Intelligent operation system of cross-border e-commerce platform

Abstract

The invention belongs to the technical field of cross-border e-commerce, and provides a cross-border e-commerce platform intelligent operation system, which comprises a data acquisition module used for acquiring data from multi-source channels and summarizing and integrating the data; the data analysis module is used for dividing user groups, analyzing association among commodities and carrying out differentiated marketing and combined sales; the intelligent recommendation module is used for calculating similarity according to user historical behaviors, recommending commodities purchased by similar users to the users, and recommending matched commodities in combination with user preferences; the supply chain management module is used for predicting commodity demands and planning an optimal logistics path based on multiple factors; the marketing automation module is used for automatically triggering personalized mail sending according to the behavior and preference information of the user, setting an automatic reply rule and automatically replying common questions of the user; according to the invention, through group division of the users, personalized commodity recommendation and service are provided for different user groups, so that the user satisfaction and the purchase conversion rate are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of cross-border e-commerce, and specifically relates to an intelligent operation system for a cross-border e-commerce platform. Background Art

[0002] With the rapid development of global e-commerce, cross-border e-commerce platforms have played an important role in promoting international trade and increasing the diversity of consumer choices.

[0003] However, many existing platforms lack effective strategies for user group segmentation and personalized recommendations. Traditional recommendation systems often rely on simple content-based recommendations or historical purchase records, but fail to fully consider the similarities and group characteristics between users. This single recommendation method leads to poor user experience and fails to meet the personalized needs of different users, thereby reducing user satisfaction and purchase conversion rates. In addition, the lack of effective association rule mining makes the platform's ability in product combination sales and differentiated marketing insufficient, and it is unable to fully explore potential sales opportunities. Most traditional recommendation systems are based on simple user behavior data, such as browsing records, purchase records, etc., and make extensive recommendations, which makes it difficult to accurately capture users' real needs and preferences. This results in the recommended products received by users often not matching their own interests, reducing the user's shopping experience and satisfaction.

[0004] At the same time, inventory management and logistics distribution efficiency also lag behind in existing technologies. Many platforms rely on static historical sales data for inventory forecasting and fail to fully utilize time series analysis algorithms for dynamic forecasting. This approach results in inflexible inventory management, which is prone to inventory backlogs or out-of-stock situations, affecting users' purchasing decisions and the platform's operating efficiency. In addition, logistics route planning mostly lacks comprehensive consideration of multiple factors and fails to effectively optimize delivery routes, resulting in high transportation costs and low delivery efficiency, which further affects customer satisfaction.

[0005] To this end, those skilled in the art have proposed an intelligent operation system for a cross-border e-commerce platform, which aims to improve the overall operational efficiency and user experience of the platform, thereby gaining an advantage in the highly competitive market. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a cross-border e-commerce platform intelligent operation system to solve the problems raised in the background technology.

[0007] A cross-border e-commerce platform intelligent operation system, comprising:

[0008] The data collection module is used to obtain transaction data, user behavior data, and package transportation status data from cross-border e-commerce platforms, social media, and third-party logistics service providers, and to aggregate and integrate multi-source data in the data storage center and pre-process the data;

[0009] The data analysis module is used to divide user groups through clustering algorithms, and to mine and analyze product associations using association rules, so as to conduct differentiated marketing and combined sales;

[0010] The intelligent recommendation module is used to calculate similarity based on user historical behaviors, recommend products that similar users have purchased, extract product attributes and description keywords in product description text, and recommend matching products based on user preferences;

[0011] Supply chain management module, which is used to predict commodity demand through time series analysis algorithms, perform inventory forecasts, and plan the optimal logistics path based on multiple factors;

[0012] The marketing automation module is used to automatically trigger the sending of personalized emails based on user behavior on the platform and user preference information, publish product content through social media on a scheduled basis, and set automatic reply rules to automatically reply to users' common questions.

