation vector into a multimodal representation vector mapping network to perform multimodal representation vector mapping to obtain a pre-debugging knowledge representation.

[0078] In a feasible design, a multimodal representation vector mapping network can be used to map the first constraint representation vector and the second constraint representation vector from the text space to the route vector space. For example, the multimodal representation vector mapping network here is a generation network based on a diffusion mechanism, such as a converter network. In a feasible design, the concatenated vector obtained by concatenating the first constraint representation vector and the second constraint representation vector is loaded into the multimodal representation vector mapping network for multimodal representation vector mapping to obtain a pre-debugging knowledge representation.

[0079] This application performs constraint condition feature mining on scheduling constraint information based on two information embedding mapping components in the pre-debugging knowledge representation mining network, which can achieve a deeper understanding of the constraint information. In addition, based on the multimodal representation vector mapping network in the pre-debugging knowledge representation mining network, the constraint feature space is mapped to the route feature space, and the pre-debugging knowledge representation of the route vector space is used as the execution representation vector of the route representation vector construction network (that is, the loading data of the network) to construct the route representation vector, which can effectively alleviate the debugging complexity of the subsequent route representation vector construction network and improve the quality of logistics scheduling data construction.

[0080] Step S130, loading the pre-debugging knowledge representation into the route characterization vector construction network, performing route line adjustment based on the pre-debugging knowledge representation, and constructing a target route characterization vector corresponding to the target logistics route data; wherein the route characterization vector construction network is a multi-level route line adjustment network based on route constraint condition samples corresponding to multiple preset route constraint levels and corresponding to multiple preset route constraint levels, and the preset route characterization vector construction network is taught to perform route line adjustment and debugging at multiple preset route constraint levels.

[0081] In the present application, the target route characterization vector is, for example, a route characterization vector constructed after adjusting the route line according to the pre-debugging knowledge representation, and the target route characterization vector can be used for the target logistics route data in the initial logistics scheduling data. In a feasible design, when constructing the route characterization vector according to the pre-debugging knowledge representation, the route line can be adjusted according to the pre-debugging knowledge representation corresponding to the route constraint condition, so that the constructed target route characterization vector can better represent the route constraint condition. In a feasible design, the pre-debugging knowledge representation is loaded into the route characterization vector construction network, and the route line is adjusted according to the pre-debugging knowledge representation. Constructing the target route characterization vector corresponding to the target logistics route data includes: loading the pre-debugging knowledge representation and the preset enhanced knowledge representation into the route characterization vector construction network, and guiding the preset enhanced knowledge representation to complete the adjustment operation at the target route constraint level according to the pre-debugging knowledge representation (this operation can be understood as a denoising operation) to obtain the target route characterization vector.

[0082] In a feasible design, the key structure of the route representation vector construction network is, for example, a random walk model (Markovrandom walk model). In an optional design, the forward link of the random walk model, namely random propagation, can sequentially inject perturbation information (also known as data enhancement or noise) into the loaded data until the loaded data becomes similar to pure noise. In the reverse link, the enhanced data is perturbed and cleaned in sequence based on the constraints to construct denoised data so that the denoised data meets the route constraints.

[0083] Because the scheduling constraint information has the route constraint conditions corresponding to the target route constraint level, the pre-debugging knowledge representation can also represent the pre-debugging knowledge representation corresponding to the route constraint conditions, and the pre-debugging knowledge representation is loaded into the route representation vector construction network. According to the pre-debugging knowledge representation, the preset enhanced knowledge representation is guided to perform adjustment operations on the target route constraint level, and the target route representation vector corresponding to the target logistics route data is obtained, which can increase the accuracy of the route integrity of the target route representation vector. The embodiment of the present application is based on route constraint condition samples (samples, i.e., training samples) corresponding to multiple preset route constraint levels and multi-level route line adjustment networks corresponding to multiple preset route constraint levels, and guides the preset route representation vector construction network to perform route line adjustment and debugging at multiple preset route constraint levels, that is, to optimize the preset route representation vector construction network in detail, so that the route representation vector construction network obtained after the detail optimization can quickly complete the detailed route construction adjustment and the re-calling of the existing route construction performance.

[0084] In a feasible design, the route representation vector construction network can be debugged based on the following methods:

[0085] Step S201 , clone the network internal variables of the preset route representation vector construction network to obtain a first route representation vector construction network and a second route representation vector construction network, wherein the first route representation vector construction network and the second route representation vector construction network are at the same level.

[0086] The first route characterization vector construction network and the second route characterization vector construction network are at the same level, indicating that the first route characterization vector construction network and the second route characterization vector construction network are connected in parallel. In a feasible design, the preset route characterization vector construction network is, for example, a basic route characterization vector construction network for pre-debugging of route characterization vector construction, and the basic route characterization vector construction network includes basic route characterization vector construction performance. As a feasible design, the preset route characterization vector construction network is, for example, any feasible neural network, such as a potential random walk model), a DALL·E model, and the preset route characterization vector construction network is, for example, a converter network. In a feasible design, the preset route characterization vector construction network is cloned for internal network variables to obtain a first route characterization vector construction network and a second route characterization vector construction network, and the first route characterization vector construction network and the second route characterization vector construction network are at the same level, including: S201a, cloning the internal network variables of some components of the preset route characterization vector construction network to obtain the first route characterization vector construction network; S201b, cloning the internal network variables of all components of the preset route characterization vector construction network to obtain the second route characterization vector construction network; S201c, connecting the first route characterization vector construction network and the second route characterization vector construction network in parallel.

[0087] In a feasible design, the first route characterization vector construction network is, for example, a network obtained by cloning the network internal variables (i.e., parameters of the construction network, such as weights, biases, etc.) of the local architecture of the preset route characterization vector construction network. In the embodiment of the present application, based on the route line adjustment and debugging of the first route characterization vector construction network, the details of the preset route characterization vector construction network are optimized to achieve fine-tuning after pre-debugging. In a feasible design, the second route characterization vector construction network is, for example, a network obtained by cloning the network internal variables of the overall network structure of the preset route characterization vector construction network. In the embodiment of the present application, the second route characterization vector construction network is used to maintain the existing route construction capability of the preset route characterization vector construction network.

[0088] Step S202: Acquire a first pre-debugging knowledge representation sample and an enhanced knowledge representation sample corresponding to a first scheduling constraint information sample having a route constraint condition sample.

[0089] Here, the process of acquiring the first pre-debugging knowledge representation example refers to the process of extracting the pre-debugging knowledge representation in steps S101 to S103, for example.

[0090] In an optional design, the enhanced knowledge representation sample is, for example, a feature map corresponding to noise information whose probability density function of the perturbation information distribution conforms to the Gaussian distribution. Alternatively, in another design, the enhanced knowledge representation sample is, for example, obtained by mining the data representation vector of the enhanced data of the first scheduling data sample, where the first scheduling data sample is, for example, a scheduling data sample aligned with the first scheduling constraint information sample.

[0091] Step S203, when adjusting the first route representation vector construction network and the second route representation vector construction network to construct the route representation vector according to the first pre-debugging knowledge representation sample, the enhanced knowledge representation sample and the preset subsidiary enhancement information corresponding to the enhanced knowledge representation sample, adding a route constraint condition sample to the first route representation vector construction network according to the multi-level route line adjustment network, so that the first route representation vector construction network performs route line adjustment and debugging to obtain a first debugging network.

