Scheme recommendation method and device, electronic equipment, storage medium and program product

Abstract

The invention provides a scheme recommendation method and device, electronic equipment, a storage medium and a program product, and relates to the technical field of machine learning. The logistics scheme recommendation method comprises the following steps: taking a reference step discount suitable for a logistics project as an initial step discount, carrying out iterative optimization, completing a round of iteration to obtain a filial step discount, and predicting a logistics order amount of a logistics user based on the filial step discount and user characteristics of the logistics user; and in at least two adjacent rounds of iterative optimization, when it is detected that the logistics order quantity predicted in the subsequent round of iterative optimization is smaller than or equal to the logistics order quantity predicted in the first round, stopping iterative optimization, and determining the filial generation step discount obtained in the first round as the recommended logistics scheme. Through the technical scheme disclosed by the invention, the logistics demand of the user can be predicted more accurately, so that the logistics scheme which better conforms to the actual condition of the user is formulated.

Description

Background Art

[0002] Logistics users refer to individuals, enterprises or other organizations that use logistics services. In the logistics user contract scenario, it is necessary to recommend logistics prices for logistics users. Currently, one implementation scheme for logistics price recommendation is to recommend based on the standard price and the standard discount template. The number of orders issued by logistics users is inversely proportional to the discount obtained. Due to the universality of this scheme, it is impossible to adjust logistics prices according to the characteristics of different logistics users. As a result, the current logistics price recommendation scheme is not reasonable enough, causing logistics users to spend more on logistics costs.

[0003] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0004] The purpose of the present disclosure is to provide a logistics plan recommendation method, a logistics plan recommendation device, an electronic device, a computer-readable storage medium and a computer program product, which can at least to a certain extent improve the problem of many driving planning restrictions in the automatic driving control process in the related technology.

[0005] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0006] According to one aspect of the present disclosure, a logistics solution recommendation method is provided, comprising: taking a reference step discount applicable to a logistics project as a first-generation step discount, performing iterative optimization, wherein a sub-generation step discount is obtained after completing one round of iteration, and the logistics order quantity of the logistics user is predicted based on the sub-generation step discount and the user characteristics of the logistics user; in at least two adjacent rounds of iterative optimization, if it is detected that the logistics order quantity predicted in subsequent iterative optimization rounds is less than or equal to the logistics order quantity predicted in the first round, the iterative optimization is stopped, and the sub-generation step discount obtained in the first round is determined as the recommended logistics solution.

[0007] In some embodiments of the present disclosure, the user characteristics include user portrait characteristics and user fulfillment characteristics, and based on the child-generation tiered discount and the user characteristics of the logistics user, predicting the logistics order quantity of the logistics user includes: inputting the user portrait characteristics into a logistics market information model, and outputting a first feature vector, wherein the user portrait characteristics include a user attribute portrait and a competitor portrait that quotes the logistics user; inputting the user fulfillment characteristics and the child-generation tiered discount into a fulfillment information model, and outputting a second feature vector, wherein the user fulfillment characteristics include logistics items and agreed data on the logistics items; splicing the first feature vector and the second feature vector to obtain a spliced ​​feature; inputting the spliced ​​feature into an order quantity prediction model, and outputting the logistics order quantity of the logistics user.

[0008] In some embodiments of the present disclosure, before predicting the logistics order quantity of the logistics user based on the child-generation tiered discount and the user characteristics of the logistics user, it also includes: performing prompt instruction processing on the collected portrait features of multiple historical logistics users to obtain instruction features, and the instruction features include the corresponding feature items and feature values ​​in the portrait features of the historical logistics users; pre-training the language model based on the instruction features to obtain a pre-trained model with pre-trained weights; and performing low-rank adaptive fine-tuning on the pre-trained model to obtain the logistics market information model.

[0009] In some embodiments of the present disclosure, the pre-trained model is subjected to low-rank adaptive fine-tuning to obtain the logistics market information model, including: initializing a dimensionality reduction matrix based on a random Gaussian distribution, and initializing a dimensionality increase matrix based on a 0 matrix; fixing the pre-trained weights, and training the dimensionality reduction matrix and the dimensionality increase matrix based on the instructed features to obtain fine-tuning weights; and superimposing the fine-tuning weights on the pre-trained weights to fine-tune the pre-trained model into the logistics market information model.

[0010] In some embodiments of the present disclosure, before predicting the logistics order quantity of the logistics user based on the child-generation step discount and the user characteristics of the logistics user, it also includes: training a deep neural network model based on the collected historical user fulfillment characteristics and the corresponding contracted step discounts to obtain the fulfillment information model.

[0011] In some embodiments of the present disclosure, a deep neural network model is trained based on collected historical user performance features and corresponding contract step discounts to obtain the performance information model, including: inputting the historical user performance features and the corresponding contract step discounts into the input layer of the deep neural network model, and outputting an input vector; inputting the input vector into the encoding layer for encoding to obtain an embedded vector, wherein the embedded vector includes a logistics item embedding vector and an agreement data embedding vector; inputting the embedded vector into the hidden layer to output a performance prediction vector; and training the model parameters of the deep neural network model based on the gap between the prediction vector and the actual performance vector of the historical user until the training is completed to obtain the performance information model.

[0012] In some embodiments of the present disclosure, the input vector is input into the encoding layer for encoding to obtain an embedded vector, which includes a logistics item embedding vector and an agreement data embedding vector, and also includes: bucketing the historical user fulfillment features, wherein if the feature value of the historical user fulfillment feature is less than or equal to a feature threshold, the input vector corresponding to the historical user fulfillment feature is allocated to the corresponding bucket for processing; if the feature value of the historical user fulfillment feature is greater than the feature threshold, the input vector corresponding to the historical user fulfillment feature is allocated to the same bucket for processing.

