An erp background management method and system based on a multi-user mall

Abstract

The present invention discloses an erp background management method and system based on a multi-user mall, relating to the technical field of e-commerce, including: constructing an elastic expansion mechanism based on a cloud architecture, and the elastic expansion mechanism dynamically adjusts server resources by monitoring the real-time load data of the erp background management system according to the server resource allocation strategy; adopting a machine learning model to analyze the historical data of merchants and generate behavior prediction results, and optimizing the server resource allocation strategy according to the behavior prediction results; developing an intelligent erp configuration recommendation system based on a deep reinforcement learning algorithm, and the intelligent erp configuration recommendation system automatically identifies the business models of merchants and generates personalized erp management configuration suggestions by using the behavior prediction results; integrating multi-channel payment interfaces and logistics APIs, and automatically selecting payment and logistics strategies according to the personalized erp management configuration suggestions to construct a unified order management process. The present invention significantly improves the performance, resource utilization efficiency and intelligence level of the erp background management system of the multi-user mall.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce, and specifically to an erp background management method and system based on a multi-user mall. Background Art

[0002] With the rapid development of e-commerce, multi-user malls, as a new business model, are becoming increasingly popular. To effectively manage this complex business ecosystem, the enterprise resource planning (erp) system plays a crucial role in the background management of multi-user malls. Traditional erp systems mainly focus on the resource integration and process optimization of a single enterprise. However, in a multi-user mall environment, the erp system faces more complex challenges. These challenges include, but are not limited to: dynamic allocation of multi-merchant resources, real-time synchronization of cross-platform data, personalized merchant management requirements, and processing of large-scale concurrent transactions. Existing erp background management systems often appear inadequate when dealing with these challenges, especially in aspects such as elastic resource expansion, intelligent configuration, and global order management.

[0003] Existing multi-user mall erp background management systems usually adopt static resource allocation strategies, making it difficult to adapt to the dynamic changes of mall business. At the same time, these systems lack the ability to deeply analyze and predict merchant historical data and cannot provide personalized erp configuration suggestions for merchants. In addition, in terms of order management, existing systems often separate functions such as payment and logistics, lacking a unified management process, resulting in low order processing efficiency and poor user experience. These problems seriously restrict the operation efficiency and expansion ability of multi-user malls, and there is an urgent need for a new erp background management solution. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: how to provide an erp background management method based on a multi-user mall?

[0006] To solve the above technical problems, the present invention provides the following technical solutions: An erp background management method based on a multi-user mall, including: constructing an elastic expansion mechanism based on a cloud architecture, where the elastic expansion mechanism dynamically adjusts server resources by monitoring the real-time load data of the erp background management system according to the server resource allocation strategy; using a machine learning model to analyze the historical data of merchants and generate behavior prediction results, and optimizing the server resource allocation strategy according to the behavior prediction results; developing an intelligent erp configuration recommendation system based on a deep reinforcement learning algorithm, where the intelligent erp configuration recommendation system automatically identifies the business models of merchants and generates personalized erp management configuration suggestions using the behavior prediction results; integrating multi-channel payment interfaces and logistics APIs, and automatically selecting payment and logistics strategies according to the personalized erp management configuration suggestions to construct a unified order management process.

[0007] As a preferred solution of the erp background management method based on a multi-user mall according to the present invention, where: constructing an elastic expansion mechanism based on a cloud architecture, where the elastic expansion mechanism dynamically adjusts server resources by monitoring the real-time load data of the erp background management system according to the server resource allocation strategy, includes the following steps: constructing a distributed load monitoring system based on a cloud architecture, and collecting various real-time load data of the server cluster through the distributed load monitoring system; formulating a server resource allocation strategy; the server resource allocation strategy includes a resource utilization threshold and a resource adjustment rule; establishing a multi-dimensional resource scoring model; the multi-dimensional resource scoring model calculates the server resource utilization score according to the real-time load data; comparing the server resource utilization score with the resource utilization threshold, and when the server resource utilization score exceeds the resource utilization threshold, triggering a resource adjustment; after triggering the resource adjustment, dynamically increasing or decreasing the number of server instances according to the resource adjustment rule, adjusting the calculation resource allocation, and completing the elastic expansion mechanism of the server resources.

[0008] As a preferred solution of the erp background management method based on a multi-user mall according to the present invention, wherein: each of the real-time load data corresponds to a type of resource; the resource utilization threshold includes a dynamic high threshold and a dynamic low threshold; the resource adjustment rules include a resource expansion rule, a resource contraction rule, an intelligent reservation rule, and an elastic adjustment rule; wherein, the resource expansion rule is that when the resource utilization rate of a certain type of real-time load data exceeds its dynamic high threshold for T1 consecutive minutes, resource expansion is triggered, and the expansion amount is: the resource amount of the current real-time load data * (the resource utilization rate of the current real-time load data - the dynamic high threshold) / the dynamic high threshold, rounded up, but not less than the minimum expansion unit; the resource contraction rule is that when the resource utilization rate of a certain type of real-time load data is lower than its dynamic low threshold for T2 consecutive minutes, resource contraction is triggered, and the contraction amount is: the resource amount of the current real-time load data * (the dynamic low threshold - the resource utilization rate of the current real-time load data) / the dynamic low threshold, rounded up, but not less than the minimum contraction unit; the intelligent reservation rule is that when the erp background management system predicts that the resource demand within a preset time will exceed the current capacity, that is, the total available resource amount at present, the reserved resource amount is: the predicted peak usage * (1 + safety factor) + volatility factor * historical fluctuation range; the elastic adjustment rule is that if the number of resource adjustments within the recent M hours is greater than the preset number, then the new resource utilization threshold = the original resource utilization threshold * (1 + X%); the multi-dimensional resource scoring model is to calculate the ratio of the current usage amount of each real-time load data to the maximum capacity, multiply by the weight of each real-time load data, and add up the weighted resource utilization rates of all real-time load data, and the obtained score is used as the server resource utilization score.