[0013] Preferably, the data analysis module numerically processes the user's purchase frequency, consumption amount, and purchase category preference characteristic data to form a characteristic vector, and constructs a data set X = {x1, x2, ..., x n}, where x i is the feature vector of the i-th user, n is the total number of users, and k initial cluster centers C = {c1, c2, ..., c k}, k is the number of user groups that are expected to be divided; it is expressed using the Euclidean distance formula:

[0014]

[0015] Among them, m is the dimension of the feature vector. By calculating each user feature vector x i With each cluster center c j The user is assigned to the cluster to which the nearest cluster center belongs based on the distance between them. For each cluster, its cluster center is recalculated, and the new cluster center is the mean of the feature vectors of all users in the cluster, that is:

[0016]

[0017] Among them, n j is the number of users in the jth cluster, S jis the set of users in the jth cluster; repeat the steps of calculating distance, assigning clusters and updating cluster centers until the cluster center no longer changes or the change is less than the set threshold, and the user group division is completed;

[0018] The Apriori algorithm is used to analyze the association between commodities. Frequent item sets are generated from transaction data, and the support (A) is calculated as:

[0019]

[0020] Among them, |D(A)| represents the number of transactions containing item set A, and |D| represents the total number of transactions. The association rule A→B is generated from the frequent item set, which is expressed as:

[0021]

[0022] The confidence level is calculated to measure the strength of the rule, indicating the probability of purchasing B after purchasing A, and to conduct differentiated marketing and combination sales.

[0023] Preferably, the intelligent recommendation module is used to calculate similarity based on user historical behaviors, and organize the user's historical purchase and browsing records into a user-product matrix. The rows of the matrix represent users, the columns represent products, and the matrix elements are the user's behavior data on the products. When the user set is U = {u1,u2,...,u m}, the commodity set is I = {i1,i2,...,i n}, then the element r of the user-item matrix R ij Represents user u i Behavior data for product j, for user u a and user u b The cosine similarity is calculated as follows:

[0024]

[0025] Among them, r ai and r bi They are user u a and user u b For the behavior data of product i, the closer the cosine similarity value is to 1, the more similar the preferences of the two users are;

[0026] After obtaining the similarity between all users, for each target user, according to the similarity, select k users with high similarity as its similar user group, and its similar user set is S ux; For the target user, filter out the products that the target user has not purchased from the purchase records of similar user groups, and perform weighted calculation based on the behavior data of similar users on the products to obtain the recommendation score of each product, which is expressed as:

[0027]

[0028] Among them, score(j) is the recommendation score of product j, sim(u x ,u) is the target user u x Similarity with similar user u, r uj Based on the behavior data of similar user u on product j, we recommend products that similar users have purchased.

[0029] Preferably, the intelligent recommendation module also recommends matching products according to user preferences, and converts the user's historical purchase and browsing records into a user preference vector by extracting product attributes and descriptive keywords in the product description text. The user preference keyword set is P = {p1, p2, ..., p s}, for each keyword p j , giving weights based on the user's behavior data on products containing the keyword Form the user preference vector V p ;

[0030] For each product, the product feature vector V is formed based on the extracted keyword set. g , use cosine similarity to calculate the similarity between the user preference vector and the product feature vector, expressed as:

[0031]

[0032] in, is the keyword p in the product feature vector corresponding to the user preference j The corresponding TF-IDF value, the higher the similarity, the more the product matches the user's preference. The products are sorted from high to low according to the similarity, and the products with high similarity are recommended to the user.

[0033] Preferably, the supply chain management module collects historical sales data and arranges them in chronological order to form a time series {y t}, t=1,2,...,T, where y t represents the sales volume of the product at time t, where T is the total duration of the historical data;

[0034] Use historical sales data to train the ARIMA model and estimate the model parameters by minimizing the error between the predicted value and the true value; Based on the trained ARIMA model, combine future seasonal information and promotion activity plans to predict the demand for goods in the next h moments During the forecasting process, the model parameters, known historical data, and future input features are substituted into the model formula for calculation to predict future commodity demand and complete the inventory forecast.

[0035] Preferably, the supply chain management module also plans the optimal logistics path based on multiple factors, by abstracting the logistics distribution network into a directed graph G = (V, E), where V is a node set including warehouses, distribution centers and customer addresses, and E is an edge set representing the connection between nodes, and each edge (i, j) ∈ E corresponds to a transportation distance d ij , transportation cost ij , delivery time t ij ; When there are m paths, the kth path P k By node sequence where v k1 Represents the warehouse as the starting node, Indicates the client address as the terminating node;

[0036] Based on multiple factors, the objective function is expressed as:

[0037] Z k =ω1D k +ω2C k +ω3T k

[0038] Among them, ω1, ω2, ω3 are corresponding weight coefficients, and ω1+ω2+ω3=1; the optimal logistics path is obtained by solving.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. The present invention divides users into groups and analyzes product associations through association rule mining, which helps the platform implement differentiated marketing and combined sales strategies, and provides personalized product recommendations and services for different user groups, thereby improving user satisfaction and purchase conversion rate.