[0013] In some embodiments of the present disclosure, before predicting the logistics order quantity of the logistics user based on the child-generation step discount and the user characteristics of the logistics user, it also includes: inputting the portrait characteristics of the historical logistics user into the logistics market information model, and outputting a first historical feature vector; inputting the historical user performance characteristics and the corresponding contract step discount into the performance information model, and outputting a second historical feature vector; splicing the first historical feature vector and the second historical feature vector to obtain a historical splicing feature; inputting the historical splicing feature into the regression model to output an order quantity prediction value; calculating the loss between the order quantity prediction value and the actual order quantity value of the historical logistics user based on the mean square error loss function, so as to update the model parameters of the regression model based on the loss, until the training is completed and the order quantity prediction model is obtained.

[0014] In some embodiments of the present disclosure, the splicing features are input into an order quantity prediction model to output the logistics order quantity of the logistics user, and the splicing features generated in a first historical period are input into the order quantity prediction model to output the logistics order quantity of the logistics user in a second future period.

[0015] In some embodiments of the present disclosure, a reference step discount applicable to a logistics project is used as a first-generation step discount and iteratively optimized, including: predicting the logistics order quantity of the logistics user based on the reference step discount as a fitness value; selecting a preferred step discount from the reference step discount based on the fitness value; and performing crossover and / or mutation operations on the preferred step discount to obtain the child step discount.

[0016] According to another aspect of the present disclosure, a logistics solution recommendation device is provided, including: an iterative optimization module, used to use a reference step discount applicable to a logistics project as a first-generation step discount for iterative optimization, wherein a sub-generation step discount is obtained after completing one round of iteration, and the logistics order quantity of the logistics user is predicted based on the sub-generation step discount and the user characteristics of the logistics user; a determination module, used to detect in at least two adjacent rounds of iterative optimization that the logistics order quantity predicted in subsequent iterative optimization rounds is less than or equal to the logistics order quantity predicted in the first round, stop the iterative optimization, and determine the sub-generation step discount obtained in the first round as the recommended logistics solution.

[0017] According to another aspect of the present disclosure, there is provided an electronic device, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-mentioned logistics solution recommendation methods by executing the executable instructions.

[0018] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, any one of the above-mentioned logistics solution recommendation methods is implemented.

[0019] According to another aspect of the present disclosure, a computer program product is provided, on which a computer program is stored, and when the computer program is executed by a processor, any one of the above-mentioned logistics solution recommendation methods is implemented.

[0020] The logistics solution recommendation scheme provided by the embodiment of the present disclosure, in three consecutive rounds of iterative optimization of step discounts, if it is found that the logistics order quantity predicted by the second and third rounds of iterative optimization is less than or equal to the logistics order quantity predicted in the first round, this means that the subsequent iterations have not brought better results, and may have reached a local optimum or overfitting. At this time, the iterative optimization is stopped, and the child step discount obtained in the first round is determined as the recommended logistics solution. By combining user characteristics and iterative optimization of step discounts, the user's logistics needs can be more accurately predicted, thereby formulating a logistics solution that is more in line with the user's actual situation. By combining user characteristics and iterative optimization of step discounts, the user's logistics needs can be more accurately predicted, thereby formulating a logistics solution that is more in line with the user's actual situation.

[0021] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 A schematic diagram showing the structure of a logistics solution recommendation system in an embodiment of the present disclosure;

[0024] Figure 2 A flow chart showing a method for recommending a logistics solution in an embodiment of the present disclosure is shown;

[0025] Figure 3 A schematic diagram showing another method for recommending a logistics solution in an embodiment of the present disclosure;

[0026] Figure 4 A schematic diagram of a model showing a logistics solution recommendation scheme in an embodiment of the present disclosure;

[0027] Figure 5 A schematic diagram showing a model of another logistics solution recommendation scheme in an embodiment of the present disclosure;

[0028] Figure 6 A schematic diagram of a model showing another logistics solution recommendation scheme in an embodiment of the present disclosure;

[0029] Figure 7 A flow chart showing another method for recommending a logistics solution in an embodiment of the present disclosure is shown;

[0030] Figure 8 A schematic diagram showing a logistics solution recommendation device in an embodiment of the present disclosure;

[0031] Fig. 9 A schematic diagram of an electronic device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0032] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0033] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0034] Logistics service providers need to provide logistics solutions for logistics users. In the logistics user contract signing scenario, the logistics solution method currently adopted by the industry is to recommend tiered discounts based on standard prices according to standard discount templates, that is, to recommend solutions according to standard tiered discounts (which can be understood as an implementation method of reference tiered discounts in the present disclosure). As the number of orders and order revenue of logistics users increases, lower discounts can be obtained. However, if the standard discount template is used to quote for all logistics users, the tiered discounts are not targeted. For logistics users of different sizes, the recommendation results generated by the standard template are not guiding, and the discounts are not attractive. In the long run, the churn rate of logistics users increases. In addition, if the logistics solution is not combined with information from the external market, the recommendation results will not be obviously competitive.

[0035] Based on the above defects, in the present disclosure, different tiered discounts can be recommended according to the size, industry, city distribution, estimated order quantity and other information of logistics users. The price patterns of group logistics users are learned by combining the operations research model and the deep learning model to make the recommendation results more credible. In addition, through the form of a large model, prompt words are input to obtain features, and external market information is implicitly introduced to improve the rationality of logistics plan recommendations.

[0036] To facilitate understanding, several terms involved in this application are first explained below.

[0037] Tiered discount: a preferential method that divides products into different levels according to criteria such as purchase quantity or consumption amount, and sets different discount amounts for each level.