[0009] As a preferred solution of the erp background management method based on a multi-user mall according to the present invention, wherein: a machine learning model is used to analyze the historical data of merchants and generate a behavior prediction result, and the server resource allocation strategy is optimized according to the behavior prediction result, including the following steps: collecting and preprocessing the historical data of merchants; the historical data of merchants includes the order volume, transaction amount, access traffic, inventory change, and seasonal fluctuation; constructing a time series prediction model based on the long short-term memory network algorithm LSTM and the attention mechanism, and training the preprocessed historical data of merchants through the time series prediction model; using the time series prediction model to generate a merchant behavior prediction result; the merchant behavior prediction result includes the expected order volume, transaction amount, and access traffic; adjusting the server resource allocation strategy according to the merchant behavior prediction result; wherein, adjusting the server resource allocation strategy includes updating the resource utilization threshold, optimizing the resource adjustment rule, and formulating a resource reservation plan.

[0010] As a preferred solution of the erp background management method based on a multi-user mall according to the present invention, wherein: an intelligent erp configuration recommendation system is developed based on a deep reinforcement learning algorithm, and the intelligent erp configuration recommendation system automatically identifies the business model of a merchant and generates personalized erp management configuration suggestions by using the behavior prediction results, including the following steps: constructing a business model feature matrix of the merchant including the merchant behavior prediction results, historical erp usage data, and product category information; constructing a multi-task deep neural network model, which includes a shared layer and multiple task-specific layers, wherein the shared layer is used to extract the general features of the merchant business, and the multiple task-specific layers are respectively used for merchant business model identification and epr configuration parameter prediction; adopting a transfer learning method to initialize the parameters of the multi-task deep neural network model by using a pre-trained e-commerce domain knowledge graph; training the initialized multi-task deep neural network model based on the business model feature matrix of the merchant to obtain a trained multi-task deep neural network model for merchant business model identification and epr configuration parameter prediction; taking the business model feature matrix of the merchant as input, passing it into the trained multi-task deep neural network model, extracting the general features of the merchant business through the shared layer, and then respectively using the task-specific layer for merchant business model identification to identify the merchant business model, and using the task-specific layer for epr configuration parameter prediction to generate corresponding personalized erp management configuration suggestions.

[0011] As a preferred solution of the erp background management method based on a multi-user mall according to the present invention, wherein: integrating multiple-channel payment interfaces and logistics APIs, and automatically selecting payment and logistics strategies according to the personalized erp management configuration suggestions to construct a unified order management process, including the following steps: integrating APIs of multiple third-party payment platforms and logistics service providers; establishing a payment and logistics strategy database, and storing the payment and logistics parameters in the personalized erp management configuration suggestions into the logistics strategy database; constructing an order processing module; constructing an order status tracking system, which can obtain the payment status and logistics information of the order in real time; constructing a unified order management interface.

[0012] As a preferred solution of the erp background management method based on a multi-user mall according to the present invention, wherein: the real-time load data includes CPU, memory, network bandwidth, and disk I / O.

[0013] To further solve the above technical problems, the present invention provides the following technical solutions: An erp background management system based on a multi-user mall, including: an elastic resource management module, used to build an elastic expansion mechanism based on a cloud architecture, and the elastic expansion mechanism dynamically adjusts server resources by monitoring the real-time load data of the erp background management system according to the server resource allocation strategy; a prediction optimization module, used to analyze the historical data of merchants by using a machine learning model and generate a behavior prediction result, and optimize the server resource allocation strategy according to the behavior prediction result; an intelligent erp configuration recommendation module, used to develop an intelligent erp configuration recommendation system based on a deep reinforcement learning algorithm, and the intelligent erp configuration recommendation system automatically identifies the business model of merchants and generates personalized erp management configuration suggestions by using the behavior prediction result; a unified order management module, used to integrate multi-channel payment interfaces and logistics APIs, and automatically select payment and logistics strategies according to the personalized erp management configuration suggestions, and build a unified order management process.

[0014] A computer device includes a memory and a processor, the memory stores a computer program, and it is characterized in that when the processor executes the computer program, the steps of the above-mentioned erp background management method based on a multi-user mall are implemented.

[0015] A computer-readable storage medium stores a computer program thereon, and it is characterized in that when the computer program is executed by a processor, the steps of the above-mentioned erp background management method based on a multi-user mall are implemented.

[0016] The beneficial effects of the present invention: The present invention provides an erp background management method for a multi-user mall based on a cloud architecture, which realizes intelligent and dynamic server resource allocation by building an elastic expansion mechanism and effectively responds to load fluctuations. This method uses a machine learning model to analyze the historical data of merchants, generates a behavior prediction result, and optimizes the resource allocation strategy, improving the foresight and adaptability of the system. At the same time, the intelligent erp configuration recommendation system developed based on the deep reinforcement learning algorithm can automatically identify the business model of merchants and generate personalized management configuration suggestions, greatly improving the intelligent level of the system. In addition, by integrating multi-channel payment interfaces and logistics APIs, a unified order management process is built, realizing the automatic selection of payment and logistics strategies and the efficient management of the entire life cycle of orders. This comprehensive solution not only significantly improves resource utilization efficiency and system performance, but also better meets the personalized needs of different merchants, provides strong technical support for the operation and management of multi-user malls, and effectively improves the user experience and operation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0018] Figure 1 This is the overall flowchart of the erp background management method based on a multi-user mall provided by an embodiment of the present invention.

[0019] Figure 2 This is the computer device diagram of the erp background management method based on a multi-user mall of the present invention. Specific Embodiments

[0020] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the specific embodiments of the present invention in detail with reference to the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0022] Embodiment 1, referring to Figure 1 , which is an embodiment of the present invention, provides an erp background management method based on a multi-user mall.