[0092] In a feasible design, the preset auxiliary enhancement information is added enhancement information, such as random Gaussian enhancement information. In a feasible design, the multi-level route line adjustment network can be used to extract the representation vector of the route constraint condition sample, and add the obtained route constraint condition representation vector sample to the first route representation vector construction network.

[0093] In a feasible design, the first debugging network is, for example, a network obtained by performing route line adjustment debugging on the first route characterization vector construction network. When the first route characterization vector construction network is performing route line adjustment debugging, the network internal variables of the second route characterization vector construction network are locked and do not need to be optimized, so that the construction effect of the existing network can be effectively retained.

[0094] In a feasible design, in the embodiment of the present application, the implementation method of the first route characterization vector construction network and the second route characterization vector construction network at the same level can refer to the following description:

[0095] The first route representation vector construction network includes a first embedding mapping component, and the second route representation vector construction network includes a second embedding mapping component and a restoration mapping component.

[0096] In a feasible design, when adjusting the first route representation vector construction network and the second route representation vector construction network to construct the route representation vector according to the first pre-debugging knowledge representation sample, the enhanced knowledge representation sample and the preset subsidiary enhancement information corresponding to the enhanced knowledge representation sample, adding a route constraint condition sample to the first route representation vector construction network according to the multi-level route line adjustment network, so that the first route representation vector construction network performs route line adjustment and debugging, obtaining the first debugging network may include:

[0097] Step S301, loading route constraint condition samples corresponding to a plurality of preset route constraint levels into a multi-level route line adjustment network to perform route constraint condition representation vector mining to obtain route constraint condition representation vector samples.

[0098] For example, the route constraint condition representation vector sample is, for example, a feature obtained after mining the route constraint condition representation vectors of the route constraint condition samples corresponding to multiple preset route constraint levels.

[0099] In a feasible design, a multi-level route regulation network may include route regulation components corresponding to multiple preset route constraint levels, and the route regulation component corresponding to each preset route constraint level may be used to perform route constraint condition representation vector mining on route constraint condition samples of the corresponding preset route constraint level.

[0100] In a feasible design, route constraint condition samples corresponding to multiple preset route constraint levels are loaded into a multi-level route line adjustment network to perform route constraint condition representation vector mining, and route constraint condition representation vector samples are obtained, including: S401, loading route constraint condition samples corresponding to each preset route constraint level in multiple preset route constraint levels into the route adjustment component of the corresponding preset route constraint level to perform route constraint condition representation vector mining, and obtain the target route constraint condition representation vector corresponding to each preset route constraint level; S402, eccentrically adjusting and integrating the target route constraint condition representation vectors corresponding to multiple preset route constraint levels, and obtaining route constraint condition representation vector samples.

[0101] For example, the eccentricity coefficients corresponding to the multiple preset route constraint levels are first determined, that is, a weight value, and the target route constraint condition representation vectors corresponding to the multiple preset route constraint levels are eccentrically adjusted and integrated according to the weight value (that is, weighted and then the weighted results are summed up) to obtain a sample of the route constraint condition representation vector. The weight values ​​corresponding to the multiple preset route constraint levels can be adaptively configured in combination with the influence of the multiple preset route constraint levels in the process of constructing the route representation vector.

[0102] Based on this, the route constraint condition samples corresponding to each preset route constraint level in multiple preset route constraint levels are loaded into the route adjustment component of the corresponding preset route constraint level to perform route constraint condition representation vector mining, obtain the target route constraint condition representation vector corresponding to each preset route constraint level, and perform eccentric adjustment and integration on the target route constraint condition representation vectors corresponding to the multiple preset route constraint levels to obtain route constraint condition representation vector samples, which can increase the accuracy of each target route constraint condition representation vector in representing the route details of the corresponding preset route constraint level.

[0103] In a feasible design, the route adjustment component corresponding to each preset route constraint level includes a coordinate inverse combination operator and a cascaded residual representation vector extraction operator of multiple feature granularities (the granularity here refers to the scale of the data, and one scale corresponds to a data extraction density. The larger the granularity, the denser the data, and the smaller the granularity, the sparser the data). The target route constraint condition representation vector corresponding to each preset route constraint level may include a residual route constraint condition representation vector of multiple feature granularities. For example, suppose the route adjustment component n corresponding to the n-th preset route constraint level, when the residual representation vector extraction operator of multiple feature granularities is a residual representation vector extraction operator of three-level feature granularity, the target route constraint condition representation vector corresponding to the n-th preset route constraint level includes residual route constraint condition representation vectors corresponding to three feature granularities.

[0104] As a feasible design, the residual representation vector extraction operators of adjacent feature granularities in the residual representation vector extraction operators of multiple feature granularities are connected based on the pooling operator. In the feasible design, the route constraint condition samples corresponding to each preset route constraint level in multiple preset route constraint levels are loaded into the route adjustment component of the corresponding preset route constraint level to perform route constraint condition representation vector mining, and the target route constraint condition representation vector corresponding to each preset route constraint level is obtained, including:

[0105] S401a, load the route constraint condition samples corresponding to each preset route constraint level into the coordinate decombination operator corresponding to the preset route constraint level for coordinate decombination processing, and obtain the initial constraint condition representation vector corresponding to each preset route constraint level. Coordinate decombination processing is to perform a reverse reorganization of the input features to generate new features. Specifically, by performing convolution operations and channel expansion on the input features, and then performing reshaping operations, features with richer information are generated. In an example, assuming that the input is a feature map with a size of W*H*C, the input size is converted to W*H*4C through a convolution operation, and then the feature map size is converted to 2W*2H*C through a reshaping operation. The coordinate decombination operator can be used to perform coordinate decombination processing on the route constraint condition samples. The initial constraint condition representation vector corresponding to each preset route constraint level is, for example, a feature obtained after performing coordinate decombination processing on the route constraint condition samples corresponding to each preset route constraint level, so as to reduce the amount of network calculation.

[0106] S401b, based on the residual representation vector extraction operator of multi-feature granularity corresponding to each preset route constraint level, perform multi-feature granularity residual representation vector extraction on the initial constraint condition representation vector to obtain the multi-feature granularity residual route constraint condition representation vector corresponding to each preset route constraint level.

[0107] Specifically, the multi-feature granularity residual representation vector extraction operator can be used to perform multi-feature granularity residual representation vector extraction on the initial constraint condition representation vector. The multi-feature granularity residual route constraint condition representation vector is, for example, the multi-feature granularity feature obtained after performing multi-feature granularity residual representation vector extraction on the initial constraint condition representation vector.

[0108] As a feasible design, the residual representation vector extraction operator of each feature granularity in the residual representation vector extraction operator of multiple feature granularities includes a first convolution operator and two residual representation vector extraction units. The residual representation vector extraction unit is, for example, a unit of a residual structure, so that the operator can propagate information across layers.

[0109] Examples of route constraint condition representation vectors include route constraint condition representation vectors of multiple feature granularities. The target route constraint condition representation vectors corresponding to multiple preset route constraint levels are eccentrically adjusted and integrated to obtain route constraint condition representation vector examples including:

[0110] S402a, performing eccentric adjustment and integration on the residual route constraint condition representation vectors of each characteristic granularity corresponding to the plurality of preset route constraint levels, to obtain the route constraint condition representation vector of each characteristic granularity.