[0038] DNN (Deep Neural Network): A deep neural network is an artificial neural network that contains multiple hidden layers. It can automatically learn complex patterns and features from large amounts of data and is widely used in image recognition, speech processing, natural language processing and other fields.

[0039] Qwen1.5: is a language model developed and trained by relevant teams. It has certain language understanding and generation capabilities and can handle a variety of natural language processing tasks.

[0040] LORA (Low-Rank Adaptation): refers to low-rank adaptation, a method of model fine-tuning. By injecting a small amount of trainable low-rank matrices into the model to adapt to new tasks or data, the number of training parameters is reduced, the computing resource requirements are reduced, and better performance improvement and new task adaptation can be achieved.

[0041] Figure 1 1 is a schematic diagram of a computer system provided by an exemplary embodiment of the present application. The system includes: a plurality of terminals 120 and a server cluster 140.

[0042] The terminal 120 may be a mobile terminal such as a mobile phone, a game console, a tablet computer, an e-book reader, smart glasses, an MP4 (Moving Picture Experts Group Audio Layer IV) player, a smart home device, an AR (Augmented Reality) device, a VR (Virtual Reality) device, or the terminal 120 may also be a personal computer (PC), such as a laptop computer and a desktop computer.

[0043] The terminal 120 may be installed with an application for providing logistics solution recommendations.

[0044] The terminal 120 is connected to the server cluster 140 via a communication network. Optionally, the communication network is a wired network or a wireless network.

[0045] The server cluster 140 is a server, or is composed of several servers, or is a virtualization platform, or is a cloud computing service center. The server cluster 140 is used to provide background services for the application program that provides logistics solution recommendations. Optionally, the server cluster 140 undertakes the main computing work, and the terminal 120 undertakes the secondary computing work; or, the server cluster 140 undertakes the secondary computing work, and the terminal 120 undertakes the main computing work; or, the terminal 120 and the server cluster 140 adopt a distributed computing architecture for collaborative computing.

[0046] In some optional embodiments, the server cluster 140 is used to store logistics solution recommendation program information.

[0047] Optionally, the logistics user terminals of the application programs installed in different terminals 120 are the same, or the logistics user terminals of the application programs installed on the two terminals 120 are logistics user terminals of the same type of application programs on different control system platforms. Based on the different terminal platforms, the specific forms of the logistics user terminals of the application programs may also be different, for example, the logistics user terminals of the application programs may be mobile phone logistics user terminals, PC logistics user terminals, or global wide area network (World Wide Web, Web) logistics user terminals, etc.

[0048] Those skilled in the art will appreciate that the number of the terminals 120 may be more or less. For example, there may be only one terminal, or there may be dozens or hundreds of terminals, or a greater number. The embodiment of the present application does not limit the number and device type of the terminals.

[0049] Optionally, the system may also include a management device ( Figure 1 (not shown), the management device is connected to the server cluster 140 via a communication network. Optionally, the communication network is a wired network or a wireless network.

[0050] Optionally, the above-mentioned wireless network or wired network uses standard communication technology and / or protocol. The network is usually the Internet, but it can also be any network, including but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a dedicated network or any combination of a virtual private network). In some embodiments, the data exchanged through the network is represented by technologies and / or formats including Hyper Text Mark-up Language (HTML), Extensible Markup Language (XML), etc. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec) can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above-mentioned data communication technologies.

[0051] The logistics solution recommendation method in this example implementation will be described in more detail below with reference to the accompanying drawings and examples.

[0052] like Figure 2 As shown, a logistics solution recommendation method according to an embodiment of the present disclosure includes:

[0053] Step S202, taking the reference step discount applicable to the logistics project as the first-generation step discount, and performing iterative optimization, wherein a sub-generation step discount is obtained after completing one round of iteration, and the logistics order quantity of the logistics user is predicted based on the sub-generation step discount and the user characteristics of the logistics user.

[0054] Among them, "step discount" refers to different discount ratios or discount amounts set according to the different logistics order quantity ranges of logistics users.

[0055] The reference tiered discount may be a preliminary tiered discount scheme set based on logistics market experience, market conditions, historical data or preliminary analysis.

[0056] The first-generation step discount refers to the initial parent-generation step discount scheme that is iteratively optimized.

[0057] Iterative optimization refers to modifying the current step discount according to specified rules or algorithms in the hope of obtaining better results. After each round of iteration, a new offspring step discount will be obtained. The specified rules or algorithms may include genetic algorithms, simulated annealing algorithms, particle swarm optimization algorithms, and gradient descent methods.

[0058] User characteristics of logistics users: refers to relevant information that can describe the characteristics and behavior patterns of logistics users.

[0059] The logistics order volume of a logistics user refers to the number of logistics orders that the logistics user expects or may generate within a specified time period.

[0060] In addition, based on the sub-generation tiered discounts and the user characteristics of the logistics users, a method for predicting the logistics user's logistics order quantity is as follows: based on the above-mentioned sub-generation tiered discounts and these characteristics of the logistics users, data analysis, machine learning models or other prediction methods are used to estimate or infer the number of logistics orders that the logistics user may generate under given conditions, that is, the logistics order quantity.

[0061] Step S204: In at least two adjacent rounds of iterative optimization, if it is detected that the logistics order quantities predicted in subsequent iterative optimization rounds are less than or equal to the logistics order quantities predicted in the first round, the iterative optimization is stopped, and the child-generation ladder discount obtained in the first round is determined as the recommended logistics solution.

[0062] In some embodiments, at least two adjacent rounds of iterative optimization are three adjacent rounds of iterative optimization, and the three adjacent rounds of iterative optimization refer to three consecutive iterative processes. In these three iterations, the child-generation step discounts of the first round, the second round and the third round are obtained respectively, and it is detected that the logistics order quantities predicted by the second and third rounds of iterative optimization are less than or equal to the logistics order quantity predicted in the first round.