[0023] In view of the problems existing in the existing erp background management technology of multi-user malls, such as inflexible resource allocation, lack of intelligent configuration, and scattered order management processes, the present invention proposes an erp background management method and system based on a multi-user mall. This method aims to solve the above problems through technical means such as constructing an elastic expansion mechanism based on a cloud architecture, using a machine learning model for merchant behavior prediction, developing an intelligent erp configuration recommendation system, and integrating multi-channel payment and logistics APIs. The present invention belongs to the fields of e-commerce technology and enterprise resource planning systems, and provides new technical support for the efficient operation and intelligent management of multi-user malls.

[0024] S1: Construct an elastic expansion mechanism based on a cloud architecture. The elastic expansion mechanism dynamically adjusts server resources by monitoring the real-time load data of the erp background management system according to the server resource allocation strategy.

[0025] S1.1: Build a distributed load monitoring system based on a cloud architecture, and collect real-time load data such as CPU utilization, memory occupancy, network throughput, and disk I / O of the server cluster through the distributed load monitoring system.

[0026] Specifically, in the present invention, the distributed load monitoring system based on a cloud architecture is the foundation of the entire elastic expansion mechanism. It adopts a distributed architecture and deploys multiple monitoring nodes in the cloud environment to achieve real-time and efficient monitoring of large-scale server clusters. Each monitoring node is responsible for collecting real-time load data such as CPU utilization, memory occupancy, network throughput, and disk I / O of the servers within its jurisdiction. These real-time load data are transmitted to the central data processing unit through a secure network channel, where they are aggregated, cleaned, and preliminarily analyzed. The distributed load monitoring system uses lightweight data collection agents to minimize the impact on the performance of the monitored servers. At the same time, to address the challenges of large-scale data processing, the distributed load monitoring system integrates stream processing technology to be able to process massive amounts of monitoring data in real time. In addition, the system also has an adaptive sampling rate adjustment function, which can dynamically adjust the data collection frequency according to the load situation, optimizing the use of system resources while ensuring monitoring accuracy. To improve reliability, the distributed load monitoring system adopts a data multi-copy storage and automatic failure switching mechanism to ensure continuous provision of accurate monitoring data in the event of partial node failures. The distributed load monitoring system provided by the present invention provides a reliable data basis for subsequent resource evaluation and dynamic adjustment and is a key component for realizing intelligent resource management.

[0027] S1.2: Formulate a server resource allocation strategy. The server resource allocation strategy includes resource utilization thresholds and resource adjustment rules.

[0028] Specifically, the server resource allocation strategy is a method for dynamically adjusting cloud server resources, aiming to optimize resource utilization efficiency and system performance. In the erp background management system of the multi-user mall provided by the present invention, the server resource allocation strategy needs to cope with load changes at different times, such as a sharp increase in traffic during promotions or load fluctuations during daily operations. Therefore, the server resource allocation strategy plays an important role.

[0029] In this embodiment, the resource utilization threshold is dynamically set, and the resource utilization threshold includes a dynamic high threshold and a dynamic low threshold. Before setting the resource utilization threshold, it is necessary to determine a baseline threshold, which also includes a baseline high threshold and a baseline low threshold. The baseline threshold is usually determined based on the experience of the system administrator or historical data analysis. By setting different baseline thresholds for different types of resources (i.e., real-time load data: CPU, memory, network bandwidth, disk I / O), resource allocation can be more finely controlled. In one embodiment, the rules for setting the baseline threshold can be as follows: a) The CPU utilization is set relatively low to reserve sufficient processing power to handle sudden requests. b) The memory usage is slightly higher because memory resources are usually easier to expand quickly. c) The network bandwidth threshold is moderate to ensure good response speed. d) The disk I / O threshold is relatively low because I / O operations are usually the performance bottleneck.

[0030] Exemplarily, based on the setting rules, the baseline high thresholds for these resource types can be: CPU utilization threshold: 75%, memory usage threshold: 80%, network bandwidth usage threshold: 70%, and disk I / O usage threshold: 60%.

[0031] Specifically, the calculation formula for the dynamic high threshold is:

[0032] Dynamic high threshold = baseline high threshold + α * standard deviation + β * trend factor;

[0033] The calculation formula for the dynamic low threshold is:

[0034] Dynamic low threshold = baseline low threshold - α * standard deviation - β * trend factor;

[0035] It should be noted that the trend factor is an indicator reflecting the changing trend of resource utilization. In this embodiment, the resource utilization data of a certain type of real-time load data at the most recent N time points is used to obtain the slope through linear regression as the trend factor. α and β respectively represent the weights of the standard deviation and the trend factor. The standard deviation refers to the standard deviation of the resource utilization of a certain type of real-time load data.

[0036] Furthermore, the resource adjustment rules include a resource expansion rule, a resource contraction rule, an intelligent reservation rule, and an elastic adjustment rule.

[0037] Resource expansion rule: When the resource utilization of a certain type of real-time load data exceeds its dynamic high threshold for continuously T1 minutes, resource expansion is triggered, and the expansion amount is: the resource amount of the current real-time load data * (the resource utilization of the current real-time load data - dynamic high threshold) / dynamic high threshold, rounded up, but not less than the minimum expansion unit.

[0038] Resource contraction rule: When the resource utilization rate of a certain type of real-time load data is lower than its dynamic low threshold for T2 consecutive minutes, resource contraction is triggered, and the contraction amount is:

[0039] The resource amount of the current real-time load data * (dynamic low threshold - resource utilization rate of the current real-time load data) / dynamic low threshold, rounded up, but not less than the minimum contraction unit.

[0040] Intelligent reservation rule: The intelligent reservation rule is that when the resource demand within the preset time of the erp back-end management system (such as predicting the next N hours) will exceed the total available resources, the reserved resource amount is: predicted peak usage * (1 + safety factor) + volatility factor * historical volatility range; where the safety factor is a value greater than 0, used to increase a certain proportion of additional resources on the basis of prediction to cope with potential prediction errors; the volatility factor is a weight factor used to adjust the impact of historical volatility on the reserved resources; the historical volatility range refers to the change range of resource usage in the past period of time.

