Logistics resource allocation method and device for multi-dimensional dynamic feature engineering, equipment and storage medium

Abstract

The invention relates to the technical field of logistics, and discloses a logistics resource allocation method for multi-dimensional dynamic feature engineering, which comprises the following steps: acquiring multi-source data for acquiring various types of original data of a user, and fusing the original data; calculating a dynamic feature, and generating a real-time updated feature based on a preset algorithm rule; dividing potential levels, namely dividing the potential levels of the users based on dynamic characteristics; generating a user incentive strategy, inputting user features and resource states according to the deep reinforcement learning model, and outputting a user personalized incentive strategy; and optimizing resource allocation, establishing a mixed integer programming model, and solving an optimal resource allocation scheme by taking overall profit maximization as a target. According to the invention, the transformation of distribution from static extensive distribution to dynamic accurate distribution is realized, the limitation of dependence on static historical data is overcome, and real-time perception of user demands, market environments and behavior preferences is realized; and the marginal benefits and the return on investment of resource release are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of logistics, and in particular to a multi-dimensional dynamic feature engineering logistics resource allocation method, device, equipment and storage medium. BACKGROUND

[0002] The current user stratification strategy of the logistics industry mainly relies on static indicators such as transaction amount and purchase frequency to divide users, ignoring the dynamic potential and demand of users and other key characteristics. Dividing users according to static indicators has the following problems: insufficient identification of potential users, multi-dimensional dynamic feature engineering logistics resource allocation relies on manual experience, resulting in insufficient resources for high-value users, and relatively excessive resources for low-efficiency users, local resource squeeze, because user data analysis is limited to basic transaction data, and the value of user behavior data and external data is ignored, resulting in that the traditional clustering model cannot realize real-time updating of user grouping.

[0003] Therefore, the prior art still needs to be improved and developed. SUMMARY

[0004] The present application provides a multi-dimensional dynamic feature engineering logistics resource allocation method, device, equipment and storage medium, which is used for accurate identification and effective operation of high-potential and high-value users, and improves resource utilization and response speed, without relying on manual experience to realize real-time updating of user grouping.

[0005] The first aspect of the present application provides a multi-dimensional dynamic feature engineering logistics resource allocation method, which comprises: Obtaining multi-source data is used to obtain various types of original data of users, and the original data is fused and processed; Calculating dynamic features, generating real-time updated features based on preset algorithm rules; Dividing potential levels, dividing the potential levels of the users based on the dynamic features, and generating a visual atlas; Generating user incentive strategies, inputting user features and resource states according to a deep reinforcement learning model, and outputting user personalized incentive strategies; Optimizing resource allocation, establishing a mixed integer programming model, taking the maximization of overall profit as the target, and solving the optimal resource allocation scheme under the constraints of budget and transportation capacity.

[0006] Optionally, in the first implementation manner of the first aspect of the present application, the multi-source data is obtained to obtain various types of original data of users, and the original data is fused and processed, comprising: Respectively obtaining internal data, external data and behavior data of the users; Data format unification, outlier cleaning and spatio-temporal dimension alignment processing are performed on the internal data, external data and behavior data; The internal data includes order data, position trajectory data and service interaction data; the external data includes meteorological influence data, regional economic data and competitor pricing data; and the behavior data includes user access frequency data and function interaction preference data.

[0007] Optionally, in the second implementation manner of the first aspect, the calculation of the dynamic feature includes: The real-time feature engine is constructed to calculate the cost elasticity index CE and the service sensitivity SS. The calculation formula of the cost elasticity index CE is: S_p represents the historical price sensitivity, C_i represents the regional competition intensity coefficient, and w_i is the weight coefficient, and d_i is the competitor discount strength. The calculation formula of the service sensitivity SS is: T_r represents the complaint response time requirement, F_c represents the historical claim frequency, and F_s represents the customer service exclusive service request frequency. The time decay mechanism is set, and the feature weight update formula is: W_t represents the feature weight at time t, W_0 represents the initial weight, λ is the decay coefficient, and t is the time interval. The automatic feature derivation is performed, and the high-order combined feature is generated through feature cross.

[0008] Optionally, in the third implementation manner of the first aspect, the division of the potential level includes: The user potential level is preset. The improved spectral clustering and isolated forest algorithm are used to divide the user potential level based on the dynamic feature. The similarity matrix is constructed, and the calculation formula is: X_i is the numerical feature vector, and Z_i is the category feature vector. The isolated forest algorithm is used to filter abnormal users. The user clustering visualization graph is generated through the t-SNE dimension reduction algorithm.