[0111] The residual route constraint condition representation vectors of each feature granularity corresponding to multiple preset route constraint levels are respectively eccentrically adjusted and integrated to obtain the route constraint condition representation vector of each feature granularity, including: the residual route constraint condition representation vectors of the target feature granularity corresponding to multiple preset route constraint levels are respectively eccentrically adjusted and integrated to obtain the route constraint condition representation vector of the target feature granularity, wherein the target feature granularity is, for example, one feature granularity among multiple feature granularities.

[0112] Based on this, the residual representation vector extraction operator based on multiple feature granularities is used to extract the residual route constraint condition representation vector of multiple feature granularities corresponding to each preset route constraint level, and the residual route constraint condition representation vector of each feature granularity corresponding to multiple preset route constraint levels is eccentrically adjusted and integrated to obtain the route constraint condition representation vector of each feature granularity. Subsequently, the route constraint condition representation vectors of multiple levels of feature granularity are loaded into the first pooling operator of the feature granularity corresponding to the first embedding mapping component, and the first embedding mapping component is taught to perform route line adjustment and debugging of multiple feature granularities to increase the debugging effect of the route line adjustment and debugging of the first embedding mapping component.

[0113] Step S302, load the enhanced knowledge representation sample, the first pre-debugging knowledge representation sample and the route constraint condition representation vector sample into the first embedding mapping component, embed and map the enhanced knowledge representation sample according to the first pre-debugging knowledge representation sample and the route constraint condition representation vector sample, and obtain the first embedding mapping representation vector.

[0114] The first embedding mapping representation vector is, for example, a feature obtained after embedding and mapping the enhanced knowledge representation sample based on the first pre-debugging knowledge representation sample and the route constraint condition representation vector sample.

[0115] According to the first pre-debugging knowledge representation sample and the route constraint condition representation vector sample, the enhanced knowledge representation sample is embedded and mapped to obtain a first embedded mapping representation vector, including: according to the concatenation features of the first pre-debugging knowledge representation sample and the route constraint condition representation vector sample, the enhanced knowledge representation sample is embedded and mapped to obtain a first embedded mapping representation vector.

[0116] Step S303: load the enhanced knowledge representation sample and the first pre-debugging knowledge representation sample into the second embedding mapping component, and perform embedding mapping on the enhanced knowledge representation sample according to the first pre-debugging knowledge representation sample to obtain a second embedding mapping representation vector.

[0117] The second embedding mapping representation vector is, for example, a feature obtained after embedding and mapping the enhanced knowledge representation sample according to the first pre-debugging knowledge representation sample.

[0118] Step S304, loading the first embedded mapping representation vector, the second embedded mapping representation vector and the first pre-debugging knowledge representation sample into the restoration mapping component, and performing restoration mapping on the first embedded mapping representation vector and the second embedded mapping representation vector according to the first pre-debugging knowledge representation sample to obtain a route representation vector sample.

[0119] The route representation vector sample is, for example, a feature obtained after restoring and mapping the first embedded mapping representation vector and the second embedded mapping representation vector according to the first pre-debugging knowledge representation sample. Performing restoration mapping on the first embedded mapping representation vector and the second embedded mapping representation vector according to the first pre-debugging knowledge representation sample to obtain the route representation vector sample includes: performing restoration mapping on the concatenation feature of the first embedded mapping representation vector and the second embedded mapping representation vector according to the first pre-debugging knowledge representation sample to obtain the route representation vector sample.

[0120] In a feasible design, the first embedding mapping component includes a cascaded first pooling operator of multi-feature granularity, the route constraint representation vector sample includes a route constraint representation vector of multi-feature granularity, the first embedding mapping representation vector includes a first pooling representation vector of multi-feature granularity, the second embedding mapping component includes a cascaded second pooling operator of multi-feature granularity, the second embedding mapping representation vector includes a second pooling representation vector of multi-feature granularity, and the restoration mapping component includes a cascaded unpooling operator of multi-feature granularity.

[0121] Assume that the multi-feature granularity includes three types of feature granularities. In another parallel structure of a first route representation vector construction network and a second route representation vector construction network in an embodiment of the present application, in a feasible design, the enhanced knowledge representation sample, the first pre-debugging knowledge representation sample and the route constraint condition representation vector sample are loaded into a first embedding mapping component, and the enhanced knowledge representation sample is embedded and mapped according to the first pre-debugging knowledge representation sample and the route constraint condition representation vector sample to obtain the first embedding mapping representation vector, including: S302a, based on the first pooling operator of each feature granularity in the multi-feature granularity, according to the first pre-debugging knowledge representation sample, the execution representation vector of the first pooling operator of each feature granularity is pooled and extracted to obtain each The first pooling representation vector of the feature granularity, wherein the execution representation vector of the first pooling operator of the first feature granularity is the enhanced knowledge representation sample, and the execution representation vector of the first pooling operator from the second feature granularity to the last feature granularity is the first pooling representation vector of the previous feature granularity itself and the route constraint representation vector of the previous feature granularity itself; the above-mentioned loading the enhanced knowledge representation sample and the first pre-debugging knowledge representation sample into the second embedding mapping component, embedding and mapping the enhanced knowledge representation sample according to the first pre-debugging knowledge representation sample, and obtaining the second embedding mapping representation vector includes: S303b, based on the second pooling operator of each feature granularity, according to the first pre-debugging knowledge representation sample, embedding and mapping the second pooling operator of each feature granularity The execution representation vector of the first feature granularity is pooled and the second pooled representation vector of each feature granularity is extracted, wherein the execution representation vector of the second pooling operator of the first feature granularity is the enhanced knowledge representation sample, and the execution representation vector of the second pooling operator from the second feature granularity to the last feature granularity is the second pooled representation vector of the previous feature granularity itself; the first embedding mapping representation vector, the second embedding mapping representation vector and the first pre-debugging knowledge representation sample are loaded into the restoration mapping component, and the first embedding mapping representation vector and the second embedding mapping representation vector are restored and mapped according to the first pre-debugging knowledge representation sample to obtain the route representation vector sample, including: S304a, based on the reverse pooling operator of each feature granularity, according to the first pre-debugging knowledge representation sample Try the knowledge representation example, extract the de-pooling representation vector of the execution representation vector of the de-pooling operator of each feature granularity, and obtain the de-pooling representation vector of each feature granularity, wherein the execution representation vector of the de-pooling operator of the last feature granularity connected to the second pooling operator of the last feature granularity is: the second pooling representation vector of the last feature granularity, the first pooling representation vector of the same feature granularity as itself, and the route constraint representation vector of the same feature granularity as itself, and the execution representation vector of the de-pooling operator from the first feature granularity to the last second feature granularity includes: the de-pooling representation vector of the next feature granularity itself, the first pooling representation vector of the same feature granularity as itself, and the route constraint representation vector of the same feature granularity as itself;S304b, using the de-pooling representation vector output by the de-pooling operator of the first feature granularity as a route representation vector sample. ;

[0122] Based on this, based on the same-level implementation design provided in this embodiment, a route constraint condition representation vector with multiple feature granularity is added to the first embedded mapping component, so that the first embedded mapping component can accurately adjust the route details on the premise of learning how to represent the route details with multiple feature granularity. Based on the second embedded mapping component and the restoration mapping component, the existing route construction effect of the preset route representation vector construction network is maintained, so that the route representation vector construction network with optimized details can quickly complete the detailed route construction adjustment and the re-call of the existing route construction performance.

[0123] Step S305, debugging the first embedding mapping component to obtain a first debugging network according to a first enhancement error between the first enhancement information sample and the preset subsidiary enhancement information; the first enhancement information sample is the enhancement information between the enhancement knowledge representation sample and the route representation vector sample.