[0041] Elastic adjustment rule: If the number of resource adjustments within the last M hours is greater than the preset number, the new resource utilization rate threshold = original resource utilization rate threshold * (1 + X%), to reduce frequent adjustments.

[0042] S1.3: Establish a multi-dimensional resource scoring model, and the multi-dimensional resource scoring model calculates the server resource utilization rate score based on the real-time load data collected in S1.1.

[0043] Specifically, the multi-dimensional resource scoring model calculates the ratio of the current usage of each type of real-time load data to the maximum capacity, then multiplies it by the weight of each type of real-time load data, and finally adds up the weighted resource utilization rates of all real-time load data to obtain the score, which is used as the server resource utilization rate score. This server resource utilization rate score reflects the overall resource usage of the server. The formula is:

[0044]

[0045] Among them, S is the server resource utilization rate score; n is the number of resource types considered (such as CPU, memory, network, disk I / O, etc.); w i is the weight of the i-th type of resource, and R i is the current usage of the i-th type of resource; R i,max is the maximum capacity of the i-th type of resource.

[0046] It should be noted that in this embodiment, the application layer response time can also be regarded as the (n + 1)-th type of "resource", that is, as an additional item in the formula:

[0047]

[0048] Among them, w app is the weight of the application layer metrics; T current is the current response time; T target is the target response time.

[0049] By expanding the calculation of the server resource utilization score by treating the application layer response time as the (n + 1)-th "resource", a more comprehensive and intelligent resource allocation strategy can be implemented in the erp back-end management system of a multi-user mall. Because in the e-commerce environment, the response speed of the system directly affects the user experience and the transaction completion rate, so it is crucial to incorporate it into the resource allocation decision-making. Through this expansion, the erp back-end management system not only considers the traditional hardware resource utilization (such as CPU, memory, storage, and network), but also takes into account the performance metrics at the software level. This means that even when the hardware resources do not reach a high load, if the application layer response time slows down, the erp back-end management system can detect it in time and make adjustments. For example, during a promotional event, there may be a large number of concurrent requests resulting in an increase in the response time, but the CPU and memory utilization do not reach the threshold. In this case, the traditional resource allocation strategy may not be able to respond in time, while the extended multi-dimensional resource scoring model can trigger resource expansion faster to ensure the user experience. In addition, the addition of this extra item also provides the erp back-end management system with more refined tuning capabilities. Administrators can balance the relationship between hardware resource utilization and software performance by adjusting the weight of the application layer metrics, so as to better meet the needs of specific business scenarios. For example, for the module that processes high-value transactions, a higher weight can be given to the response time to ensure that these critical operations can obtain more resource guarantees.

[0050] This method of comprehensively considering hardware and software performance makes the resource allocation strategy closer to the actual business needs, can better cope with the dynamic load characteristics of the e-commerce platform, and improves the overall efficiency of the system and user satisfaction. At the same time, it also provides an extensible framework for introducing more application layer metrics (such as transaction processing speed, query response time, etc.) in the future, enabling the resource allocation strategy to evolve and optimize continuously to adapt to the changing business needs.

[0051] S1.4: Compare the server resource utilization score calculated in S1.3 with the resource utilization threshold in S1.2. When the server resource utilization score exceeds the resource utilization threshold, trigger resource adjustment.

[0052] S1.5: After triggering resource adjustment, according to the resource adjustment rules in S1.2, dynamically increase or decrease the number of server instances, adjust the calculation resource allocation, and achieve elastic expansion of server resources.

[0053] Preferably, S1 realizes an intelligent and dynamic server resource allocation strategy by constructing an elastic expansion mechanism based on a cloud architecture. First, a distributed load monitoring system is used to collect multi-dimensional resource usage data in real time, providing a comprehensive and accurate basis for decision-making. Second, by introducing a dynamic threshold system and a multi-dimensional resource scoring model, the erp back-end management system can flexibly adjust the resource allocation strategy according to the real-time load situation and historical trends, effectively coping with load fluctuations. In particular, incorporating the application layer response time into the scoring model makes the resource allocation more in line with the actual business needs and can respond in a timely manner to changes in the user experience. In addition, intelligent reservation rules and elastic adjustment rules further enhance the system's forward-looking and stability, effectively preventing problems such as resource shortages or over-allocation. This method that comprehensively considers the hardware resource utilization rate and software performance indicators not only improves the resource utilization efficiency but also significantly improves the system's adaptability in the face of complex and changing load situations, especially suitable for application scenarios with high concurrency and large load fluctuations such as e-commerce platforms. Generally speaking, this solution can realize the economical and efficient utilization of resources while ensuring system performance and user experience, providing strong technical support for the erp back-end management system of multi-user malls.

[0054] S2: Use a machine learning model to analyze the historical data of merchants and generate behavior prediction results, and optimize the server resource allocation strategy according to the behavior prediction results.

[0055] S2.1: Collect the historical data of merchants. The historical data of merchants consists of multi-dimensional information, specifically including order volume, transaction amount, access traffic, inventory changes, and seasonal fluctuations, etc.

[0056] Collect the historical data of merchants, the order volume O(t), transaction amount V(t), access traffic F(t), inventory changes I(t), and the seasonal fluctuation index S(t), where t represents the time point.

[0057] S2.2: Preprocess the historical data of merchants collected in S2.1, including data cleaning, outlier handling, and feature standardization.

[0058] S2.3: Construct a time series prediction model based on the long short-term memory network algorithm LSTM and the attention mechanism, and train the preprocessed historical data of merchants in S2.2 through the time series prediction model.