[0009] Optionally, in the fourth implementation manner of the first aspect, the generation of the user incentive includes: Generating a personalized incentive strategy based on a deep reinforcement learning model; The state space of the deep reinforcement learning model is composed of a user feature vector and a real-time resource state, and the real-time resource state includes a regional transport capacity remaining amount and a subsidy budget balance. The action space of the deep reinforcement learning model is a structured output, including a price discount rate, a distribution priority, and whether to allocate a dedicated customer service. The reward function of the deep reinforcement learning model is designed as: Wherein γ1, γ2, γ3 are weight coefficients, β1 is the single quantity, β2 is the satisfaction score, and β3 is the subsidy cost. The deep reinforcement learning model is pre-trained using historical data, and user responses are simulated through a SMARTS simulation environment.

[0010] Optionally, in the fifth implementation manner of the first aspect of the present application, the optimization of resource allocation, the establishment of a mixed integer programming model, the maximization of the overall profit as the target, and the solving of the optimal resource allocation scheme under the budget and transport capacity constraints include: Establishing a mixed integer programming model; The objective function of the mixed integer programming model is: Wherein x i is a binary decision variable for whether to incentivize customer i; the constraint conditions of the mixed integer programming model include budget constraints, transport capacity constraints, and service balance; The mixed integer programming model is solved using a solver.

[0011] Optionally, in the sixth implementation manner of the first aspect of the present application, the method further includes strategy deployment and dynamic optimization visualization: The personalized incentive strategy is issued to the business system through an interface to trigger resource scheduling; Monitoring and feedback effects, presetting high-potential indicators and thresholds, and determining whether the user reaches the high-potential indicators; the high-potential indicators include delivery volume growth, subsidy return rate, and user retention rate; Model iteration, updating feature weights by using incremental learning, retraining the clustering model at regular intervals, and evaluating through an A / B testing platform.

[0012] The second aspect of the present application provides a multi-dimensional dynamic feature engineering logistics resource allocation device, which includes: A multi-source data acquisition module is used to acquire various types of original data of users and fuse the original data; A dynamic feature calculation module is used to generate real-time updated features based on preset algorithm rules; The potential level classification module is used to classify the users into potential levels based on dynamic characteristics and generate a visual map. The module for generating user incentive strategies is used to output personalized user incentive strategies based on the deep reinforcement learning model, taking user characteristics and resource status as input. The resource allocation optimization module is used to establish a mixed integer programming model, aiming to maximize overall profit, and to solve for the optimal resource allocation scheme under budget and capacity constraints.

[0013] A third aspect of the present invention provides a logistics resource allocation device for multidimensional dynamic feature engineering, comprising a memory and at least one processor, wherein the memory stores computer-readable instructions; The at least one processor invokes the computer-readable instructions in the memory to execute the various steps of the logistics resource allocation method of the multidimensional dynamic feature engineering described above.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the various steps of the logistics resource allocation method of the multidimensional dynamic feature engineering described above.

[0015] The technical solution provided by this invention constructs a real-time feature engine through multi-source data fusion and dynamic feature engineering to accurately perceive user needs and market environment, realize the accurate identification of high-value users and optimized resource allocation, and innovatively combine deep reinforcement learning and operations research optimization models to maximize profits under global resource constraints while ensuring personalized services. Furthermore, it continuously optimizes decision-making efficiency through a self-evolving closed-loop system, ultimately significantly reducing operating costs, improving resource utilization efficiency and user satisfaction, and achieving the core goal of cost reduction and revenue increase for enterprises. Attached Figure Description

[0016] Figure 1 A first flowchart of a logistics resource allocation method for multi-dimensional dynamic feature engineering provided in an embodiment of the present invention; Figure 2 A second flowchart of the logistics resource allocation method for multi-dimensional dynamic feature engineering provided in this embodiment of the invention; Figure 3 A third flowchart of the logistics resource allocation method for multi-dimensional dynamic feature engineering provided in an embodiment of the present invention; Figure 4 A fourth flowchart of the logistics resource allocation method for multi-dimensional dynamic feature engineering provided in this embodiment of the invention; Figure 5 The fifth flowchart of the logistics resource allocation method for multi-dimensional dynamic feature engineering provided in this embodiment of the invention; Figure 6 The sixth flowchart of the logistics resource allocation method for multi-dimensional dynamic feature engineering provided in this embodiment of the invention; Figure 7 A schematic diagram of the structure of the logistics resource allocation device for multi-dimensional dynamic feature engineering provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of a logistics resource allocation device for multi-dimensional dynamic feature engineering provided in an embodiment of the present invention. Detailed Implementation

[0017] This invention provides a method, apparatus, equipment, and storage medium for allocating logistics resources using multidimensional dynamic feature engineering. The method is used to establish dynamic user profiles and manage users based on these profiles.