[0124] In a feasible design, the first enhancement error represents an error between the first enhancement information sample and the preset subsidiary enhancement information. In a feasible design, the first enhancement error between the first enhancement information sample and the preset subsidiary enhancement information is determined according to a preset error function, for example, the preset error function is a cross entropy error function.

[0125] In a feasible design, based on the first enhancement error between the first enhancement information sample and the preset subsidiary enhancement information, debugging the first embedded mapping component to obtain the first debugging network includes: optimizing the network internal variables of the first embedded mapping component based on the first enhancement error to obtain the optimized first embedded mapping component; based on the optimized first embedded mapping component, returning to step S301 until the network converges; and using the first embedded mapping component at the time of convergence as the first debugging network.

[0126] As a feasible design, the method also includes: debugging the multi-level route line adjustment network and the first embedded mapping component according to the first enhancement error between the first enhancement information sample and the preset subsidiary enhancement information, and obtaining the debugged multi-level route line adjustment network and the first debugging network.

[0127] In a feasible design, based on the first enhancement error between the first enhancement information sample and the preset subsidiary enhancement information, the multi-level route line adjustment network and the first embedded mapping component are debugged to obtain the debugged multi-level route line adjustment network and the first debugged network, including: based on the first enhancement error, optimizing the network internal variables of the multi-level route line adjustment network and the network internal variables of the first embedded mapping component to obtain the optimized multi-level route line adjustment network and the optimized first embedded mapping component; based on the optimized multi-level route line adjustment network and the optimized first embedded mapping component, returning to step S301 until the network converges, and determining the converged multi-level route line adjustment network and the first embedded mapping component as the debugged multi-level route line adjustment network and the first debugged network.

[0128] Based on this, when performing detail optimization (route line adjustment and debugging) on ​​the preset route representation vector construction network, only some network internal variables of the embedded mapping component need to be learned and optimized, and other network internal variables are locked, thereby increasing the network's detail optimization speed while maintaining the network's existing construction performance.

[0129] Step S204 , the network internal variables of the first debugging network and the network internal variables of the second route representation vector construction network are sequentially merged (ie, the variables are spliced ​​layer by layer according to the network layer order) to obtain the target network internal variables.

[0130] Step S205 , optimizing the network internal variables of the preset route representation vector construction network according to the target network internal variables, and obtaining the route representation vector construction network.

[0131] In a feasible design, the above-mentioned preset route representation vector construction network includes a preset embedding mapping component (also called an encoding component) and a preset restoration mapping component (also called a decoding component), the first route representation vector construction network includes a first embedding mapping component, the second route representation vector construction network includes a second embedding mapping component and a restoration mapping component, the first debugging network includes a first debugging embedded mapping component, the target network internal variables include embedding mapping component parameters and restoration mapping component parameters, and the network internal variables of the first debugging network and the network internal variables of the second route representation vector construction network are sequentially merged to obtain the target network internal variables, including: S204a, splicing the network internal variables of the first debugging embedded mapping component and the network internal variables of the second embedded mapping component to obtain the embedding mapping component parameters; S204b, using the network internal variables of the restoration mapping component as restoration mapping component parameters.

[0132] According to the internal variables of the target network, the network internal variables of the preset route representation vector construction network are optimized, and obtaining the route representation vector construction network includes: according to the embedding mapping component parameters and the restoration mapping component parameters, respectively optimizing the network internal variables of the preset embedding mapping component and the network internal variables of the preset restoration mapping component, and obtaining the route representation vector construction network.

[0133] Based on this, based on the multi-level route adjustment network, the preset route representation vector construction network is guided to perform route adjustment and debugging at multiple preset route constraint levels, that is, the preset route representation vector construction network is optimized in detail, so that the route representation vector construction network obtained after the detail optimization can quickly complete the detailed route construction adjustment and the re-call of the existing route construction performance, so as to increase the accuracy of the route integrity of the target route representation vector. At the same time, the route representation vector construction network after the detail optimization can directly adjust the route according to the route constraint condition representation vector contained in the pre-debugging knowledge representation in the application link, without the need for the multi-level route adjustment network to coordinate the addition of route constraints. In the application link, the network structure can be simplified, the computing power overhead can be reduced, and the cost can be saved.

[0134] Step S140, loading the initial logistics scheduling data and the target route characterization vector into the logistics scheduling data construction network to construct the logistics scheduling data and obtain the target scheduling logistics scheduling data.

[0135] The target scheduling logistics scheduling data is, for example, the logistics scheduling data constructed by fusing the initial logistics scheduling data and the target route representation vector. As a feasible design, the target scheduling logistics scheduling data is, for example, the logistics scheduling data having the same data density (scale) as the initial logistics scheduling data. It can be understood that the process of constructing the logistics scheduling data is to construct the logistics route therein, such as adding, deleting, and changing the transit point therein (i.e., the corresponding two-dimensional coordinates) to meet the route constraints.

[0136] In a feasible design, the logistics scheduling data construction network includes a scheduling data feature extraction component, a representation vector integration component, and a data restoration mapping component. The initial logistics scheduling data and the target route representation vector are loaded into the logistics scheduling data construction network to construct the logistics scheduling data. Obtaining the target scheduling logistics scheduling data may include:

[0137] Step S141, loading the initial logistics scheduling data into the scheduling data feature extraction component to perform data representation vector mining to obtain the initial logistics scheduling data representation vector.

[0138] The initial logistics scheduling data representation vector represents the route vector corresponding to the initial logistics scheduling data. In a feasible design, the network structure of the scheduling data feature extraction component is, for example, a variational autoencoder. As a feasible design, the scheduling data feature extraction component is, for example, an embedding mapping component of multiple feature granularities.

[0139] Step S142, loading the initial logistics scheduling data representation vector and the target route representation vector into the representation vector integration component to perform a representation vector integration operation to obtain an integrated logistics scheduling data representation vector.

[0140] The integrated logistics scheduling data characterization vector is, for example, a feature obtained after performing a characterization vector integration operation on the initial logistics scheduling data characterization vector and the target route characterization vector. In a feasible design, the integrated logistics scheduling data characterization vector is, for example, an integrated logistics scheduling data characterization vector distribution. For example, the integrated logistics scheduling data characterization vector distribution includes the distribution mean and variance of the integrated logistics scheduling data characterization vector.

[0141] As a feasible design, the initial logistics scheduling data representation vector and the target route representation vector are loaded into the representation vector integration component for representation vector integration operation, and obtaining the integrated logistics scheduling data representation vector may include: splicing the initial logistics scheduling data representation vector and the target route representation vector to obtain the integrated logistics scheduling data representation vector. As a feasible design, the initial logistics scheduling data representation vector and the target route representation vector are loaded into the representation vector integration component for representation vector integration operation, and obtaining the integrated logistics scheduling data representation vector may include: performing vector integration based on the attention mechanism on the initial logistics scheduling data representation vector and the target route representation vector to obtain the integrated logistics scheduling data representation vector.

[0142] S143, loading the integrated logistics scheduling data representation vector into the data restoration mapping component for data restoration mapping to obtain target scheduling logistics scheduling data.

[0143] In a feasible design, the data restoration mapping component can be used to perform data restoration mapping on the integrated logistics scheduling data representation vector. For example, the network structure of the data restoration mapping component is, for example, a decoder in a variational autoencoder. As a feasible design, the data restoration mapping component is, for example, a restoration mapping component of multiple feature granularity.