[0059] Specifically, the time series prediction model is expressed as:

[0060] 1. Calculation of the LSTM unit:

[0061] f t =σ(W f ·[h t-1 ,x t +bf )

[0062] i t = σ(W i · [h t-1 , x t + b i )

[0063]

[0064] o t = σ(W o · [h t-1 , x t + b o )

[0065] h t = o t * tanh(C t )

[0066] Among them, f t is the forget gate; i t is the input gate; o t is the output gate; C t is the cell state, storing long-term memory; is the candidate cell state; h t is the hidden state, serving as the output of the current time step; x t is the input of the current time step; W f is the weight matrix of the forget gate; W i is the weight matrix of the input gate; W C is the weight matrix of the cell state; W o is the weight matrix of the output gate; b f , b i , b C , b o are the corresponding bias terms respectively; σ is the activation function; tanh is the hyperbolic tangent activation function.

[0067] 2. Attention mechanism:

[0068]

[0069] Among them, e ij is the attention score of the i-th decoding time step for the j-th encoded hidden state; α ij is the normalized attention weight; c i is the context vector, representing the weighted encoder hidden state; V a , W a and U a are the learnable parameters of the attention mechanism; s i-1is the decoder state at the previous time step; h j is the j-th encoder hidden state; The T in represents transpose; The T in represents the length of the input sequence, and k is an index ranging from 1 to T.

[0070] 3. Introduce the time-aware attention mechanism:

[0071]

[0072] where τ ij is the time decay factor; θ is the adjustable time scale parameter; is the attention weight after adding time awareness.

[0073] 4. Final prediction:

[0074]

[0075] where, is the prediction output; W y and b y are the weights and biases of the output layer.

[0076] S2.4: Use the time series prediction model in S2.3 to generate the merchant behavior prediction results. The merchant behavior prediction results include the expected order volume, transaction amount, and access traffic within a certain future time period.

[0077] Specifically, use the trained model for prediction:

[0078]

[0079] where, is the predicted future order volume; is the predicted future transaction amount; is the predicted future access traffic; f LSTM represents the trained LSTM model; represents the historical data of merchant i; Δt is the prediction time span.

[0080] S2.5: According to the merchant behavior prediction results in S2.4, adjust the server resource allocation strategy in S1.2, including updating the resource utilization threshold, optimizing the resource adjustment rule, and formulating the resource reservation plan.

[0081] Specifically, update the resource utilization threshold, specifically by dynamically adjusting the dynamic high threshold and dynamic low threshold in S1.2 according to the prediction results:

[0082]

[0083] where, are the updated dynamic high threshold and low threshold; are the original dynamic high threshold and low threshold; δ high , δ low are the adjustment coefficients of the high threshold and low threshold; is the predicted future access traffic; F i (t) is the current access traffic.

[0084] Preferably, when updating the threshold, the change in access traffic is mainly considered because it is usually the most direct factor affecting the server load.

[0085] Furthermore, adjust the resource expansion and contraction rules in S1.2 according to the prediction results.

[0086] Resource expansion rule:

[0087]

[0088] Resource contraction rule:

[0089]

[0090] where E is the expansion amount; P is the contraction amount; R current is the current resource amount; E min is the minimum expansion unit; P min is the minimum contraction unit.

[0091] Preferably, in the resource adjustment rule, the access traffic is the main reference index, but in actual applications, the impacts of the order volume and transaction amount can be considered according to specific situations.

[0092] Formulate a resource reservation plan:

[0093]

[0094] where R reserve is the reserved resource amount; is the safety factor, corresponding to the safety factor in S1.2; ω is the volatility factor, corresponding to the volatility factor in S1.2; V hist is the historical volatility range, corresponding to the historical volatility range in S1.2.

[0095] Preferably, in the resource reservation plan, the predicted access traffic and order volume are comprehensively considered, and the maximum value is taken as the benchmark, so as to better cope with possible peak loads.

[0096] S3: Develop an intelligent erp configuration recommendation system based on the deep reinforcement learning algorithm. The intelligent erp configuration recommendation system automatically identifies the business models of merchants and generates personalized erp management configuration suggestions using the behavior prediction results.

[0097] S3.1: Construct a merchant business model feature matrix that includes the merchant behavior prediction results, historical erp usage data, and product category information in S2.4.

[0098] Specifically, for the merchant business model feature matrix X i (t) of merchant i at time t, it is expressed as:

[0099] X i (t) = [P i (t), H i (t), C i ;

[0100]

[0101] Among them, P i (t) is the merchant behavior prediction result vector from S2.4; H i (t) is the historical erp usage data vector; C i is the product category information vector.

[0102] S3.2: Construct a multi-task deep neural network model. This multi-task deep neural network model includes a shared layer and multiple task-specific layers. The shared layer is used to extract general features of the merchant business, and the multiple task-specific layers are respectively used for merchant business model recognition and epr configuration parameter prediction.

[0103] Multi-task learning objective function:

[0104] L total = αL business + βL config + γL shared + λΩ(W);

[0105] Among them, L business is the loss function for merchant business model recognition; L config is the loss function for epr configuration parameter prediction; L shared is the loss function for the shared layer; Ω(W) is the regularization term; α, β, γ, λ are weight coefficients.

[0106] S3.3: Adopt the transfer learning method to initialize the parameters of the multi-task deep neural network model in S3.2 using the pre-trained e-commerce domain knowledge graph.

[0107] Specifically, the way the knowledge graph is embedded into the model:

[0108] W init = W pretrained + ∈;

[0109] Among them, Winit is the initialized model weights; W pretrained are the pre-trained weights learned from the knowledge graph; ∈ is a small random perturbation used to break symmetry.

[0110] S3.4: Train the initialized multi-task deep neural network model in S3.3 based on the merchant business model feature matrix in S3.1 to obtain a trained multi-task deep neural network model for merchant business model recognition and erp configuration parameter prediction.

[0111] Specifically, the model training process is expressed as:

[0112]

[0113] where are the optimized model parameters; X is the input feature; Y is the true label; are the model parameters.