[0018] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The first embodiment of a logistics resource allocation method for multidimensional dynamic feature engineering in this invention includes: S101. Obtain multi-source data, used to obtain various types of raw data from users and perform fusion processing on the raw data; S102. Calculate dynamic features and generate real-time updated features based on preset algorithm rules; S103. Classify potential levels: Classify the users into potential levels based on dynamic characteristics and generate a visual map. S104. Generate user incentive strategies. Based on the deep reinforcement learning model, input user characteristics and resource status, and output personalized user incentive strategies. S105. Optimize resource allocation by establishing a mixed integer programming model. With the goal of maximizing overall profit, solve for the optimal resource allocation scheme under budget and capacity constraints.

[0020] This invention, through multi-source data fusion and dynamic feature engineering, constructs a real-time feature engine to accurately perceive user needs and market environment, enabling precise identification of high-value users and optimized resource allocation. It innovatively combines deep reinforcement learning and operations research optimization models to maximize profits under global resource constraints while ensuring personalized services. Furthermore, it continuously optimizes decision-making efficiency through a self-evolving closed-loop system, ultimately significantly reducing operating costs, improving resource utilization efficiency and user satisfaction, and achieving the core goal of cost reduction and revenue increase for enterprises.

[0021] Please see Figure 2 The second embodiment of the logistics resource allocation method for multidimensional dynamic feature engineering in this invention includes: The process of acquiring multi-source data involves obtaining various types of raw data from the user and performing fusion processing on the raw data, including: S201. Obtain the user's internal data, external data, and behavioral data respectively; S202, Perform data format unification, outlier cleaning, and spatiotemporal dimension alignment on the internal data, external data, and behavioral data; The internal data includes order data, location trajectory data, and service interaction data; the external data includes meteorological impact data, regional economic data, and competitor pricing data; and the behavioral data includes user access frequency data and function interaction preference data.

[0022] This invention, through its clearly defined internal data (including order data, location trajectory data, and service interaction data), accurately reflects users' historical transaction habits, logistics route preferences, and service quality needs. External data (including meteorological impact data, regional economic data, and competitor pricing data) effectively perceives the impact of environmental factors on logistics efficiency, regional differences in consumption capacity, and market competition. Behavioral data (including user access frequency data and functional interaction preference data) deeply mines users' potential needs, preferences, and activity characteristics. Through the multi-dimensional collaboration of these three types of data, a comprehensive data system covering users' intrinsic attributes, external environment, and behavioral characteristics is constructed. After processing, this provides a high-purity, highly correlated data foundation for subsequent dynamic feature engineering, significantly improving the accuracy of user profiles and the reliability of decision support.

[0023] Please see Figure 3 A third embodiment of a logistics resource allocation method for multidimensional dynamic feature engineering in this invention includes: calculating dynamic features, based on preset algorithm rules, to generate real-time updated features, including: S301. Construct a real-time feature engine to calculate the cost elasticity index (CE) and service sensitivity (SS). The formula for calculating the cost elasticity index CE is as follows: S_p represents historical price sensitivity, and C_i represents the regional competition intensity coefficient. w_i is the weighting coefficient, and d_i is the discount level of the competing products; The formula for calculating the service sensitivity SS is: T_r represents the timeliness requirement for complaint response, F_c represents the frequency of historical claims, and F_s represents the frequency of customer service requests. S302. Set a time decay mechanism and use the feature weight update formula: Where W_t represents the feature weight at time t, W_0 represents the initial weight, λ is the decay coefficient, and t is the time interval; S303. Perform automated feature derivation and generate higher-order combined features through feature cross-interaction.