[0144] As a feasible design, the scheduling data feature extraction component, the representation vector integration component and the data restoration mapping component are debugged collaboratively or separately.

[0145] Based on this, because of the previous separation operation on the constraint information and logistics scheduling data, the initial logistics scheduling data representation vector and the target route representation vector corresponding to the initial logistics scheduling data are integrated to obtain the integrated logistics scheduling data representation vector, and then data restoration mapping is performed based on the integrated logistics scheduling data representation vector to obtain the target scheduling logistics scheduling data, so that the output target scheduling logistics scheduling data maintains the existing route characteristics of the initial logistics scheduling data on the basis of including accurate route details.

[0146] As a feasible design, the method may also include:

[0147] Step S210, obtain an initial logistics scheduling data sample having a logistics route data sample, a second scheduling constraint information sample having a route constraint condition sample, and a logistics scheduling data representation vector sample corresponding to the scheduling logistics scheduling data sample matching the second scheduling constraint information sample, wherein the second scheduling constraint information sample is used to indicate the construction of logistics scheduling data with the logistics route data sample as the logistics route data.

[0148] In a feasible design, the scheduling logistics scheduling data sample is, for example, the scheduling data sample obtained after performing graphic matching on the second scheduling constraint information sample, and the logistics scheduling data representation vector sample corresponding to the scheduling logistics scheduling data sample is, for example, the route vector obtained after performing data representation vector mining on the scheduling logistics scheduling data sample.

[0149] As a feasible design, the pre-debugging knowledge representation mining network includes a first information embedding mapping component, which is, for example, an information embedding mapping component in a pre-debugging multimodal alignment network. The logistics scheduling data representation vector samples corresponding to the scheduling logistics scheduling data samples can be obtained by mining the data representation vectors of the scheduling logistics scheduling data samples based on the scheduling data feature extraction component in the pre-debugging multimodal alignment network.

[0150] Step S220, performing pre-debugging knowledge representation mining debugging on a preset pre-debugging knowledge representation mining network according to the second scheduling constraint information sample and the logistics scheduling data representation vector sample, to obtain a pre-debugging knowledge representation mining network.

[0151] In a feasible design, the preset pre-debugging knowledge representation mining network includes a first preset information embedding mapping component, a second preset information embedding mapping component and a preset multimodal representation vector mapping network. The first preset information embedding mapping component is, for example, an information embedding mapping component pre-debugging with multimodal alignment, and the second preset information embedding mapping component is, for example, an information embedding mapping component pre-debugging with constraint feature mining. Accordingly, the preset pre-debugging knowledge representation mining network is pre-debugging knowledge representation mining and debugging is performed based on the second scheduling constraint information sample and the logistics scheduling data representation vector sample. Obtaining the pre-debugging knowledge representation mining network includes: S1, loading the second scheduling constraint information sample into the first preset information embedding mapping component for constraint feature mining to obtain a first sample constraint representation vector; S2, loading the second scheduling constraint information sample into the first The second preset information embedding mapping component performs constraint feature mining to obtain a second sample constraint representation vector; S3, loads the first sample constraint representation vector and the second sample constraint representation vector into a preset multimodal representation vector mapping network for multimodal representation vector mapping to obtain a prediction pre-debugging knowledge representation; S4, debugs the preset multimodal representation vector mapping network based on the pre-trained knowledge representation error between the logistics scheduling data representation vector sample and the prediction pre-debugging knowledge representation to obtain a multimodal representation vector mapping network; S5, uses the first preset information embedding mapping component as the first information embedding mapping component; uses the second preset information embedding mapping component as the second information embedding mapping component; S6, obtains a pre-debugging knowledge representation mining network based on the first information embedding mapping component, the second information embedding mapping component and the multimodal representation vector mapping network.

[0152] Step S230 , determining a second pre-debugging knowledge representation example corresponding to the second scheduling constraint information example based on the pre-debugging knowledge representation mining network.

[0153] The process of obtaining the second pre-debugging knowledge representation sample here may refer to the process of extracting the pre-debugging knowledge representation in steps S101 to S103.

[0154] Step S240, based on the route constraint condition sample, the second pre-debugging knowledge representation sample, the initial logistics scheduling data sample and the scheduled logistics scheduling data sample, the preset route characterization vector construction network, the multi-level route line adjustment network and the preset neural network are collaboratively debugged to obtain the route characterization vector construction network, the debugged multi-level route line adjustment network and the logistics scheduling data construction network.

[0155] In a feasible design, based on the route constraint condition sample, the second pre-debugging knowledge representation sample, the initial logistics scheduling data sample and the scheduling logistics scheduling data sample, the preset route representation vector construction network, the multi-level route line adjustment network and the preset neural network are collaboratively debugged to obtain the route representation vector construction network, the debugged multi-level route line adjustment network and the logistics scheduling data construction network, including: S1, cloning the network internal variables of the preset route representation vector construction network to obtain the first route representation vector construction network and the second route representation vector construction network, the first route representation vector construction network and the second route representation vector construction network are at the same level; S2, when adjusting the first route representation vector construction network and the second route representation vector construction network to construct the route representation vector according to the second pre-debugging knowledge representation sample, the enhanced knowledge representation sample and the preset subsidiary enhancement information corresponding to the enhanced knowledge representation sample, According to the multi-level route regulation network, route constraint condition samples are added to the first route characterization vector construction network to obtain a predicted route characterization vector; S3, the initial logistics scheduling data sample and the predicted route characterization vector are loaded into the preset neural network to construct the logistics scheduling data to obtain the predicted logistics scheduling data; S4, according to the loss information between the predicted logistics scheduling data and the scheduled logistics scheduling data samples, the first route characterization vector construction network, the multi-level route regulation network and the preset neural network are debugged to obtain the first debugged network, the debugged multi-level route regulation network and the logistics scheduling data construction network; S5, the network internal variables of the first debugged network and the network internal variables of the second route characterization vector construction network are sequentially merged to obtain the target network internal variables; S6, according to the target network internal variables, the network internal variables of the preset route characterization vector construction network are optimized to obtain the route characterization vector construction network.

[0156] Based on this, on the basis of obtaining the debugged pre-debugging knowledge representation mining network, the preset route representation vector construction network, the multi-level route line adjustment network and the preset neural network are collaboratively debugged. The debugging samples required for the second step can be collected according to the network debugged in the first step, thereby increasing the sample collection speed and improving the network debugging speed.