[0063] In this embodiment, in three consecutive rounds of iterative optimization of step discounts, if it is found that the logistics order quantity predicted by the second and third rounds of iterative optimization is less than or equal to the logistics order quantity predicted by the first round, this means that the subsequent iterations have not brought better results, and may have reached a local optimum or overfitting. At this time, the iterative optimization is stopped, and the child step discount obtained in the first round is determined as the recommended logistics plan. By combining user characteristics and iterative optimization of step discounts, the user's logistics needs can be more accurately predicted, thereby formulating a logistics plan that is more in line with the user's actual situation. By combining user characteristics and iterative optimization of step discounts, the user's logistics needs can be more accurately predicted, thereby formulating a logistics plan that is more in line with the user's actual situation.

[0064] like Figure 3 As shown, in some embodiments of the present disclosure, user characteristics include user portrait characteristics and user performance characteristics. Based on the child-generation tiered discount and the user characteristics of the logistics user, the logistics order quantity of the logistics user is predicted, including:

[0065] Step S302, input the user portrait features into the logistics market information model, and output a first feature vector, wherein the user portrait features include a user attribute portrait and a portrait of competitors who quote to the logistics user.

[0066] For example, the user portrait features include: logistics user name, primary industry, secondary industry, tertiary industry of the logistics user, contracted province, contracted city, declaration time, other companies providing logistics services (competing company 1, competing company 2...), quotations of other logistics services (competing quotation 1, competing quotation 2...), and other information.

[0067] Among them, the first-level industry is a larger category in the entire economic activities or market. The second-level industry is a further subdivision based on the first-level industry, which divides out the parts of the first-level industry with certain unique business types, market characteristics, operation models, etc. The tertiary industry is a more detailed and specific classification under the second-level industry, which can usually more accurately reflect the specific business scope, product or service type, etc. of logistics users.

[0068] The logistics market information model is used to process and analyze user portrait features to extract valuable information about the logistics market situation and convert it into the first feature vector.

[0069] Step S304, input the user's fulfillment characteristics and the child-generation ladder discount into the fulfillment information model, and output a second feature vector. The user's fulfillment characteristics include logistics items and agreed data on the logistics items.

[0070] Exemplarily, the user fulfillment characteristics are logistics items and the agreed data on logistics items, including: logistics user contracted products, promised order quantity, promised revenue, genetic algorithm offspring step discounts, etc.

[0071] The fulfillment information model is used to process user fulfillment characteristics and child tiered discounts. For the logistics items in the user fulfillment characteristics, the model can understand the agreed data of the corresponding logistics items to parse out relevant information such as the promised order quantity and promised revenue.

[0072] Step S306: concatenate the first feature vector and the second feature vector to obtain a concatenated feature.

[0073] Step S308, input the splicing features into the order quantity prediction model, and output the logistics order quantity of the logistics user.

[0074] Among them, the order quantity prediction model is a tool used to predict the logistics user's logistics order quantity based on the input feature information. It may be built based on machine learning or deep learning algorithms, such as using neural networks, decision trees, regression analysis and other methods. The model has been trained with a large amount of historical data and has learned the potential relationship between different feature combinations and logistics order quantities, so that it can make accurate order quantity predictions based on newly input features.

[0075] In this embodiment, by introducing the profile of competitors who quote to the logistics user, the information of the external logistics market can be taken into account when making logistics plan recommendations. The first feature vector and the second feature vector are spliced ​​to form a spliced ​​feature that integrates the user's situation in the market and the specific fulfillment situation. The spliced ​​feature is input into the order quantity prediction model, and the model outputs the predicted value of the logistics user's logistics order quantity based on the learned patterns and relationships. By integrating various information such as user portrait features and fulfillment features, it can more comprehensively consider the factors affecting the logistics order quantity, thereby improving the reliability of order quantity prediction, so as to further infer a reasonable tiered discount plan based on the predicted order quantity as a logistics recommendation plan.

[0076] In some embodiments of the present disclosure, before predicting the logistics order quantity of the logistics user based on the child-generation tiered discount and the user characteristics of the logistics user, the method further includes:

[0077] The collected portrait features of multiple historical logistics users are processed into prompt instructions to obtain instructional features, which include corresponding feature items and feature values ​​in the portrait features of historical logistics users.

[0078] Among them, the prompt instruction processing is to classify and name the collected original, diverse and possibly chaotic portrait features of multiple historical logistics users, and give each feature a clear feature item.

[0079] Exemplarily, the portrait features of historical logistics users include: logistics user name, primary industry, secondary industry, tertiary industry, contracted province, contracted city, declaration time, competing companies (competing company 1, competing company 2...), competing quotations (competing quotation 1, competing quotation 2...) and other information.

[0080] Prompt instructions are processed and generated: "Logistics user <name>, belongs to the industry <first-level industry-second-level industry-third-level industry>, its contract signing place is <contract signing province-contract signing city>, the time for applying for quotation is <declaration time>, the competing company is <competing company>, and the quotation of the competing company is <competing quotation>".

[0081] like Figure 4 As shown, the language model is pre-trained based on the d-dimensional instruction features to obtain a pre-trained model with a pre-trained weight matrix.

[0082] Among them, the pre-training weights are the parameter values ​​obtained by the model during the pre-training process.

[0083] The language model can be Qwen1.5, which is used to process various types of natural language processing tasks. For example, Qwen1.5-0.5B is selected as the initial weight for experiments. The large model is trained based on multiple data sources, multiple disciplines, and multiple modalities. Therefore, it implicitly integrates the price information of the logistics market. The user portrait is input into the large model to obtain the first eigenvector, which is introduced into the subsequent training module, so that external market information can be well obtained.