[0041] 2. The present invention calculates similarity based on user historical behaviors, recommends products purchased by similar users to users, and recommends matching products based on user preferences. This personalized recommendation method not only improves the user's shopping experience, but also deepens the connection between the user and the platform, and helps to promote brand loyalty.

[0042] 3. The present invention predicts commodity demand and inventory forecasts through time series analysis algorithms, which helps the platform to arrange inventory reasonably and avoid inventory backlogs or out-of-stock phenomena. At the same time, the optimal logistics path is planned based on multiple factors, which improves logistics distribution efficiency, reduces transportation costs, and further improves customer satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a block diagram of the cross-border e-commerce platform intelligent operation system of the present invention. DETAILED DESCRIPTION

[0044] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0045] As attached Figure 1 As shown:

[0046] Embodiment 1: The present invention provides a cross-border e-commerce platform intelligent operation system, including: a data collection module, used to obtain transaction data, user behavior data and package transportation status data from the cross-border e-commerce platform, social media and third-party logistics service providers, and summarize and integrate multi-source data in a data storage center; at the same time, pre-process the data, including identifying and deleting duplicate transaction records, correcting erroneous user information formats, filling in missing logistics track data, etc., to ensure the accuracy, consistency and integrity of the data and improve data availability.

[0047] The data analysis module is used to divide user groups through clustering algorithms, and to mine and analyze product associations using association rules, so as to conduct differentiated marketing and combined sales;

[0048] The intelligent recommendation module is used to calculate similarity based on user historical behaviors, recommend products that similar users have purchased, extract product attributes and description keywords in product description text, and recommend matching products based on user preferences;

[0049] Supply chain management module, which is used to predict commodity demand through time series analysis algorithms, perform inventory forecasts, and plan the optimal logistics path based on multiple factors;

[0050] The marketing automation module is used to automatically trigger the sending of personalized emails based on user behavior on the platform and user preference information, publish product content through social media on a scheduled basis, and set automatic reply rules to automatically reply to users' common questions.

[0051] The data analysis module digitizes the user's purchase frequency, consumption amount, and purchase category preference feature data to form a feature vector and construct a data set X = {x1, x2, ..., x n}, where x i is the feature vector of the i-th user, n is the total number of users, and k initial cluster centers C = {c1, c2, ..., c k}, k is the number of user groups to be divided, and k=3 is set to represent high-value users, potential users, and ordinary users respectively; the Euclidean distance formula is used to express it:

[0052]

[0053] Among them, m is the dimension of the feature vector. By calculating each user feature vector x i With each cluster center c j The user is assigned to the cluster to which the nearest cluster center belongs based on the distance between them. For each cluster, its cluster center is recalculated, and the new cluster center is the mean of the feature vectors of all users in the cluster, that is:

[0054]

[0055] Among them, n j is the number of users in the jth cluster, S j is the set of users in the jth cluster; repeat the steps of calculating distance, assigning clusters and updating cluster centers until the cluster center no longer changes or the change is less than the set threshold, and the user group division is completed;

[0056] The Apriori algorithm is used to analyze the association between commodities. Frequent item sets are generated from transaction data, and the support (A) is calculated as:

[0057]

[0058] Among them, |D(A)| represents the number of transactions containing item set A, and |D| represents the total number of transactions. The association rule A→B is generated from the frequent item set, which is expressed as:

[0059]

[0060] The confidence level is calculated to measure the strength of the rule, indicating the probability of purchasing B after purchasing A, and to conduct differentiated marketing and combined sales. By analyzing these association rules, if it is found that users who purchase laptops are likely to purchase computer bags, combined sales can be conducted, and promotional information about computer bags can be pushed to users who purchase laptops to achieve differentiated marketing. By segmenting users, companies can develop personalized marketing strategies based on the characteristics of different groups to improve marketing effectiveness and user satisfaction.