[0024] This invention addresses the shortcomings of traditional feature engineering by constructing a real-time feature engine to quantify two key indicators: cost elasticity index and service sensitivity. This transforms user price preferences and service needs into calculable feature values. Furthermore, it dynamically adjusts feature weights through a time decay mechanism to ensure the model always reflects the latest user behavior characteristics. Finally, it generates higher-order combined features through automated feature derivation and feature cross-pollination to deeply mine potential user behavior patterns. These three techniques collectively solve the problems of lag and one-sidedness in traditional feature engineering, providing real-time, accurate, and in-depth feature data support for subsequent user segmentation and resource allocation.

[0025] Please see Figure 4 A fourth embodiment of a logistics resource allocation method based on multidimensional dynamic feature engineering in this invention includes: classifying potential levels, classifying users by potential levels based on dynamic features, and generating a visual map, including: S401, Preset user potential level; S402. Using an improved spectral clustering and isolated forest algorithm, the user potential levels are divided based on dynamic features; S403. Construct the similarity matrix, the calculation formula is as follows: , where X_i is the numerical feature vector and Z_i is the categorical feature vector; S404. Use the isolated forest algorithm to filter out abnormal users; S405. A user group visualization map will be generated using the t-SNE dimensionality reduction algorithm.

[0026] This invention improves the spectral clustering algorithm to handle complex data distribution characteristics, achieving refined segmentation of user groups; it combines the isolated forest algorithm to effectively filter outlier data points, avoiding interference from noisy data on the segmentation results; it employs a similarity matrix calculation method that integrates numerical and categorical features to comprehensively improve the accuracy and robustness of user segmentation; and finally, it uses t-SNE dimensionality reduction visualization technology to intuitively present the segmentation results in the high-dimensional feature space, enabling operators to quickly understand the structural characteristics of user groups and providing a reliable decision-making basis for subsequent precise resource allocation. This effectively solves the problems of ambiguous boundaries, abnormal interference, and poor interpretability in traditional user segmentation methods.

[0027] Please see Figure 5 The fifth embodiment of a logistics resource allocation method for multidimensional dynamic feature engineering in this invention includes: S501. Generate personalized incentive strategies based on deep reinforcement learning models; The state space of the deep reinforcement learning model consists of user feature vectors and real-time resource status, which includes the remaining regional transportation capacity and the remaining subsidy budget. The action space of the deep reinforcement learning model is a structured output, which includes price discount rate, delivery priority, and whether a dedicated customer service representative is assigned. The reward function of the deep reinforcement learning model is designed as follows: Where γ1, γ2, and γ3 are weighting coefficients, β1 is the number of customers per order, β2 is the satisfaction score, and β3 is the subsidy cost; S502. Pre-train the deep reinforcement learning model using historical data and simulate user responses using the SMARTS simulation environment.

[0028] This invention, through the construction of a state space containing user feature vectors and real-time resource status, ensures that the deep reinforcement learning model can make decisions based on multi-dimensional dynamic features and the system's real-time carrying capacity. By designing a structured action space including price discount rates, delivery priorities, and dedicated customer service allocation, it achieves the coordinated output of multi-dimensional incentive strategies. By integrating a multi-objective reward function that considers order volume, satisfaction scores, and subsidy costs, it improves business revenue while also taking into account user experience and cost control. Combined with historical data pre-training and SMARTS simulation environment simulation, it significantly reduces the trial-and-error costs in the real environment, ultimately achieving the linkage optimization of personalized incentive strategies and global resource status. This effectively solves the problems of weak targeting, low resource matching, and poor overall efficiency inherent in traditional incentive strategies.

[0029] Please see Figure 6 The sixth embodiment of a logistics resource allocation method for multidimensional dynamic feature engineering in this invention includes: The optimization of resource allocation involves establishing a mixed-integer programming model to maximize overall profit and solve for the optimal resource allocation scheme under budget and capacity constraints. This includes: S601. Establish a mixed integer programming model; The objective function of the mixed-integer programming model is: , where x i The variable is a binary decision variable for whether to incentivize customer i; the constraints of the mixed integer programming model include: budget constraints, capacity constraints, and service balance. S602. Solve the mixed integer programming model using a solver.

[0030] In this embodiment of the invention, a mixed-integer programming model with the goal of maximizing overall profit is established, transforming the personalized strategies generated by deep reinforcement learning into globally optimal decisions. Binary decision variables are used to accurately represent the customer's incentive state, and multiple business constraints such as budget constraints, capacity constraints, and service balance are strictly met through mathematical modeling. Optimization calculations are performed using a solver to ensure that the optimal resource allocation scheme is obtained quickly under complex constraints, achieving a seamless connection from personalized strategies to global resource optimization, and effectively solving the contradiction between micro-incentives and macro-resource allocation.