[0084] The pre-trained model is fine-tuned by low-rank adaptation to obtain the logistics market information model.

[0085] In this embodiment, the collected portrait features are carefully analyzed and classified to determine the feature items and feature values ​​corresponding to each feature. In this way, the original and diverse portrait features are converted into imperative features with clear structure and definition, and the obtained imperative features are used to pre-train the language model. After completing the pre-training and obtaining the model with pre-training weights, the pre-trained model is low-rank adaptively fine-tuned to obtain the logistics market information model according to the specific needs and tasks of the logistics market information. Through in-depth learning and optimization of historical logistics user portrait features, the model can more accurately capture the user's behavior patterns, demand preferences and market trends, so as to ensure the reliability of the output of the logistics market information model.

[0086] In some embodiments of the present disclosure, a pre-trained model is subjected to low-rank adaptive fine-tuning to obtain a logistics market information model, including:

[0087] Initialize the dimension reduction matrix based on random Gaussian distribution, and initialize the dimension increase matrix based on the 0 matrix; fix the pre-training weights, train the dimension reduction matrix and the dimension increase matrix based on the command features to obtain the fine-tuning weights; superimpose the fine-tuning weights on the pre-training weights to fine-tune the pre-training model into a logistics market information model.

[0088] like Figure 4 As shown in the figure, LORA weight refers to adding a bypass next to the pre-trained weight to perform a dimensionality increase and then dimensionality reduction operation; A is initialized with a random Gaussian distribution, and B is initialized with a 0 matrix to ensure that the bypass matrix is ​​a 0 matrix when the training is initialized; during the training process, the pre-trained weight matrix W0 is fixed, and only the dimensionality reduction matrix A and the dimensionality increase matrix B are trained to obtain the fine-tuning weight The dimension of the model remains unchanged. During calculation, the parameters of BA and PLM are superimposed to obtain the target weight matrix h, as shown in equations (1) and (2).

[0089]

[0090] Among them, d refers to the vector dimension of the instruction feature, d×r refers to the matrix size of B, r×k refers to the matrix size of A, and R represents the matrix dimension.

[0091] Exemplarily, adding a bypass next to the pre-trained weights and performing the operation of dimensionality increase and then dimensionality reduction indicates that an additional calculation path is added in addition to the original pre-trained weights. This path first increases the dimension (increases the number of parameters or dimensions) and then reduces the dimension (returns to the original dimension). The random Gaussian distribution is used to initialize A in order to give A some random but normal distribution initial values; B is initialized with a 0 matrix so that at the beginning of training, the overall effect of the bypass is 0, that is, it has no effect on the pre-trained weights. During the training process, the pre-trained weights W0 are fixed, and only the dimensionality reduction matrix A and the dimensionality increase matrix B are trained, that is, only the A and B matrices in the newly added bypass are trained and adjusted. Since only A and B are trained, the overall dimension (number of parameters) of the model does not change, but by adjusting the parameters of A and B, the model can adapt to new tasks or data.

[0092] In some embodiments of the present disclosure, before predicting the logistics order quantity of logistics users based on the child-generation tiered discounts and the user characteristics of logistics users, it also includes: training a deep neural network model based on the collected historical user fulfillment characteristics and the corresponding contracted tiered discounts to obtain a fulfillment information model.

[0093] Among them, the deep neural network model structure can be a deep neural network (MLP), a convolutional neural network (CNN) or a recurrent neural network (RNN) and their variants.

[0094] In this embodiment, the fulfillment information model is trained based on historical logistics user data so that the fulfillment information model can output accurate fulfillment information of logistics users.

[0095] In some embodiments of the present disclosure, a deep neural network model is trained based on the collected historical user performance characteristics and the corresponding contract step discounts to obtain a performance information model, including:

[0096] The historical user performance characteristics and the corresponding contract step discounts are input into the input layer of the deep neural network model, and the input vector is output.

[0097] like Figure 5 As shown, taking the DNN model as an example, the structure of the DNN model usually consists of the following main parts:

[0098] The input layer is used to input historical user performance characteristics, including the contracted products, promised order quantity and promised revenue of logistics users, the actual contracted tiered discount of contracted users, and output the input vector. The input vector is input into the encoding layer for encoding to obtain an embedded vector, which includes the logistics project embedding vector and the agreed data embedding vector.

[0099] Hidden layer: Set N hidden layers with n nodes in each layer, input the embedded vector into the hidden layer and output the fulfillment prediction vector.

[0100] The output layer is used to output the second eigenvector, namely eigenvector 2.

[0101] The model parameters of the deep neural network model are trained based on the gap between the fulfillment prediction vector and the actual fulfillment vector of historical users until the training is completed to obtain the fulfillment information model.

[0102] In this embodiment, a large amount of historical user performance feature data and corresponding child-generation ladder discount data are collated and collected. These performance features may cover various details of logistics projects. The prepared historical user performance features and corresponding child-generation ladder discount data are used as input and passed to the input layer of the deep neural network model. The vector output by the input layer is sent to the encoding layer for encoding. The encoding process converts the original performance features and discount data into embedded vectors. For logistics projects and agreed data, corresponding embedded vectors are generated respectively. The encoded embedded vectors are then input into the hidden layer. The neurons in the hidden layer process the input embedded vectors through complex calculations and weight adjustments, and finally output the performance prediction vector. Through the powerful learning ability of the deep neural network and the ability to capture complex relationships, the user's performance can be predicted more accurately.

[0103] In some embodiments of the present disclosure, an input vector is input into a coding layer for encoding to obtain an embedded vector, which includes a logistics item embedding vector and an agreement data embedding vector, and also includes: bucketing historical user fulfillment features, wherein if the feature value of the historical user fulfillment feature is less than or equal to the feature threshold, the input vector corresponding to the historical user fulfillment feature is allocated to the corresponding bucket for processing; if the feature value of the historical user fulfillment feature is greater than the feature threshold, the input vector corresponding to the historical user fulfillment feature is allocated to the same bucket for processing.