[0061] The intelligent recommendation module is used to calculate similarity based on user historical behaviors and organize the user's historical purchase and browsing records into a user-product matrix. The rows of the matrix represent users, the columns represent products, and the matrix elements are the user's behavior data on the products. When the user set is U = {u1,u2,...,u m}, the commodity set is I = {i1,i2,...,i n}, then the element r of the user-item matrix R ijRepresents user u i Behavior data for product j, for user u a and user u b The cosine similarity is calculated as follows:

[0062]

[0063] Among them, r ai and r bi They are user u a and user u b For the behavior data of product i, the closer the cosine similarity value is to 1, the more similar the preferences of the two users are;

[0064] After obtaining the similarity between all users, for each target user, according to the similarity, select k users with high similarity as its similar user group, and its similar user set is For the target user, we filter out the products that the target user has not purchased from the purchase records of similar user groups, and perform weighted calculation based on the behavior data of similar users on the products to obtain the recommendation score of each product, which is expressed as:

[0065]

[0066] Among them, score(j) is the recommendation score of product j, sim(u x ,u) is the target user u x Similarity with similar user u, r uj Based on the behavior data of similar user u for product j, we recommend products that similar users have purchased. Recommendations are made based on the user's historical behavior and the preferences of similar users. The recommendation results are more in line with the user's personalized needs, increasing the user's interest in the recommended products and the possibility of purchasing, increasing user stickiness and platform sales; at the same time, we tap into the user's potential needs and discover new consumption growth points.

[0067] Based on the user's historical purchase and browsing history, the similarity between users is calculated to find user groups with similar interests and hobbies. When a user browses or prepares to buy a product, the system refers to the purchase behavior of similar users and recommends products that they have purchased before but the current user has not purchased yet, thereby increasing the probability of users discovering products of interest and promoting transaction conversion.

[0068] The intelligent recommendation module also recommends matching products based on user preferences. By extracting product attributes and descriptive keywords from product description texts, the user's historical purchase and browsing records are converted into user preference vectors. The user preference keyword set is P = {p1, p2, ..., p s}, for each keyword p j, giving weights based on the user's behavior data on products containing the keyword Form the user preference vector V p ;

[0069] For each product, the product feature vector V is formed based on the extracted keyword set. g , use cosine similarity to calculate the similarity between the user preference vector and the product feature vector, expressed as:

[0070]

[0071] in, is the keyword p in the product feature vector corresponding to the user preference j The corresponding TF-IDF value, the higher the similarity, the more the product matches the user's preference. The products are sorted from high to low according to the similarity, and the products with high similarity are recommended to the user.

[0072] By extracting product attribute information, such as brand, material, function, style, etc., as well as keywords and key phrases in product description text, when a user expresses a preference for a certain type of product, the system will filter out matching products based on the product content features for recommendation, thus satisfying the user's needs for products with specific attributes. Recommendations are made directly based on the user's clearly expressed preferences, with high accuracy; helping users quickly find products that meet specific attribute requirements, thus improving the user's shopping experience; for newly listed products or products in the cold start stage, recommendations can also be made through content matching, thus solving the cold start problem.

[0073] The two recommendation methods in the intelligent recommendation module can significantly improve the shopping experience and satisfaction of e-commerce platform users. This personalized recommendation method not only promotes users' purchasing behavior, but also deepens the connection between users and the platform, thereby promoting brand loyalty.

[0074] The supply chain management module collects historical sales data and arranges them in chronological order to form a time series {y t}, t=1,2,...,T, where y t represents the sales volume of the product at time t, where T is the total duration of the historical data;

[0075] Use historical sales data to train the ARIMA model and estimate the model parameters by minimizing the error between the predicted value and the true value; Based on the trained ARIMA model, combine future seasonal information and promotion activity plans to predict the demand for goods in the next h moments During the forecasting process, the model parameters, known historical data, and future input features are substituted into the model formula for calculation to predict future commodity demand and complete the inventory forecast.

[0076] The ARIMA model combines historical sales data, seasonal factors, promotional activities and other information to predict future demand for goods. Based on the forecast results, companies can formulate reasonable replenishment plans to avoid inventory backlogs or out-of-stock situations, reduce inventory holding costs, and improve capital turnover.