[0157] In summary, based on obtaining the initial logistics scheduling data and scheduling constraint information with the target logistics route data, the scheduling constraint information is used to indicate the construction of logistics scheduling data with the target logistics route data as the logistics route data, and the information of the scheduling constraint information has route constraints corresponding to the target route constraint level, and the target route constraint level is, for example, one or more preset route constraint levels among multiple preset route constraint levels. Then, the scheduling constraint information is loaded into the pre-debugging knowledge representation mining network for pre-debugging knowledge representation mining to obtain the pre-debugging knowledge representation corresponding to the scheduling constraint information. Because the scheduling constraint information has the route constraints corresponding to the target route constraint level, the pre-debugging knowledge representation can also represent the pre-debugging knowledge representation corresponding to the route constraints. At the same time, the pre-debugging knowledge representation mining network is based on realizing the projection of the constraint feature space to the route feature space, and the process of building the route based on the constraint vector is refined into two links from the constraint vector to the route vector and then to the route construction, which can effectively alleviate the debugging complexity of the subsequent route representation vector construction network and increase the quality of logistics scheduling data construction. Then, the pre-debugging knowledge representation is loaded into the pre-debugging knowledge representation mining network. A route characterization vector is constructed to adjust the route according to the pre-debugging knowledge representation, and a target route characterization vector corresponding to the target logistics route data is constructed, wherein the route characterization vector construction network is a multi-level route line adjustment network corresponding to the route constraint condition samples corresponding to multiple preset route constraint levels and the multiple preset route constraint levels, and the preset route characterization vector construction network is taught to perform route line adjustment and debugging at multiple preset route constraint levels, and based on the multi-level route line adjustment network, the preset route characterization vector construction network is guided to perform route line adjustment and debugging at multiple preset route constraint levels, that is, the preset route characterization vector construction network is optimized in detail, so that the route characterization vector construction network obtained after the detail optimization can quickly complete the detailed route construction adjustment and the re-calling of the existing route construction performance, so as to increase the accuracy of the route integrity of the target route characterization vector, and finally the initial logistics scheduling data and the target route characterization vector are loaded into the logistics scheduling data construction network to construct the logistics scheduling data, and the target scheduling logistics scheduling data is obtained, so that the output target scheduling logistics scheduling data has an accurate route scheduling result.

[0158] See also Figure 3 , is a schematic diagram of the structure of a big data processing device provided in an embodiment of the present application. The above-mentioned big data processing device can be a computer program (including program code) running in a network device, for example, the big data processing device is an application software; the device can be used to execute the corresponding steps of the method provided in an embodiment of the present application. Figure 3 As shown, the big data processing device may include: a data acquisition module 310, a priori mining module 320, a line adjustment module 330, and a data construction module 340.

[0159] The data acquisition module 310 is used to acquire initial logistics scheduling data and scheduling constraint information having target logistics route data, wherein the scheduling constraint information is used to indicate the construction of logistics scheduling data using the target logistics route data as logistics route data, and the scheduling constraint information includes route constraint conditions corresponding to the target route constraint level;

[0160] A priori mining module 320 is used to load the scheduling constraint information into a pre-debugging knowledge representation mining network to perform pre-debugging knowledge representation mining to obtain a pre-debugging knowledge representation corresponding to the scheduling constraint information;

[0161] A route adjustment module 330 is used to load the pre-commissioning knowledge representation into a route representation vector construction network, perform route adjustment according to the pre-commissioning knowledge representation, and construct a target route representation vector corresponding to the target logistics route data;

[0162] A data construction module 340 is used to load the initial logistics scheduling data and the target route representation vector into a logistics scheduling data construction network to construct logistics scheduling data and obtain target scheduling logistics scheduling data;

[0163] Among them, the route characterization vector construction network is based on route constraint condition samples corresponding to multiple preset route constraint levels and a multi-level route line adjustment network corresponding to the multiple preset route constraint levels, and the preset route characterization vector construction network is taught to perform route line adjustment and debugging of the multiple preset route constraint levels; the target route constraint level is one or more preset route constraint levels among the multiple preset route constraint levels.

[0164] According to one embodiment of the present application, Figure 2 The steps involved in the big data processing method for smart logistics scheduling shown in FIG. Figure 3 It is executed by the various modules in the big data processing device shown.

[0165] According to one embodiment of the present application, Figure 3 The various modules in the shown big data processing device can be separately or all combined into one or several units to constitute, or one (some) of the units can be further split into at least two smaller sub-units in function, and the same operation can be achieved without affecting the realization of the technical effects of the embodiments of the present application. The above modules are divided based on logical functions. In practical applications, the functions of one module can also be implemented by at least two units, or the functions of at least two modules can be implemented by one unit. In other embodiments of the present application, the big data processing device may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of at least two units.

[0166] According to one embodiment of the present application, the program can be executed by running on a general-purpose computer device such as a computer including a central processing unit (CPU), a random access memory medium (RAM), a read-only memory medium (ROM), and other processing components and storage components. Figure 2 A computer program (including program code) for each step involved in the corresponding method shown in Figure 3 The big data processing device shown in the embodiment of the present application is used to implement the big data processing method for smart logistics scheduling. The above computer program can be recorded on a computer-readable recording medium, for example, and loaded into the above computing device through the computer-readable recording medium and run therein.

[0167] See also Figure 4 , is a schematic diagram of a structure of a big data processing system implemented as a computer device according to an embodiment of the present application. Figure 4 As shown, the above-mentioned computer device 1000 may include: a processor 1001, a network interface 1004 and a memory 1005. In addition, the above-mentioned computer device 1000 may also include: a user interface 1003, and at least one communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. Among them, the user interface 1003 may include a display screen (Display), a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or it may be a non-volatile memory (non-volatile memory), such as at least one disk storage. The memory 1005 may optionally also be at least one storage device away from the aforementioned processor 1001. As Figure 4 As shown, the memory 1005 as a computer-readable storage medium may include an operating system, a network communication module, a user interface module, and a device control application program.

[0168] exist Figure 4 In the computer device 1000 shown, the network interface 1004 can provide a network communication function; the user interface 1003 is mainly used to provide an input interface; and the processor 1001 can be used to call the device control application stored in the memory 1005 to implement the method provided in the above embodiment.

[0169] It should be understood that the computer device 1000 described in the embodiment of the present application can execute the above Figure 2 The description of the big data processing method applied to smart logistics scheduling in the corresponding embodiment can also be performed as described above. Figure 3The description of the big data processing device in the corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of the same method will not be repeated here.

[0170] In addition, it should be pointed out here that: the embodiment of the present application also provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program executed by the aforementioned big data processing device, and the computer program includes program instructions. When the processor executes the program instructions, the aforementioned Figure 2 The description of the above-mentioned big data processing method applied to smart logistics scheduling in the corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of using the same method will not be repeated. For technical details not disclosed in the computer-readable storage medium embodiment involved in this application, please refer to the description of the method embodiment of this application.

[0171] As an example, the above program instructions may be deployed on a computer device for execution, or deployed on at least two computer devices at one location for execution, or executed on at least two computer devices distributed at at least two locations and interconnected through a communication network. At least two computer devices distributed at at least two locations and interconnected through a communication network may constitute a blockchain network.

[0172] The computer-readable storage medium may be the large data processing device provided in any of the aforementioned embodiments or the central storage unit of the computer device, such as a hard disk or a memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (smart media card, SMC), a secure digital (secure digital, SD) card, a flash card (flash card), etc. equipped on the computer device. Further, the computer-readable storage medium may also include both the central storage unit of the computer device and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0173] The terms "first", "second", etc. in the description, claims, and drawings of the embodiments of the present application are used to distinguish different contents, rather than to describe a specific order. In addition, the term "comprising" and any variation thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product, or equipment comprising a series of steps or units is not limited to the listed steps or modules, but may optionally include steps or modules that are not listed, or may optionally include other step units inherent to these processes, methods, devices, products, or equipment.

[0174] The present application also provides a computer program product, including a computer program / instruction, which implements the above-mentioned Figure 2 The description of the above-mentioned big data processing method applied to smart logistics scheduling in the corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of using the same method will not be repeated. For technical details not disclosed in the embodiments of the computer program product involved in this application, please refer to the description of the method embodiment of this application.