[0104] Among them, the bucket feature is generated based on data preprocessing. It divides discrete committed quantities or committed revenues with many different values ​​into several intervals or buckets to group discrete values.

[0105] Exemplarily, historical user fulfillment features include the user's historical committed order quantities. The order quantity data should be divided into 10002 different categories or intervals from 0 to 10001, that is, buckets. For orders between 0 and 10000, the numerical value of the order quantity is directly used to represent the bucket. For example, if the order quantity is 5000, it is in the 5000th bucket, and for orders exceeding 10000, they are all classified into the 10001th bucket.

[0106] Does the historical user fulfillment feature also include the user's historical promised income? Map the order quantity into 0-2001 buckets, 0-100000 orders, the bucket number is 100000 / 50, and the mapping of more than 100000 orders is 2001, and initialize the embedding of 2001.

[0107] In this embodiment, when the feature value of the historical user performance feature is less than or equal to the feature threshold, the corresponding input vector is assigned to the corresponding bucket for processing. In this way, smaller or more common feature values ​​can be subdivided into different buckets, and when the feature value is greater than the feature threshold, it indicates that these feature values ​​are relatively rare feature values, and the corresponding input vector is assigned to the same bucket for processing. Through this bucketing strategy, for rare feature values, by classifying them into the same bucket, the model is prevented from being excessively affected by extreme values ​​that are too sparse and unrepresentative, thereby preventing the long tail problem of data from causing learning bias in the model.

[0108] In some embodiments of the present disclosure, before predicting the logistics order quantity of the logistics user based on the child-generation tiered discount and the user characteristics of the logistics user, the method further includes:

[0109] Input the portrait features of historical logistics users into the logistics market information model to output the first historical feature vector; input the historical user performance features and the corresponding contract step discount performance information model to output the second historical feature vector; splice the first historical feature vector and the second historical feature vector to obtain the historical splicing features; input the historical splicing features into the regression model to output the order quantity prediction value; calculate the loss between the order quantity prediction value and the actual order quantity value of the historical logistics users based on the mean square error loss function, and update the model parameters of the regression model based on the loss until the training is completed to obtain the order quantity prediction model.

[0110] like Figure 6 As shown, the first feature vector output by the large model is concatenated with the second feature vector output by the DNN model and input into the regression model as the input feature vector.

[0111] The regression model includes N hidden layers. After being processed by the hidden layers, the output layer outputs the predicted logistics order volume of the logistics user, such as the order volume in the next three months.

[0112] The MSE loss function is used to calculate the loss of the predicted value and the actual value. MSE Loss is the mean square error (MSE) loss function, as shown in formula (3). The function of the mean square error loss function is to measure the average difference between the predicted value of the model and the true value. The smaller the value, the better the prediction effect of the model.

[0113]

[0114] in, is the true value, y i is the predicted value and n is the number of samples.

[0115] In this embodiment, the portrait features of historical logistics users are input into the logistics market information model, which is processed to output a first historical feature vector. The fulfillment features and the contract step discount are input into the fulfillment information model to obtain a second historical feature vector. By combining the user portrait features and the fulfillment features, the first historical feature vector and the second historical feature vector are spliced ​​to form a comprehensive historical splicing feature. The regression model is trained based on the historical splicing feature, and the obtained order quantity prediction model comprehensively considers various factors affecting the logistics order quantity, thereby improving the accuracy and reliability of the prediction.

[0116] In some embodiments of the present disclosure, the splicing features are input into the order quantity prediction model to output the logistics order quantity of the logistics user, which also includes: inputting the splicing features generated in the first historical period into the order quantity prediction model to output the logistics order quantity of the logistics user in the second future period.

[0117] In this embodiment, the first historical period refers to a period of time that has passed, during which relevant data including portrait features and fulfillment features of historical logistics users are collected, and these data are processed and spliced ​​to form spliced ​​features. These spliced ​​features generated in the first historical period are input into the trained order quantity prediction model, so that the model outputs the prediction results of the logistics order quantity of logistics users in the second future period, so as to understand the expected logistics order quantity in advance as the main factor in evaluating the logistics plan.

[0118] In some embodiments of the present disclosure, a reference tiered discount applicable to a logistics project is used as a first-generation tiered discount, and iterative optimization is performed, including:

[0119] The logistics user's logistics order quantity is predicted based on the reference step discount as the fitness value; the preferred step discount is selected from the reference step discount based on the fitness value; and the preferred step discount is crossover and / or mutation operations are performed to obtain a child step discount.

[0120] The reference discount ladder, that is, the standard discount ladder, can be understood as different companies will set different standard discount ladder templates for different logistics products, as shown in Table 1 below.

[0121] Table 1

[0122]

[0123] Genetic algorithms are used based on the standard discount ladder (initial population) to generate offspring (including selection, crossover, mutation, etc.) based on the company's discount bottom line. The offspring's step discount is input as a feature into the DNN model, and other features are concatenated to obtain the model output. The order volume in the next three months is predicted as the individual fitness score of the genetic algorithm, and the score is retained. When no individual fitness is greater than that of the parent generation within the next two generations, the solution process is terminated and the step discount of the parent generation is retained as the final recommended discount.