[0031] The method also includes strategy deployment and dynamic optimization visualization: Personalized incentive strategies are sent to the business system via an interface to trigger resource scheduling. The system monitors and provides feedback on the results, presets high-potential indicators and thresholds, and determines whether users have reached the high-potential indicators. The high-potential indicators include the increase in shipment volume, the rate of return on subsidies, and the user retention rate. The model is iterated by updating feature weights through incremental learning, retraining the clustering model periodically, and evaluating it through an A / B testing platform.

[0032] This invention employs automated strategy delivery via interfaces to ensure rapid implementation of decision-making results; it establishes an effective and scientific strategy performance evaluation system by pre-setting high-potential indicators across multiple dimensions, such as shipment volume growth, subsidy return rate, and user retention rate; it uses incremental learning to dynamically update feature weights, combined with a timed retraining mechanism to ensure the model continuously adapts to changes in data distribution; and finally, it verifies the strategy performance through an A / B testing platform, forming a complete closed-loop system that enables continuous self-optimization and dynamic tuning of logistics resource allocation strategies, significantly improving the system's long-term adaptability and operational efficiency.

[0033] The above describes the logistics resource allocation method of multi-dimensional dynamic feature engineering in the embodiments of the present invention. The following describes the apparatus in the embodiments of the present invention. Please refer to [link / reference]. Figure 7The implementation methods of the logistics resource allocation device for multidimensional dynamic feature engineering in this invention include: The multi-source data acquisition module 701 is used to acquire various types of raw data from users and perform fusion processing on the raw data; The dynamic feature calculation module 702 is used to generate real-time updated features based on preset algorithm rules; The potential level classification module 703 is used to classify the user's potential level based on dynamic characteristics and generate a visual map; The user incentive strategy generation module 704 is used to generate personalized incentive strategies for users based on the deep reinforcement learning model, inputting user characteristics and resource status. The resource allocation optimization module 705 is used to establish a mixed integer programming model to solve for the optimal resource allocation scheme under budget and capacity constraints with the goal of maximizing overall profit.

[0034] In some embodiments, the multi-source data acquisition module 701 includes: The acquisition unit 7011 is used to acquire the user's internal data, external data, and behavioral data respectively. Data processing unit 7012 is used to perform data format unification, outlier cleaning, and spatiotemporal dimension alignment on the internal data, external data, and behavioral data; The internal data includes order data, location trajectory data, and service interaction data; the external data includes meteorological impact data, regional economic data, and competitor pricing data; and the behavioral data includes user access frequency data and function interaction preference data.

[0035] This invention, through its clearly defined internal data (including order data, location trajectory data, and service interaction data), accurately reflects users' historical transaction habits, logistics route preferences, and service quality needs. External data (including meteorological impact data, regional economic data, and competitor pricing data) effectively perceives the impact of environmental factors on logistics efficiency, regional differences in consumption capacity, and market competition. Behavioral data (including user access frequency data and functional interaction preference data) deeply mines users' potential needs, preferences, and activity characteristics. Through the multi-dimensional collaboration of these three types of data, a comprehensive data system covering users' intrinsic attributes, external environment, and behavioral characteristics is constructed. After processing, this provides a high-purity, highly correlated data foundation for subsequent dynamic feature engineering, significantly improving the accuracy of user profiles and the reliability of decision support.

[0036] In some embodiments, the dynamic feature calculation module 702 includes: The first building unit 7021 is used to build a real-time feature engine and calculate the cost elasticity index CE and service sensitivity SS. The formula for calculating the cost elasticity index CE is as follows: S_p represents historical price sensitivity, and C_i represents the regional competition intensity coefficient. w_i is the weighting coefficient, and d_i is the discount level of the competing products; The formula for calculating the service sensitivity SS is: T_r represents the timeliness requirement for complaint response, F_c represents the frequency of historical claims, and F_s represents the frequency of customer service requests. Unit 7022 is used to configure the time decay mechanism, which employs the feature weight update formula: Where W_t represents the feature weight at time t, W_0 represents the initial weight, λ is the decay coefficient, and t is the time interval; The first generation unit 7023 performs automated feature derivation and generates higher-order combined features through feature cross-interaction.