[0114] S3.5: Use the merchant business model feature matrix in S3.1 as the input and pass it into the trained multi-task deep neural network model in S3.4. Extract the general features of the merchant business through the shared layer, and then use the task-specific layer for merchant business model recognition to identify the merchant business model, and use the task-specific layer for erp configuration parameter prediction to generate corresponding personalized erp management configuration suggestions.

[0115] Specifically, merchant business model recognition:

[0116]

[0117] where p(y i |X i ) is the probability distribution of predicting that merchant i belongs to each business model given the input feature matrix X i of merchant i; y i is the business model category of merchant i; W b is the weight matrix of the business model recognition task; h is the output feature vector of the shared layer; b b is the bias vector of the business model recognition task; softmax is the activation function used to convert the output into a probability distribution;

[0118] erp configuration parameter prediction:

[0119]

[0120] where h is the output feature vector of the shared layer; W b , b b are the weights and biases of the business model recognition task; W c , bc are the weights and biases for the configuration parameter prediction task; f c is the activation function (such as ReLU). is the predicted erp configuration parameter vector;

[0121] It should be noted that to further improve the accuracy of erp configuration parameter prediction, an attention mechanism can be introduced.

[0122] It should be noted that the attention mechanism is used in both S2.3 and S3.5, but there are some important differences in their application scenarios and specific implementations:

[0123] 1. Differences in application scenarios:

[0124] S2.3 (time series prediction model): Used to process time series data, aiming to predict future order volumes, transaction amounts, and access traffic. The attention mechanism is mainly used to capture the importance of different time points.

[0125] S3.5 (erp configuration parameter prediction): Used to process multi-dimensional merchant feature data, aiming to generate personalized erp configuration suggestions. The attention mechanism is mainly used to capture the importance of different erp configuration parameters.

[0126] 2. Differences in specific implementations: The attention mechanism of S2.3 is in the time dimension, while the attention mechanism of S3.5 is between different erp configuration parameters.

[0127] 3. Differences in input and output: S2.3: The input is time series data, and the output is the predicted value at future time points; S3.5: The input is a feature matrix containing prediction results, historical data, and product information, and the output is the predicted value of erp configuration parameters.

[0128] 4. Differences in computational complexity: The attention mechanism of S2.3 has a higher computational complexity because it needs to calculate attention weights at each time step. The attention mechanism of S3.5 is relatively simple and only needs to calculate attention weights once for a fixed number of erp configuration parameters.

[0129] 5. Differences in objectives: The attention mechanism of S2.3 aims to improve the accuracy of time series prediction. The attention mechanism of S3.5 aims to improve the personalization and accuracy of erp configuration suggestions.

[0130] Generally speaking, although both use the attention mechanism technology, due to differences in application scenarios, input and output, calculation methods, and ultimate goals, there are obvious differences in their specific implementations and functions. The attention mechanism in S2.3 focuses more on capturing long-term dependencies in time series, while the attention mechanism in S3.5 focuses more on finding the most relevant parts among multiple configuration parameters. The combination of these two attention mechanisms enables the entire system to accurately predict future trends and provide precise personalized erp configuration suggestions.

[0131] S4: Integrate multi-channel payment interfaces and logistics APIs, and automatically select payment and logistics strategies according to the personalized erp management configuration suggestions to build a unified order management process.

[0132] S4.1: Integrate payment interfaces and logistics APIs.

[0133] Integrate APIs of multiple third-party payment platforms and logistics service providers.

[0134] Establish a payment and logistics strategy database, and store the payment and logistics-related parameters in the personalized erp management configuration suggestions generated in S3.5 into this database.

[0135] S4.2: Build an order processing module and an order status tracking system.

[0136] Build an order processing module, which automatically selects the most suitable payment method and logistics plan for each order according to the payment and logistics strategy database in S4.1.

[0137] Build an order status tracking system, which obtains the payment status and logistics information of orders in real time through the multi-channel payment interfaces and logistics APIs integrated in S4.1.

[0138] S4.3: Build a unified order management interface.

[0139] Build a unified order management interface, which integrates the order processing module and the order status tracking system in S4.2.

[0140] Implement the management functions of the entire order life cycle in the order management interface, including order creation, payment, shipping, logistics tracking, and completion, etc.

[0141] In summary, the present invention realizes intelligent and dynamic server resource allocation by constructing an elastic expansion mechanism, effectively coping with load fluctuations. This method uses a machine learning model to analyze the historical data of merchants, generates behavior prediction results, optimizes the resource allocation strategy, and improves the forward-looking and adaptability of the system. At the same time, the intelligent erp configuration recommendation system developed based on the deep reinforcement learning algorithm can automatically identify the business models of merchants and generate personalized management configuration suggestions, greatly enhancing the intelligence level of the system. In addition, by integrating multi-channel payment interfaces and logistics APIs, a unified order management process is constructed, realizing the automatic selection of payment and logistics strategies and the efficient management of the entire order life cycle. This comprehensive solution not only significantly improves resource utilization efficiency and system performance, but also better meets the personalized needs of different merchants, provides strong technical support for the operation and management of multi-user shopping malls, and effectively improves user experience and operation efficiency.

[0142] Embodiment 2, which is an embodiment of the present invention, provides an erp background management system based on a multi-user shopping mall, including: an elastic resource management module for constructing an elastic expansion mechanism based on a cloud architecture, and the elastic expansion mechanism dynamically adjusts server resources by monitoring the real-time load data of the erp background management system according to the server resource allocation strategy; a prediction optimization module for analyzing the historical data of merchants using a machine learning model and generating behavior prediction results, and optimizing the server resource allocation strategy according to the behavior prediction results; an intelligent erp configuration recommendation module for developing an intelligent erp configuration recommendation system based on the deep reinforcement learning algorithm, and the intelligent erp configuration recommendation system automatically identifies the business models of merchants and generates personalized erp management configuration suggestions using the behavior prediction results; a unified order management module for integrating multi-channel payment interfaces and logistics APIs, and automatically selecting payment and logistics strategies according to the personalized erp management configuration suggestions to construct a unified order management process.