[0124] In this embodiment, based on the reference step discounts, combined with the relevant characteristics of the logistics users and the market conditions, the logistics order volume that the logistics users may generate is predicted, and the predicted logistics order volume is used as the fitness value. Based on the calculated fitness value, a discount plan with better performance is screened out from the reference step discounts, namely the preferred step discounts, and the preferred step discounts are crossover and / or mutation operations are performed. The crossover operation refers to exchanging and combining partial elements of two or more preferred step discounts to generate a new discount plan. Through continuous selection, crossover and mutation operations, different step discount combinations can be explored, and the discount strategy that can best stimulate logistics users to increase the number of orders can be found, thereby improving the market competitiveness of the enterprise.

[0125] like Figure 7As shown, a logistics solution recommendation method according to another embodiment of the present disclosure includes:

[0126] The user portrait is processed in an instructional manner to obtain instructional features, the language model is trained using the instructional features to obtain the pre-trained model weights, the pre-trained model is fine-tuned through LORA to obtain the LORA weights, the logistics market information model is obtained, and the first feature vector is output by the logistics market information model.

[0127] The DNN model is trained to obtain the fulfillment information model, the fulfillment cumulative features and the child step discounts are input into the fulfillment information model, the second feature vector is output, the first feature vector and the second feature vector are spliced ​​to obtain the spliced ​​features; the spliced ​​features are input into the order quantity prediction model obtained based on the regression model, and the model outputs the predicted logistics order quantity of the logistics user, that is, the order quantity for the next month.

[0128] Among them, the offspring step discount is generated by iterative optimization of the reference step discount based on genetic algorithm.

[0129] In some embodiments, during the model training phase, historical logistics user quotation data and fulfillment data are used as training data, totaling more than 5 million items.

[0130] In some embodiments, in the prediction stage, the genetic algorithm is used to generate offspring step discounts, and the starting value of the 1-8 step interval is selected, corresponding to the discount value of the 1-8 steps (8 is a hyperparameter in the experimental process and can be adjusted). By averaging 10 generations of offspring step discounts, with an average of 20 offspring results in each generation, the quantity-price relationship model is used to select the offspring step discount with the largest number of orders in the future and recommend it to sales for signing. The experimental process is divided into an experimental group and a control group. The logistics user data in November 2023 and December 2023 are used, and the step discount of the most recently submitted quotation is used to iteratively generate offspring step discounts using a genetic algorithm. The quantity-price relationship model is used to predict the order volume of the offspring step discount in January, February, and March 2024, and the total order volume and revenue in January, February, and March 2024 are recovered. The average number of orders per customer in the experimental group increased by 8.9% compared with the control group, and the average number of orders per customer in the experimental group increased by 7.8% compared with the control group.

[0131] It should be noted that the above figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0132] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods or program products. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as "circuits", "modules" or "systems".

[0133] Refer to the following Figure 8 The logistics solution recommendation device 800 according to this embodiment of the present disclosure is described. Figure 8 The logistics solution recommendation device 800 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiment of the present disclosure.

[0134] The logistics solution recommendation device 800 is presented in the form of a hardware module. The components of the logistics solution recommendation device 800 may include but are not limited to: an iterative optimization module 802, which is used to use the reference step discount applicable to the logistics project as the first-generation step discount for iterative optimization, wherein a sub-generation step discount is obtained after completing one round of iteration, and the logistics order quantity of the logistics user is predicted based on the sub-generation step discount and the user characteristics of the logistics user; a determination module 804, which is used to detect that the logistics order quantity predicted in the subsequent iterative optimization rounds is less than or equal to the logistics order quantity predicted in the first round in at least two adjacent iterative optimization rounds, stop the iterative optimization, and determine the sub-generation step discount obtained in the first round as the recommended logistics solution.

[0135] Refer to the following Fig. 9 The electronic device 900 according to this embodiment of the present disclosure is described. Fig. 9 The electronic device 900 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0136] like Fig. 9 As shown, the electronic device 900 is in the form of a general computing device. The components of the electronic device 900 may include but are not limited to: at least one processing unit 910, at least one storage unit 920, and a bus 930 connecting different system components (including the storage unit 920 and the processing unit 910).

[0137] The storage unit stores program codes, which can be executed by the processing unit 910, so that the processing unit 910 performs the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification. For example, the processing unit 910 can perform the following steps: Figure 2 Steps S202 and S204 shown in , and other steps defined in the logistics solution recommendation method of the present disclosure.

[0138] The storage unit 920 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 9201 and / or a cache storage unit 9202 , and may further include a read-only storage unit (ROM) 9203 .

[0139] The storage unit 920 may also include a program / utility 9204 having a set (at least one) of program modules 9205, such program modules 9205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0140] Bus 930 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0141] The electronic device 900 may also communicate with one or more external devices 960 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device, and / or communicate with any device that enables the electronic device 900 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 950. Furthermore, the electronic device 900 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 950. As shown, the network adapter 950 communicates with other modules of the electronic device 900 via a bus 930. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0142] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0143] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps according to various exemplary implementations of the present disclosure described in the above "Exemplary Method" section of the present specification.

[0144] According to the program product for implementing the above method in the embodiment of the present disclosure, it can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

[0145] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0146] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0147] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0148] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0149] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0150] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0151] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A logistics solution recommendation method, characterized in that: include: The reference step discount applicable to the logistics project is used as the first-generation step discount, and iterative optimization is performed, wherein a sub-generation step discount is obtained after completing one round of iteration, and the logistics order quantity of the logistics user is predicted based on the sub-generation step discount and the user characteristics of the logistics user; In at least two adjacent rounds of iterative optimization, if it is detected that the logistics order quantities predicted in subsequent iterative optimization rounds are less than or equal to the logistics order quantities predicted in the first round, the iterative optimization is stopped, and the child-generation step discount obtained in the first round is determined as the recommended logistics plan.