[0037] This invention addresses the shortcomings of traditional feature engineering by constructing a real-time feature engine to quantify two key indicators: cost elasticity index and service sensitivity. This transforms user price preferences and service needs into calculable feature values. Furthermore, it dynamically adjusts feature weights through a time decay mechanism to ensure the model always reflects the latest user behavior characteristics. Finally, it generates higher-order combined features through automated feature derivation and feature cross-pollination to deeply mine potential user behavior patterns. These three techniques collectively solve the problems of lag and one-sidedness in traditional feature engineering, providing real-time, accurate, and in-depth feature data support for subsequent user segmentation and resource allocation.

[0038] In some embodiments, the potential level classification module 703 includes: Preset unit 7031 is used to preset user potential levels; The partitioning unit 7032 uses an improved spectral clustering and isolated forest algorithm to partition the user potential level based on dynamic features; The second building unit 7033 is used to construct the similarity matrix, and the calculation formula is as follows: , where X_i is the numerical feature vector and Z_i is the categorical feature vector; Filtering unit 7034 is used to filter out abnormal users using the isolated forest algorithm; The second generation unit 7035 is used to generate a user group visualization map using the t-SNE dimensionality reduction algorithm.

[0039] This invention improves the spectral clustering algorithm to handle complex data distribution characteristics, achieving refined segmentation of user groups; it combines the isolated forest algorithm to effectively filter outlier data points, avoiding interference from noisy data on the segmentation results; it employs a similarity matrix calculation method that integrates numerical and categorical features to comprehensively improve the accuracy and robustness of user segmentation; and finally, it uses t-SNE dimensionality reduction visualization technology to intuitively present the segmentation results in the high-dimensional feature space, enabling operators to quickly understand the structural characteristics of user groups and providing a reliable decision-making basis for subsequent precise resource allocation. This effectively solves the problems of ambiguous boundaries, abnormal interference, and poor interpretability in traditional user segmentation methods.

[0040] In some embodiments, the user incentive strategy generation module 704 includes: The third generation unit 7041 is used to generate personalized incentive strategies based on a deep reinforcement learning model. The state space of the deep reinforcement learning model consists of user feature vectors and real-time resource status, which includes the remaining regional transportation capacity and the remaining subsidy budget. The action space of the deep reinforcement learning model is a structured output, which includes price discount rate, delivery priority, and whether a dedicated customer service representative is assigned. The reward function of the deep reinforcement learning model is designed as follows: Where γ1, γ2, and γ3 are weighting coefficients, β1 is the number of customers per order, β2 is the satisfaction score, and β3 is the subsidy cost; Training unit 7042 is used to pre-train the deep reinforcement learning model using historical data and simulate user responses through the SMARTS simulation environment.

[0041] This invention, through the construction of a state space containing user feature vectors and real-time resource status, ensures that the deep reinforcement learning model can make decisions based on multi-dimensional dynamic features and the system's real-time carrying capacity. By designing a structured action space including price discount rates, delivery priorities, and dedicated customer service allocation, it achieves the coordinated output of multi-dimensional incentive strategies. By integrating a multi-objective reward function that considers order volume, satisfaction scores, and subsidy costs, it improves business revenue while also taking into account user experience and cost control. Combined with historical data pre-training and SMARTS simulation environment simulation, it significantly reduces the trial-and-error costs in the real environment, ultimately achieving the linkage optimization of personalized incentive strategies and global resource status. This effectively solves the problems of weak targeting, low resource matching, and poor overall efficiency inherent in traditional incentive strategies.

[0042] In some embodiments, the optimized resource allocation module 705 includes: Establish unit 7051 for building a mixed integer programming model; The objective function of the mixed-integer programming model is: , where x i The variable is a binary decision variable for whether to incentivize customer i; the constraints of the mixed integer programming model include: budget constraints, capacity constraints, and service balance. The solver unit 7052 is used to solve the mixed integer programming model using a solver.

[0043] In this embodiment of the invention, a mixed-integer programming model with the goal of maximizing overall profit is established, transforming the personalized strategies generated by deep reinforcement learning into globally optimal decisions. Binary decision variables are used to accurately represent the customer's incentive state, and multiple business constraints such as budget constraints, capacity constraints, and service balance are strictly met through mathematical modeling. Optimization calculations are performed using a solver to ensure that the optimal resource allocation scheme is obtained quickly under complex constraints, achieving a seamless connection from personalized strategies to global resource optimization, and effectively solving the contradiction between micro-incentives and macro-resource allocation.