[0175] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0176] The method and related apparatus provided by the embodiment of the present application are described with reference to the method flow chart and / or structural diagram provided by the embodiment of the present application. Specifically, each process and / or box in the method flow chart and / or structural diagram, as well as the combination of the processes and / or boxes in the flow chart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable network connection device to generate a machine, so that the instructions executed by the processor of the computer or other programmable network connection device generate instructions for implementing the process in the process. Figure 1 Schematic diagram of one or more processes and / or structures Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable network-connected device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 Schematic diagram of one or more processes and / or structures Figure 1 These computer program instructions can also be loaded onto a computer or other programmable network connected device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1A process or multiple processes and / or structures illustrate the steps of the functions specified in a box or multiple boxes. The above disclosure is only the preferred embodiment of the present application, and it is certainly not used to limit the scope of the rights of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope of the present application.

Claims

1. A big data processing method applied to smart logistics scheduling, characterized in that: The method comprises: Acquire initial logistics scheduling data and scheduling constraint information with target logistics route data, wherein the scheduling constraint information is used to indicate the construction of logistics scheduling data with the target logistics route data as logistics route data, and the information of the scheduling constraint information includes route constraint conditions corresponding to the target route constraint level; wherein the initial logistics scheduling data includes the category of transport vehicles, the attributes of goods, the transport price, inventory data and the logistics transport route; the scheduling constraint information is used to indicate the construction of logistics scheduling data with the target logistics route data as logistics route data, and on the basis of the existing route of the target logistics route data, the route constraint conditions to be constructed corresponding to the target route constraint level are limited; the target route constraint level is one or more of transport route toll constraints, transport route type constraints, transport route environmental constraints and transport route smoothness constraints; Loading the scheduling constraint information into a pre-debugging knowledge representation mining network to perform pre-debugging knowledge representation mining to obtain a pre-debugging knowledge representation corresponding to the scheduling constraint information; Loading the pre-commissioning knowledge representation into a route representation vector construction network, adjusting the route according to the pre-commissioning knowledge representation, and constructing a target route representation vector corresponding to the target logistics route data; Loading the initial logistics scheduling data and the target route characterization vector into a logistics scheduling data construction network to construct logistics scheduling data, thereby obtaining target scheduling logistics scheduling data; The route representation vector construction network is based on route constraint condition samples corresponding to multiple preset route constraint levels and a multi-level route line adjustment network corresponding to the multiple preset route constraint levels, and the preset route representation vector construction network is taught to perform route line adjustment and debugging of the multiple preset route constraint levels; the target route constraint level is one or more preset route constraint levels among the multiple preset route constraint levels; the route constraint condition samples are route constraint condition training samples; The step of loading the pre-debugging knowledge representation into a route representation vector construction network, adjusting the route according to the pre-debugging knowledge representation, and constructing a target route representation vector corresponding to the target logistics route data includes: Loading the pre-debugging knowledge representation and the preset enhanced knowledge representation into the route representation vector construction network, and guiding the preset enhanced knowledge representation to perform the adjustment operation of the target route constraint level according to the pre-debugging knowledge representation to obtain the target route representation vector; The pre-debugging knowledge representation mining network is a deep learning network, including a first information embedding mapping component, a second information embedding mapping component and a multimodal representation vector mapping network. The loading of the scheduling constraint information into the pre-debugging knowledge representation mining network for pre-debugging knowledge representation mining, and obtaining the pre-debugging knowledge representation corresponding to the scheduling constraint information include: The scheduling constraint information is loaded into the first information embedding mapping component to perform constraint condition feature mining to obtain a first constraint condition representation vector, wherein the first information embedding mapping component is an information embedding mapping component using multi-modal alignment pre-debugging; The scheduling constraint information is loaded into the second information embedding mapping component to perform constraint condition feature mining to obtain a second constraint condition representation vector, wherein the second information embedding mapping component is an information embedding mapping component that adopts constraint condition feature mining pre-debugging; wherein the information embedding mapping component is a converter network; Loading the first constraint condition representation vector and the second constraint condition representation vector into the multimodal representation vector mapping network for multimodal representation vector mapping to obtain the pre-debugging knowledge representation; the multimodal representation vector mapping network is a generation network based on a diffusion mechanism; The logistics scheduling data construction network includes: a scheduling data feature extraction component, a representation vector integration component and a data restoration mapping component. The network structure of the data restoration mapping component is a decoder in a variational autoencoder. The initial logistics scheduling data and the target route representation vector are loaded into the logistics scheduling data construction network to construct the logistics scheduling data. Obtaining the target scheduling logistics scheduling data includes: Loading the initial logistics scheduling data into the scheduling data feature extraction component to perform data representation vector mining to obtain the initial logistics scheduling data representation vector; Loading the initial logistics scheduling data representation vector and the target route representation vector into the representation vector integration component to perform a representation vector integration operation to obtain an integrated logistics scheduling data representation vector; Loading the integrated logistics scheduling data representation vector into the data restoration mapping component for data restoration mapping to obtain the target scheduling logistics scheduling data; The method further comprises: Cloning the network internal variables of the preset route representation vector construction network to obtain a first route representation vector construction network and a second route representation vector construction network, wherein the first route representation vector construction network and the second route representation vector construction network are at the same level; Obtaining a first pre-debugging knowledge representation sample corresponding to a first scheduling constraint information sample having the route constraint condition sample, and an enhanced knowledge representation sample; the enhanced knowledge representation sample is a feature graph corresponding to noise information whose probability density function of the disturbance information distribution conforms to a Gaussian distribution; When adjusting the first route representation vector construction network and the second route representation vector construction network to construct a route representation vector according to the first pre-debugging knowledge representation sample, the enhanced knowledge representation sample and the preset subsidiary enhancement information corresponding to the enhanced knowledge representation sample, adding the route constraint condition sample to the first route representation vector construction network according to the multi-level route line adjustment network, so that the first route representation vector construction network is adjusted and debugged to obtain a first debugging network; wherein the preset subsidiary enhancement information is the added enhancement information; Merging the network internal variables of the first debugging network and the network internal variables of the second route representation vector construction network in order to obtain the target network internal variables; Optimizing the network internal variables of the preset route representation vector construction network according to the target network internal variables to obtain the route representation vector construction network; The first route representation vector construction network includes: a first embedding mapping component, the second route representation vector construction network includes: a second embedding mapping component and a restoration mapping component, when the first route representation vector construction network and the second route representation vector construction network are adjusted to construct a route representation vector according to the first pre-debugging knowledge representation sample, the enhanced knowledge representation sample and the preset subsidiary enhancement information corresponding to the enhanced knowledge representation sample, the route constraint condition sample is added to the first route representation vector construction network according to the multi-level route line adjustment network, so that the first route representation vector construction network performs route line adjustment and debugging, and obtaining the first debugging network includes: Loading the route constraint condition samples corresponding to the plurality of preset route constraint levels into the multi-level route line adjustment network to perform route constraint condition representation vector mining to obtain route constraint condition representation vector samples; Loading the enhanced knowledge representation sample, the first pre-debugging knowledge representation sample and the route constraint condition representation vector sample into the first embedding mapping component, and embedding and mapping the enhanced knowledge representation sample according to the first pre-debugging knowledge representation sample and the route constraint condition representation vector sample to obtain a first embedding mapping representation vector; Loading the enhanced knowledge representation sample and the first pre-debugging knowledge representation sample into the second embedding mapping component, and performing embedding mapping on the enhanced knowledge representation sample according to the first pre-debugging knowledge representation sample to obtain a second embedding mapping representation vector; Loading the first embedded mapping representation vector, the second embedded mapping representation vector and the first pre-debugging knowledge representation sample into the restoration mapping component, and performing restoration mapping on the first embedded mapping representation vector and the second embedded mapping representation vector according to the first pre-debugging knowledge representation sample to obtain a route representation vector sample; According to a first enhancement error between a first enhancement information sample and the preset subsidiary enhancement information, the first embedding mapping component is debugged to obtain the first debugging network; the first enhancement information sample is the enhancement information between the enhancement knowledge representation sample and the route representation vector sample; The first embedding mapping component includes a cascaded first pooling operator of multiple feature granularities, the route constraint representation vector sample includes the route constraint representation vector of the multiple feature granularities, the first embedding mapping representation vector includes the first pooling representation vector of the multiple feature granularities, the enhanced knowledge representation sample, the first pre-debugging knowledge representation sample and the route constraint representation vector sample are loaded into the first embedding mapping component, and the enhanced knowledge representation sample is embedded and mapped according to the first pre-debugging knowledge representation sample and the route constraint representation vector sample to obtain a first embedding mapping. The projection characterization vector includes: based on the first pooling operator of each feature granularity in the multiple feature granularities, according to the first pre-debugging knowledge representation sample, the execution characterization vector of the first pooling operator of each feature granularity is pooled characterization vector extracted to obtain the first pooling characterization vector of each feature granularity, wherein the execution characterization vector of the first pooling operator of the first feature granularity is the enhanced knowledge representation sample, and the execution characterization vectors of the first pooling operator from the second feature granularity to the last feature granularity are the first pooling characterization vector of the previous feature granularity itself and the route constraint condition characterization vector of the previous feature granularity itself; The second embedding mapping component includes a cascaded second pooling operator of the multiple feature granularities, the second embedding mapping representation vector includes the second pooling representation vector of the multiple feature granularities, the loading of the enhanced knowledge representation sample and the first pre-debugging knowledge representation sample into the second embedding mapping component, embedding and mapping the enhanced knowledge representation sample according to the first pre-debugging knowledge representation sample, and obtaining the second embedding mapping representation vector includes: based on the second pooling operator of each feature granularity, according to the first pre-debugging knowledge representation sample, performing pooling representation vector extraction on the execution representation vector of the second pooling operator of each feature granularity, and obtaining the second pooling representation vector of each feature granularity, wherein the execution representation vector of the second pooling operator of the first feature granularity is the enhanced knowledge representation sample, and the execution representation vector of the second pooling operator from the second feature granularity to the last feature granularity is the second pooling representation vector of the previous feature granularity itself; The restoration mapping component includes the cascaded anti-pooling operators of the multiple feature granularities, the first embedded mapping representation vector, the second embedded mapping representation vector and the first pre-debugging knowledge representation sample are loaded into the restoration mapping component, and the first embedded mapping representation vector and the second embedded mapping representation vector are restored and mapped according to the first pre-debugging knowledge representation sample, and the route representation vector sample is obtained, including: based on the anti-pooling operator of each feature granularity, according to the first pre-debugging knowledge representation sample, the execution representation vector of the anti-pooling operator of each feature granularity is extracted by anti-pooling representation vector, and the anti-pooling representation vector of each feature granularity is obtained, wherein , the execution characterization vector of the reverse pooling operator of the last feature granularity connected to the second pooling operator of the last feature granularity is the second pooling characterization vector of the last feature granularity, the first pooling characterization vector of the same feature granularity as itself, and the route constraint characterization vector of the same feature granularity as itself, and the execution characterization vector of the reverse pooling operator from the first feature granularity to the last second feature granularity includes the reverse pooling characterization vector of the next feature granularity itself, the first pooling characterization vector of the same feature granularity as itself, and the route constraint characterization vector of the same feature granularity as itself; the reverse pooling characterization vector output by the reverse pooling operator of the first feature granularity is used as the route characterization vector sample; The multi-level route regulation network includes route regulation components corresponding to the plurality of preset route constraint levels respectively, and the route constraint condition samples corresponding to the plurality of preset route constraint levels are loaded into the multi-level route regulation network to perform route constraint condition representation vector mining, and obtaining the route constraint condition representation vector samples includes: Loading the route constraint condition samples corresponding to each preset route constraint level in the plurality of preset route constraint levels into the route adjustment component corresponding to the preset route constraint level to perform route constraint condition representation vector mining, and obtaining the target route constraint condition representation vector corresponding to each preset route constraint level; The target route constraint condition representation vectors corresponding to the multiple preset route constraint levels are eccentrically adjusted and integrated to obtain the route constraint condition representation vector sample, wherein the eccentric adjustment and integration is weighted, and then the weighted results are summed.