[0077] The supply chain management module also plans the optimal logistics path based on multiple factors. The logistics distribution network is abstracted into a directed graph G = (V, E), where V is a node set, including warehouses, distribution centers and customer addresses, and E is an edge set, representing the connection between nodes. Each edge (i, j) ∈ E corresponds to a transportation distance d. ij , transportation cost ij , delivery time t ij ; When there are m paths, the kth path P k By node sequence where v k1 Represents the warehouse as the starting node, Indicates the client address as the terminating node;

[0078] Based on multiple factors, the objective function is expressed as:

[0079] Z k =ω1D k +ω2C k +ω3T k

[0080] Among them, ω1, ω2, ω3 are corresponding weight coefficients, and ω1+ω2+ω3=1; the optimal logistics path is obtained by solving.

[0081] By considering factors such as transportation distance, transportation cost, and delivery time constraints, we plan the optimal route for logistics distribution to ensure that goods can be delivered to customers in the shortest time and at the lowest cost, thereby improving logistics distribution efficiency and enhancing customer satisfaction.

[0082] Embodiment 2: This embodiment is basically the same as the previous embodiment, except that: the marketing automation module automatically triggers the sending of personalized emails based on the user's behavior on the platform, such as registration, product browsing, adding to the shopping cart without purchase, and completing the purchase, as well as the user's preference information, by using marketing automation tools such as Mailchimp. For example, a reminder email containing a product coupon is sent to a user who has added a product to the shopping cart but has not purchased it, prompting him to complete the purchase; a welcome email is sent to a newly registered user and introduces the platform's featured products and promotions.

[0083] We also use social media management tools such as Hootsuite to develop social media content publishing plans in advance, publish product promotion copy, pictures, videos and other content on a regular basis, and keep the brand active on social media. At the same time, we set up automatic reply rules to automatically reply to the common questions of users in the comment area and private messages with corresponding answers, improve user communication efficiency, and enhance user stickiness.

[0084] Importantly, it should be noted that the construction and arrangement of the present application shown in a plurality of different exemplary embodiments are only exemplary. Although only a few embodiments are described in detail in this disclosure, it should be readily understood by those who refer to this disclosure that many modifications are possible without substantially departing from the novel teachings and advantages of the subject matter described in the application. Without departing from the scope of the present invention, other replacements, modifications, changes and omissions may be made in the design, operating conditions and arrangement of the exemplary embodiments. Therefore, the present invention is not limited to specific embodiments, but extends to a variety of modifications still falling within the scope of the appended claims.

[0085] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment (ie, those features that are not relevant to the best mode presently contemplated for carrying out the invention or those that are not relevant to implementing the invention) may not be described.

[0086] It should be understood that in the development of any actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but for those of ordinary skill having the benefit of this disclosure, without undue experimentation, the development effort will be a routine task of design, fabrication, and production.

[0087] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A cross-border e-commerce platform intelligent operation system, characterized in that: include: The data collection module is used to obtain transaction data, user behavior data, and package transportation status data from cross-border e-commerce platforms, social media, and third-party logistics service providers, and to aggregate and integrate multi-source data in the data storage center and pre-process the data; The data analysis module is used to divide user groups through clustering algorithms, and to mine and analyze product associations using association rules, so as to conduct differentiated marketing and combined sales; The intelligent recommendation module is used to calculate similarity based on user historical behaviors, recommend products that similar users have purchased, extract product attributes and description keywords in product description text, and recommend matching products based on user preferences; Supply chain management module, which is used to predict commodity demand through time series analysis algorithms, perform inventory forecasts, and plan the optimal logistics path based on multiple factors; The marketing automation module is used to automatically trigger the sending of personalized emails based on user behavior on the platform and user preference information, publish product content through social media on a scheduled basis, and set automatic reply rules to automatically reply to users' common questions.