[0044] The method also includes strategy deployment and dynamic optimization visualization: Personalized incentive strategies are sent to the business system via an interface to trigger resource scheduling. The system monitors and provides feedback on the results, presets high-potential indicators and thresholds, and determines whether users have reached the high-potential indicators. The high-potential indicators include the increase in shipment volume, the rate of return on subsidies, and the user retention rate. The model is iterated by updating feature weights through incremental learning, retraining the clustering model periodically, and evaluating it through an A / B testing platform.

[0045] This invention employs automated strategy delivery via interfaces to ensure rapid implementation of decision-making results; it establishes an effective and scientific strategy performance evaluation system by pre-setting high-potential indicators across multiple dimensions, such as shipment volume growth, subsidy return rate, and user retention rate; it uses incremental learning to dynamically update feature weights, combined with a timed retraining mechanism to ensure the model continuously adapts to changes in data distribution; and finally, it verifies the strategy performance through an A / B testing platform, forming a complete closed-loop system that enables continuous self-optimization and dynamic tuning of logistics resource allocation strategies, significantly improving the system's long-term adaptability and operational efficiency.

[0046] Figure 7 The structure of the logistics resource allocation device for multidimensional dynamic feature engineering shown does not constitute a limitation on the logistics resource allocation device for multidimensional dynamic feature engineering, and can realize the steps of the logistics resource allocation method for multidimensional dynamic feature engineering provided in the above method embodiments.

[0047] above Figure 7The logistics resource allocation device for multi-dimensional dynamic feature engineering in this embodiment of the invention is described in detail from the perspective of modular functional entities. The logistics resource allocation equipment for multi-dimensional dynamic feature engineering in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0048] Figure 8 This is a schematic diagram of the structure of a logistics resource allocation device for multi-dimensional dynamic feature engineering provided by an embodiment of the present invention. The device 800 can vary significantly due to different configurations or performance, and may include one or more central processing units (CPUs) 810 (e.g., one or more processors) and a memory 820, and one or more storage media 830 (e.g., one or more mass storage devices) for storing application programs 833 or data 832. The memory 820 and storage media 830 can be temporary or persistent storage. The program stored in the storage media 830 may include one or more modules (not shown), each module including a series of instruction operations on the device 800. Furthermore, the processor 810 may be configured to communicate with the storage media 830 and execute the series of instruction operations in the storage media on the device 800.

[0049] Device 800 may also include one or more power supplies 840, one or more wired or wireless network interfaces 850, one or more input / output interfaces 860, and / or one or more operating systems 831, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc.

[0050] This invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of a logistics resource allocation method based on multi-dimensional dynamic feature engineering.

[0051] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0052] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0053] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A logistics resource allocation method based on multidimensional dynamic feature engineering, characterized in that, The logistics resource allocation method of the multidimensional dynamic feature engineering includes: Acquire multi-source data to obtain various types of raw data from users and perform fusion processing on the raw data; Calculate dynamic features and generate real-time updated features based on preset algorithm rules; The potential levels are divided based on dynamic characteristics, and a visual map is generated. Generate user incentive strategies by taking user characteristics and resource status as inputs based on a deep reinforcement learning model, and outputting personalized user incentive strategies. To optimize resource allocation, a mixed-integer programming model is established. With the goal of maximizing overall profit, the optimal resource allocation scheme is solved under budget and capacity constraints.

2. The logistics resource allocation method for multi-dimensional dynamic feature engineering according to claim 1, characterized in that, Acquire multi-source data, used to obtain various types of raw data from users, and perform fusion processing on the raw data, including: The user's internal data, external data, and behavioral data are acquired respectively. The internal data, external data, and behavioral data are subjected to data format unification, outlier cleaning, and spatiotemporal dimension alignment. The internal data includes order data, location trajectory data, and service interaction data; the external data includes meteorological impact data, regional economic data, and competitor pricing data; and the behavioral data includes user access frequency data and function interaction preference data.