2. The method for recommending logistics solutions according to claim 1, characterized in that: The user characteristics include user portrait characteristics and user fulfillment characteristics. Based on the child-generation tiered discount and the user characteristics of the logistics user, the logistics order quantity of the logistics user is predicted, including: Inputting the user portrait feature into the logistics market information model and outputting a first feature vector, wherein the user portrait feature includes a user attribute portrait and a competitor portrait that offers a price to the logistics user; Inputting the user fulfillment feature and the child-generation tiered discount into a fulfillment information model, and outputting a second feature vector, wherein the user fulfillment feature includes a logistics item and agreed data on the logistics item; Concatenate the first feature vector and the second feature vector to obtain a concatenated feature; The splicing features are input into an order quantity prediction model, and the logistics order quantity of the logistics user is output.

3. The method for recommending logistics solutions according to claim 2, characterized in that: Before predicting the logistics order quantity of the logistics user based on the child-generation tiered discount and the user characteristics of the logistics user, the method further includes: Perform prompt instruction processing on the collected portrait features of multiple historical logistics users to obtain instruction features, wherein the instruction features include feature items and feature values ​​corresponding to the portrait features of the historical logistics users; Pre-training the language model based on the instruction features to obtain a pre-trained model with pre-trained weights; The pre-trained model is subjected to low-rank adaptive fine-tuning to obtain the logistics market information model.

4. The method for recommending logistics solutions according to claim 3, characterized in that: The pre-trained model is subjected to low-rank adaptive fine-tuning to obtain the logistics market information model, including: Initialize the reduced dimension matrix based on random Gaussian distribution, and initialize the increased dimension matrix based on the 0 matrix; The pre-trained weights are fixed, and the dimension reduction matrix and the dimension increase matrix are trained based on the instruction features to obtain fine-tuning weights; The fine-tuning weights are added to the pre-training weights to fine-tune the pre-training model into the logistics market information model.

5. The method for recommending logistics solutions according to claim 2, characterized in that: Before predicting the logistics order quantity of the logistics user based on the child-generation tiered discount and the user characteristics of the logistics user, the method further includes: The deep neural network model is trained based on the collected historical user performance characteristics and the corresponding contract step discounts to obtain the performance information model.

6. The method for recommending logistics solutions according to claim 5, characterized in that: The deep neural network model is trained based on the collected historical user performance characteristics and the corresponding contract step discounts to obtain the performance information model, including: Input the historical user performance characteristics and the corresponding contract-based tiered discounts into the input layer of the deep neural network model, and output an input vector; Input the input vector into the encoding layer for encoding to obtain an embedded vector, wherein the embedded vector includes a logistics item embedding vector and an agreement data embedding vector; Input the embedded vector into the hidden layer to output the fulfillment prediction vector; The model parameters of the deep neural network model are trained based on the gap between the predicted vector and the actual fulfillment vector of the historical user until the training is completed to obtain the fulfillment information model.

7. The method for recommending logistics solutions according to claim 6, characterized in that: The input vector is input into the encoding layer for encoding to obtain an embedded vector, which includes a logistics item embedding vector and an agreed data embedding vector, and also includes: The historical user fulfillment features are processed in buckets, wherein if the feature value of the historical user fulfillment feature is less than or equal to the feature threshold, the input vector corresponding to the historical user fulfillment feature is allocated to the corresponding bucket for processing; if the feature value of the historical user fulfillment feature is greater than the feature threshold, the input vector corresponding to the historical user fulfillment feature is allocated to the same bucket for processing.

8. The method for recommending logistics solutions according to claim 2, characterized in that: Before predicting the logistics order quantity of the logistics user based on the child-generation tiered discount and the user characteristics of the logistics user, the method further includes: Inputting the historical logistics user portrait features into the logistics market information model, and outputting a first historical feature vector; Inputting historical user performance characteristics and corresponding contract-signing tiered discounts into the performance information model, and outputting a second historical feature vector; Splicing the first historical feature vector and the second historical feature vector to obtain a historical splicing feature; Input the historical splicing features into a regression model to output a single quantity prediction value; The loss between the order quantity prediction value and the actual order quantity value of the historical logistics user is calculated based on the mean square error loss function to update the model parameters of the regression model based on the loss until the training is completed to obtain the order quantity prediction model.

9. The method for recommending logistics solutions according to claim 2, characterized in that: Inputting the splicing features into an order quantity prediction model and outputting the logistics order quantity of the logistics user further includes: The splicing features generated in the first historical period are input into the order quantity prediction model, and the logistics order quantity of the logistics user in the second future period is output.

10. The method for recommending a logistics solution according to any one of claims 1 to 9, characterized in that: The reference tiered discount applicable to logistics projects is used as the first-generation tiered discount, and iterative optimization is performed, including: Predicting the logistics order quantity of the logistics user based on the reference step discount as a fitness value; Selecting a preferred step discount among the reference step discounts based on the fitness value; A crossover and / or mutation operation is performed on the preferred step discount to obtain the child step discount.

11. A logistics plan recommendation device, characterized in that: include: An iterative optimization module is used to use the reference step discount applicable to the logistics project as the primary step discount for iterative optimization, wherein a sub-generation step discount is obtained after completing one round of iteration, and the logistics order quantity of the logistics user is predicted based on the sub-generation step discount and the user characteristics of the logistics user; A determination module is used to detect that the logistics order quantity predicted in subsequent iterative optimization rounds is less than or equal to the logistics order quantity predicted in the first round in at least two adjacent iterative optimization rounds, stop the iterative optimization, and determine the child-generation step discount obtained in the first round as the recommended logistics plan.

12. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; Wherein, the processor is configured to execute the logistics solution recommendation method described in any one of claims 1 to 10 by executing the executable instructions.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the logistics solution recommendation method described in any one of claims 1 to 10 is implemented.

14. A computer program product having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the logistics solution recommendation method described in any one of claims 1 to 10 is implemented.

Images