2. A cross-border e-commerce platform intelligent operation system as claimed in claim 1, characterized in that: The data analysis module numerically processes the user's purchase frequency, consumption amount, and purchase category preference feature data to form a feature vector and construct a data set X = {x1, x2, ..., x n }, where x i is the feature vector of the i-th user, n is the total number of users, and k initial cluster centers C = {c1, c2, ..., c k }, k is the number of user groups that are expected to be divided; it is expressed using the Euclidean distance formula: Among them, m is the dimension of the feature vector. By calculating each user feature vector x i With each cluster center c j The user is assigned to the cluster to which the nearest cluster center belongs based on the distance between them. For each cluster, its cluster center is recalculated, and the new cluster center is the mean of the feature vectors of all users in the cluster, that is: Among them, n j is the number of users in the jth cluster, S j is the set of users in the jth cluster; repeat the steps of calculating distance, assigning clusters and updating cluster centers until the cluster center no longer changes or the change is less than the set threshold, and the user group division is completed; The Apriori algorithm is used to analyze the association between commodities. Frequent item sets are generated from transaction data, and the support (A) is calculated as: Among them, |D(A)| represents the number of transactions containing item set A, and |D| represents the total number of transactions. The association rule A→B is generated from the frequent item set, which is expressed as: The confidence level is calculated to measure the strength of the rule, indicating the probability of purchasing B after purchasing A, and to conduct differentiated marketing and combination sales.

3. The cross-border e-commerce platform intelligent operation system as claimed in claim 1, characterized in that: The intelligent recommendation module is used to calculate similarity based on user historical behaviors, and organize the user's historical purchase and browsing records into a user-product matrix. The rows of the matrix represent users, the columns represent products, and the matrix elements are the user's behavior data on the products. When the user set is U = {u1,u2,...,u m }, the commodity set is I = {i1,i2,...,i n }, then the element r of the user-item matrix R ij Represents user u i Behavior data for product j, for user u a and user u b The cosine similarity is calculated as follows: Among them, r ai and r bi They are user u a and user u b For the behavior data of product i, the closer the cosine similarity value is to 1, the more similar the preferences of the two users are; After obtaining the similarity between all users, for each target user, according to the similarity, select k users with high similarity as its similar user group, and its similar user set is For the target user, we filter out the products that the target user has not purchased from the purchase records of similar user groups, and perform weighted calculation based on the behavior data of similar users on the products to obtain the recommendation score of each product, which is expressed as: Among them, score(j) is the recommendation score of product j, sim(u x ,u) is the target user u x Similarity with similar user u, r uj Based on the behavior data of similar user u on product j, we recommend products that similar users have purchased.

4. The cross-border e-commerce platform intelligent operation system as claimed in claim 1, characterized in that: The intelligent recommendation module also recommends matching products according to user preferences. By extracting product attributes and descriptive keywords in product description text, the user's historical purchase and browsing records are converted into user preference vectors. The user preference keyword set is P = {p1, p2, ..., p s }, for each keyword p j , giving weights based on the user's behavior data on products containing the keyword Form the user preference vector V p ; For each product, the product feature vector V is formed based on the extracted keyword set. g , use cosine similarity to calculate the similarity between the user preference vector and the product feature vector, expressed as: in, is the keyword p in the product feature vector corresponding to the user preference j The corresponding TF-IDF value, the higher the similarity, the more the product matches the user's preference. The products are sorted from high to low according to the similarity, and the products with high similarity are recommended to the user.

5. The cross-border e-commerce platform intelligent operation system as claimed in claim 1, characterized in that: The supply chain management module collects historical sales data and arranges them in chronological order to form a time series {y t }, t=1,2,...,T, where y t represents the sales volume of the product at time t, where T is the total duration of the historical data; Use historical sales data to train the ARIMA model and estimate the model parameters by minimizing the error between the predicted value and the true value; Based on the trained ARIMA model, combine future seasonal information and promotion activity plans to predict the demand for goods in the next h moments During the forecasting process, the model parameters, known historical data, and future input features are substituted into the model formula for calculation to predict future commodity demand and complete the inventory forecast.

6. A cross-border e-commerce platform intelligent operation system as claimed in claim 5, characterized in that: The supply chain management module also plans the optimal logistics path based on multiple factors by abstracting the logistics distribution network into a directed graph G = (V, E), where V is a node set including warehouses, distribution centers and customer addresses, and E is an edge set representing the connection between nodes. Each edge (i, j) ∈ E corresponds to a transportation distance d ij , transportation cost ij , delivery time t ij ; When there are m paths, the kth path P k By node sequence where v k1 Represents the warehouse as the starting node, Indicates the client address as the terminating node; Based on multiple factors, the objective function is expressed as: Z k =ω1D k +ω2C k +ω3T k Among them, ω1, ω2, ω3 are corresponding weight coefficients, and ω1+ω2+ω3=1; the optimal logistics path is obtained by solving.

Images