[0143] Embodiment 3, refer to Figure 2, which is an embodiment of the present invention and is different from the previous embodiment in that: if the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or part of this 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 for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0144] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a predefined sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction execution system, apparatus, or device.

[0145] More specific examples (non-exhaustive list) of computer-readable media include the following: electrical connection parts with one or more wirings (electronic devices), portable computer disk cartridges (magnetic devices), random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memories), fiber optic devices, and portable compact disc read-only memories (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or otherwise processing it as necessary, and then storing it in a computer memory.

[0146] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logical functions on data signals, application specific integrated circuits with appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0147] Example 4 is an embodiment of the present invention, which provides an erp background management method based on a multi-user mall. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0148] First of all, this embodiment constructs a simulated multi-user mall environment, including 100 merchants of different scales and types, and about 500,000 users. The experiment is divided into two stages, each stage lasting 30 days. The first stage uses a traditional ERP system for management, and the second stage switches to this solution for management. During the whole experiment process, various scenarios such as daily operations, promotional activities, and sudden traffic are simulated to comprehensively test the performance and adaptability of the system.

[0149] During the experiment, various index data are continuously monitored and recorded. For system performance, a stress test tool is used to simulate different levels of concurrent access to test the response time and processing ability of the system. For resource management efficiency, by deploying monitoring agents, the resource usage of the server cluster is collected in real time. In terms of the evaluation of the degree of intelligence, a certain proportion of merchant samples are randomly selected to manually review the personalized configuration suggestions, resource demand predictions, and abnormal situation detection results generated by the system, and calculate their accuracy rates. Through these comprehensive tests and data collection, the above experimental results are obtained, which strongly prove the superiority of this solution in all aspects.

[0150] Table 1 System Performance Evaluation

[0151]

[0152]

[0153] As shown in Table 1, the system performance evaluation results show that the present invention has achieved significant improvements in all key indicators. The average response time has been reduced from 320 ms to 95 ms, a 70.3% improvement, which greatly enhances the user experience. The number of concurrent users has increased from 5000 to 15000, a 200% increase, indicating a substantial improvement in the system's load capacity. The number of transactions per second (TPS) has increased from 3500 to 9800, an 180% increase, which means the system can process more business requests within the same time. These data fully demonstrate the excellent performance of the present invention in handling high concurrency and large-scale transactions, providing strong technical support for the efficient operation of multi-user shopping malls.

[0154] Table 2 Resource Management Efficiency Evaluation

[0155] Index Traditional ERP system The present invention Improvement rate CPU utilization rate (%) 40 75 87.5％ Memory usage efficiency (%) 55 82 49.1％ Storage space utilization rate (%) 60 85 41.7％

[0156] As shown in Table 2, the resource management efficiency evaluation results indicate that the present invention has made significant progress in server resource utilization. The CPU utilization rate has increased from 40% to 75%, an 87.5% increase; the memory usage efficiency has increased from 55% to 82%, a 49.1% improvement; the storage space utilization rate has increased from 60% to 85%, a 41.7% increase. These data reflect that the present invention has greatly improved the utilization efficiency of system resources through intelligent resource scheduling and elastic expansion mechanisms. Efficient resource management not only reduces operating costs but also provides greater flexibility for the business growth of the shopping mall, enabling the system to better handle business peaks and sudden traffic.

[0157] Table 3: Intelligence Degree Evaluation

[0158] Index Traditional ERP system The present invention Improvement rate Personalized configuration accuracy rate (%) 65 94 44.6％ Resource prediction accuracy rate (%) 70 92 31.4％ Abnormal detection accuracy rate (%) 75 96 28.0％

[0159] As shown in Table 3, the intelligence degree evaluation results highlight the advantages of the present invention in intelligent decision-making and prediction. The accuracy rate of personalized configuration has increased from 65% to 94%, a 44.6% increase, which means the system can provide more accurate customized ERP configuration suggestions for merchants. The accuracy rate of resource prediction has increased from 70% to 92%, a 31.4% improvement, showing excellent performance of the system in resource demand prediction. The accuracy rate of anomaly detection has increased from 75% to 96%, a 28% increase, which greatly improves the system's ability to identify and handle abnormal situations. These data fully demonstrate the effectiveness of the machine learning and deep reinforcement learning algorithms adopted in the present invention, providing strong technical support for the intelligent management of multi-user shopping malls.

[0160] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An erp background management method based on a multi-user mall, characterized in that, Including: Construct an elastic scaling mechanism based on a cloud architecture. The elastic scaling mechanism dynamically adjusts server resources by monitoring real-time load data of the erp back-end management system according to a server resource allocation policy; Adopt a machine learning model to analyze historical merchant data and generate behavior prediction results, and optimize the server resource allocation policy according to the behavior prediction results; Develop an intelligent erp configuration recommendation system based on a deep reinforcement learning algorithm. The intelligent erp configuration recommendation system uses the behavior prediction results to automatically identify the merchant business model and generate personalized erp management configuration suggestions, including the following steps: Construct a merchant business model feature matrix including the merchant behavior prediction results, historical erp usage data, and product category information; Construct a multi-task deep neural network model. The multi-task deep neural network model includes a shared layer and multiple task-specific layers. The shared layer is used to extract general features of the merchant business, and the multiple task-specific layers are respectively used for merchant business model recognition and epr configuration parameter prediction; Adopt a transfer learning method to initialize the parameters of the multi-task deep neural network model using a pre-trained e-commerce domain knowledge graph; Train the initialized multi-task deep neural network model based on the merchant business model feature matrix to obtain a trained multi-task deep neural network model for merchant business model recognition and erp configuration parameter prediction; Use the merchant business model feature matrix as input, pass it into the trained multi-task deep neural network model, extract general features of the merchant business through the shared layer, and then use the task-specific layer for merchant business model recognition to perform merchant business model recognition, and use the task-specific layer for erp configuration parameter prediction to generate corresponding personalized erp management configuration suggestions; Integrate multi-channel payment interfaces and logistics APIs, and automatically select payment and logistics strategies according to the personalized erp management configuration suggestions to construct a unified order management process.