3. The logistics resource allocation method for multidimensional dynamic feature engineering according to claim 2, characterized in that, The calculation of dynamic features, based on preset algorithm rules, generates features that are updated in real time, including: Build a real-time feature engine to calculate the cost elasticity index (CE) and service sensitivity index (SS); The formula for calculating the cost elasticity index CE is as follows: S_p represents historical price sensitivity, and C_i represents the regional competition intensity coefficient. w_i is the weighting coefficient, and d_i is the discount level of the competing products; The formula for calculating the service sensitivity SS is: T_r represents the timeliness requirement for complaint response, F_c represents the frequency of historical claims, and F_s represents the frequency of customer service requests. A time decay mechanism is set up, and the feature weight update formula is used: Where W_t represents the feature weight at time t, W_0 represents the initial weight, λ is the decay coefficient, and t is the time interval; Automated feature derivation is performed, and higher-order combined features are generated through feature cross-referencing.

4. The logistics resource allocation method for multi-dimensional dynamic feature engineering according to claim 3, characterized in that, The process of classifying potential levels, based on dynamic features, involves classifying users into potential levels and generating a visual graph, including: Preset user potential levels; An improved spectral clustering and isolated forest algorithm is used to classify the user potential levels based on dynamic features; The similarity matrix is ​​constructed using the following formula: , where X_i is the numerical feature vector and Z_i is the categorical feature vector; Use the isolated forest algorithm to filter out abnormal users; The t-SNE dimensionality reduction algorithm will be used to generate a user group visualization map.

5. The logistics resource allocation method for multi-dimensional dynamic feature engineering according to claim 3, characterized in that, The generation of user incentives, based on a deep reinforcement learning model, takes user characteristics and resource status as input and outputs a personalized user incentive strategy, including: Generate personalized incentive strategies based on deep reinforcement learning models; The state space of the deep reinforcement learning model consists of user feature vectors and real-time resource status, which includes the remaining regional transportation capacity and the remaining subsidy budget. The action space of the deep reinforcement learning model is a structured output, which includes price discount rate, delivery priority, and whether a dedicated customer service representative is assigned. The reward function of the deep reinforcement learning model is designed as follows: Where γ1, γ2, and γ3 are weighting coefficients, β1 is the number of customers per order, β2 is the satisfaction score, and β3 is the subsidy cost; The deep reinforcement learning model is pre-trained using historical data, and user responses are simulated using the SMARTS simulation environment.

6. The logistics resource allocation method for multi-dimensional dynamic feature engineering according to claim 5, characterized in that, The optimization of resource allocation involves establishing a mixed-integer programming model to maximize overall profit and solve for the optimal resource allocation scheme under budget and capacity constraints. This includes: Establish a mixed-integer programming model; The objective function of the mixed-integer programming model is: , where x i The variable is a binary decision variable for whether to incentivize customer i; the constraints of the mixed integer programming model include: budget constraints, capacity constraints, and service balance. The mixed integer programming model is solved using a solver.

7. The logistics resource allocation method for multidimensional dynamic feature engineering according to claim 1, characterized in that, The method also includes strategy deployment and dynamic optimization visualization: Personalized incentive strategies are sent to the business system via an interface to trigger resource scheduling. The system monitors and provides feedback on the results, presets high-potential indicators and thresholds, and determines whether users have reached the high-potential indicators. The high-potential indicators include the increase in shipment volume, the rate of return on subsidies, and the user retention rate. The model is iterated by updating feature weights through incremental learning, retraining the clustering model periodically, and evaluating it through an A / B testing platform.

8. A logistics resource allocation device based on multi-dimensional dynamic feature engineering, characterized in that, include: The multi-source data acquisition module is used to acquire various types of raw data from users and perform fusion processing on the raw data; The dynamic feature calculation module is used to generate real-time updated features based on preset algorithm rules; The potential level classification module is used to classify the users into potential levels based on dynamic characteristics and generate a visual map. The module for generating user incentive strategies is used to output personalized user incentive strategies based on the deep reinforcement learning model, taking user characteristics and resource status as input. The resource allocation optimization module is used to establish a mixed integer programming model, aiming to maximize overall profit, and to solve for the optimal resource allocation scheme under budget and capacity constraints.

9. A logistics resource allocation device based on multi-dimensional dynamic feature engineering, characterized in that, It includes a memory and at least one processor, wherein the memory stores computer-readable instructions; The at least one processor invokes the computer-readable instructions in the memory to perform the steps of the logistics resource allocation method of the multidimensional dynamic feature engineering as described in any one of claims 1-7.

10. A computer-readable storage medium storing computer-readable instructions thereon, characterized in that, When the computer-readable instructions are executed by a processor, they implement the various steps of the logistics resource allocation method of multidimensional dynamic feature engineering as described in any one of claims 1-7.

Images