2. The method according to claim 1, characterized in that The route adjustment component includes a coordinate inverse combination operator and a cascaded multi-feature granularity residual representation vector extraction operator, the target route constraint condition representation vector includes the multi-feature granularity residual route constraint condition representation vector, the route constraint condition samples corresponding to each preset route constraint level in the multiple preset route constraint levels are loaded into the route adjustment component of the corresponding preset route constraint level to perform route constraint condition representation vector mining, and obtaining the target route constraint condition representation vector corresponding to each preset route constraint level includes: Loading the route constraint condition samples corresponding to each preset route constraint level into the coordinate decombination operator corresponding to the preset route constraint level to perform coordinate decombination processing, and obtaining the initial constraint condition representation vector corresponding to each preset route constraint level, wherein the coordinate decombination processing is to perform convolution operation and channel expansion on the input features, and then perform reshaping operation; According to the residual representation vector extraction operator of the multi-feature granularity corresponding to each preset route constraint level, the residual representation vector of the multi-feature granularity is extracted for the initial constraint condition representation vector to obtain the residual route constraint condition representation vector of the multi-feature granularity corresponding to each preset route constraint level; The route constraint condition representation vector example includes the route constraint condition representation vector of the multi-feature granularity, and the eccentric adjustment and integration of the target route constraint condition representation vectors corresponding to the multiple preset route constraint levels to obtain the route constraint condition representation vector example includes: The residual route constraint condition representation vectors of each characteristic granularity corresponding to the plurality of preset route constraint levels are respectively eccentrically adjusted and integrated to obtain the route constraint condition representation vector of each characteristic granularity.

3. The method according to claim 1, characterized in that The method further comprises: Obtaining an initial logistics scheduling data sample having a logistics route data sample, a second scheduling constraint information sample having the route constraint condition sample, and a logistics scheduling data representation vector sample corresponding to the scheduling logistics scheduling data sample matching the second scheduling constraint information sample, wherein the second scheduling constraint information sample is used to indicate the construction of logistics scheduling data with the logistics route data sample as the logistics route data; According to the second scheduling constraint information sample and the logistics scheduling data representation vector sample, pre-debugging knowledge representation mining debugging is performed on a preset pre-debugging knowledge representation mining network to obtain the pre-debugging knowledge representation mining network; Determining a second pre-debugging knowledge representation example corresponding to the second scheduling constraint information example according to the pre-debugging knowledge representation mining network; Based on the route constraint condition sample, the second pre-debugging knowledge representation sample, the initial logistics scheduling data sample and the scheduled logistics scheduling data sample, the preset route characterization vector construction network, the multi-level route line adjustment network and the preset neural network are collaboratively debugged to obtain the route characterization vector construction network, the debugged multi-level route line adjustment network and the logistics scheduling data construction network.

4. A big data processing system, characterized in that: include: processor; and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the method according to any one of claims 1 to 3.

Images