2. The erp background management method based on a multi-user mall according to claim 1, wherein: Construct an elastic scaling mechanism based on a cloud architecture. The elastic scaling mechanism dynamically adjusts server resources by monitoring real-time load data of the erp back-end management system according to a server resource allocation policy, including the following steps: Construct a distributed load monitoring system based on a cloud architecture, and collect various real-time load data of the server cluster through the distributed load monitoring system; Formulate a server resource allocation policy; the server resource allocation policy includes a resource utilization threshold and a resource adjustment rule; Establish a multi-dimensional resource scoring model; the multi-dimensional resource scoring model calculates the server resource utilization score according to the real-time load data; Compare the server resource utilization score with the resource utilization threshold. When the server resource utilization score exceeds the resource utilization threshold, trigger resource adjustment; After triggering resource adjustment, dynamically increase or decrease the number of server instances according to the resource adjustment rule, adjust the calculation resource allocation, and complete the elastic scaling mechanism of the server resources.

3. The erp background management method based on a multi-user mall according to claim 2, characterized in that: Each type of the real-time load data corresponds to a type of resource; the resource utilization threshold includes a dynamic high threshold and a dynamic low threshold; The resource adjustment rules include resource expansion rules, resource contraction rules, intelligent reservation rules, and elastic adjustment rules; Among them, the resource expansion rule is that when the resource utilization rate of a certain type of the real-time load data exceeds its dynamic high threshold for continuously T1 minutes, resource expansion is triggered, and the expansion amount is: The resource amount of the current real-time load data * (the resource utilization rate of the current real-time load data - the dynamic high threshold) / the dynamic high threshold, rounded up, but not less than the minimum expansion unit; The resource contraction rule is that when the resource utilization rate of a certain type of the real-time load data is lower than its dynamic low threshold for continuously T2 minutes, resource contraction is triggered, and the contraction amount is: The resource amount of the current real-time load data * (the dynamic low threshold - the resource utilization rate of the current real-time load data) / the dynamic low threshold, rounded up, but not less than the minimum contraction unit; The intelligent reservation rule is that when the erp background management system predicts that the resource demand within a preset time will exceed the current capacity, that is, the total available resource amount at present, the reserved resource amount is: the predicted peak usage * (1 + safety factor) + volatility coefficient * historical fluctuation range; The elastic adjustment rule is that if the number of resource adjustments within the recent M hours is greater than the preset number, then the new resource utilization rate threshold = the original resource utilization rate threshold * (1 + X%); The multi-dimensional resource scoring model is to calculate the ratio of the current usage amount of each real-time load data to the maximum capacity, multiply it by the weight of each real-time load data, and add up the weighted resource utilization rates of all real-time load data to obtain the score, which is used as the server resource utilization rate score.

4. The erp background management method based on a multi-user mall according to claim 3, characterized in that: Using a machine learning model to analyze the merchant's historical data and generate a behavior prediction result, and optimizing the server resource allocation strategy according to the behavior prediction result, including the following steps: Collect and preprocess the merchant's historical data; the merchant's historical data includes the order volume, transaction amount, access traffic, inventory changes, and seasonal fluctuations; Construct a time series prediction model based on the long short-term memory network algorithm LSTM and the attention mechanism, and train the preprocessed merchant's historical data through the time series prediction model; Generate a merchant behavior prediction result using the time series prediction model; the merchant behavior prediction result includes the expected order volume, transaction amount, and access traffic; Adjust the server resource allocation strategy according to the merchant behavior prediction result; Among them, adjusting the server resource allocation strategy includes updating the resource utilization rate threshold, optimizing the resource adjustment rules, and formulating a resource reservation plan.

5. The erp background management method based on a multi-user mall according to claim 4, characterized in that: Integrate multiple-channel payment interfaces and logistics APIs, and automatically select payment and logistics strategies according to personalized erp management configuration suggestions to construct a unified order management process, including the following steps: Integrate the APIs of multiple third-party payment platforms and logistics service providers; Establish a payment and logistics strategy database, and store the payment and logistics parameters in the personalized erp management configuration suggestions into the logistics strategy database; Construct an order processing module; Construct an order status tracking system, and the order status tracking system obtains the payment status and logistics information of the order in real time; Construct a unified order management interface.

6. The erp background management method based on a multi-user mall according to claim 5, characterized in that: The real-time load data includes CPU, memory, network bandwidth, and disk I / O.

7. A system adopting the erp background management method based on a multi-user mall as described in any one of claims 1 to 6, characterized in that, Including: An elastic resource management module, which is used to build an elastic expansion mechanism based on the cloud architecture. The elastic expansion mechanism dynamically adjusts server resources by monitoring the system load data in real time according to the server resource allocation policy. A prediction optimization module, which is used to analyze the historical data of merchants by using a machine learning model and generate a behavior prediction result, and optimize the server resource allocation policy according to the behavior prediction result. An intelligent erp configuration recommendation module, which is used to develop an intelligent erp configuration recommendation system based on a deep reinforcement learning algorithm. The intelligent erp configuration recommendation system automatically identifies the merchant's business model and generates personalized erp management configuration suggestions by using the behavior prediction result. A unified order management module, which is used to integrate multi-channel payment interfaces and logistics APIs, and automatically select payment and logistics strategies according to the personalized erp management configuration suggestions to build a unified order management process.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the erp background management method based on a multi-user mall according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the erp background management method based on a multi-user mall according to any one of claims 1 to 6.

Images