Anti-counterfeiting intelligent label, two-way prediction sales system and method

Abstract

The present invention discloses an anti-counterfeiting smart label, a two-way predictive sales system and method, comprising a first flexible material layer, a traction layer and a double-sided adhesive layer arranged in sequence from top to bottom, and also comprising a smart label, wherein the smart label is provided with a chip and an antenna circuit, the chip is placed on the upper surface of the first flexible material layer, the antenna circuit is fixed on the upper surface of the first flexible material layer, and the first flexible material layer is provided with a mechanical gap on a side close to the chip; the traction layer comprises a semicircular second flexible material layer, the second flexible material layer is connected with a semicircular tear-type safety ring, the second flexible material layer is fixedly connected to the first flexible material layer and the double-sided adhesive layer, the tear-type safety ring is connected to the chip via a traction rope passing through the mechanical gap, and the two-way predictive sales system is provided with the above-mentioned anti-counterfeiting smart label, and the present invention arranges a tear-type safety ring opening device on the smart anti-counterfeiting label to destroy the chip position, thereby improving its anti-counterfeiting performance.

Description

Technical Field

[0001] The present invention relates to the field of logistics technology, and in particular to an anti-counterfeiting intelligent label, a two-way forecasting sales system and a method. Background Art

[0002] In today's era of rapid development of commodity economy, brand competition in the market is becoming more and more fierce, which is also stimulating various brands to constantly update their own technologies. However, the imitation technology in the market is also constantly updated and iterated. As we all know, counterfeit and inferior products are simply copied or even stolen brands, thereby using inferior products or even unqualified harmful products to defraud consumers. This not only causes huge losses to merchants and consumers, but also deeply affects the country's economic order. Therefore, merchants are committed to implementing various anti-counterfeiting methods and applying them to their own products, especially valuable products, and they must make anti-counterfeiting technology difficult to copy. RFID tags and NFC tags, which are common and better anti-counterfeiting methods on the market, can use the global uniqueness of the electronic tag ID number to achieve the purpose of anti-counterfeiting, but traditional electronic tags also have the possibility of being stolen and transferred for the second time, so their anti-counterfeiting performance needs to be updated.

[0003] With the further development of the information age, the increasing use of the Internet of Things, big data and cloud databases, as well as the popularization of electronic devices that support NFC / RFID technology, the popularity of NFC / RFID technology in product anti-counterfeiting has continued to increase.

[0004] After the circuit made of liquid metal is destroyed, it is almost impossible to copy or even transfer. The new metal circuit printing technology uses flexible electronic circuit materials to quickly, efficiently and flexibly form conductive circuits and graphics on the substrate. At the same time, the product is thin and bendable. Therefore, the circuit made of this flexible material not only has the stability of ordinary copper foil material circuits, but also has a certain degree of flexibility. Based on this, it is difficult to copy and transfer after the original chip position is destroyed.

[0005] The defects of the prior art are that the circuit structures of the existing NFC electronic tags and RFID electronic tags are relatively stable, and they are easily transferred and reused maliciously, and their anti-counterfeiting performance is low. Summary of the invention

[0006] In view of at least one defect of the prior art, the object of the present invention is to provide an anti-counterfeiting smart label. The tear-type safety ring opening device arranged on the smart anti-counterfeiting label has a destructive effect on the position of the chip, thereby disconnecting the chip from the antenna, reducing the possibility of malicious transfer and secondary use, thereby greatly improving its anti-counterfeiting performance.

[0007] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: an anti-counterfeiting smart label, comprising a circular first flexible material layer, a traction layer and a double-sided adhesive layer arranged in sequence from top to bottom, and also comprising an NFC / RFID electronic smart label, the NFC / RFID electronic smart label is provided with a chip, the chip is connected to an antenna circuit, the chip is placed on the upper surface of the first flexible material layer, the antenna circuit is fixedly arranged on the upper surface of the first flexible material layer, and the first flexible material layer has a mechanical gap on one side close to the chip; the traction layer comprises a semicircular second flexible material layer, the second flexible material layer is connected to a semicircular tear-type safety ring, the second flexible material layer is fixedly connected to the first flexible material layer and the double-sided adhesive layer, the tear-type safety ring is connected to the chip via a traction rope passing through the mechanical gap, and the binding force between the traction rope, the tear-type safety ring and the chip is greater than the binding force between the chip and the antenna circuit.

[0008] The center line of the double-sided adhesive layer is a pressed die line, and the lower surface of the double-sided adhesive layer is divided into a layer with stronger adhesion and a layer with weaker adhesion by the die line. The second flexible material layer contacts and is fixedly connected to the upper surface of the double-sided adhesive layer corresponding to the layer with weaker adhesion, and the tear-off safety ring contacts the upper surface of the double-sided adhesive layer corresponding to the layer with stronger adhesion.

[0009] A circular gasket layer is adhered to the first flexible material layer, and a circular third flexible material layer is adhered to the gasket layer; the third flexible material layer, the gasket layer and the first flexible material layer together form a receiving space for receiving the NFC / RFID electronic smart tag.

[0010] The antenna circuit is made of liquid metal or a printed metal antenna circuit; a release paper layer is arranged under the double-sided adhesive layer. One end of the tear-off safety ring is fixedly connected to a corner of the second flexible material layer, and the other end is provided with a pull ring or a pull handle.

[0011] On the other hand, existing sales forecasts are mostly based on algorithmic forecasts of historical sales data and historical event data of manufacturers, without considering the information processing sales forecast at the user end; but it is worth noting that the data at the user end can directly reflect the actual usage data of consumers, which cannot be ignored. In order to successfully capture the actual usage data of consumers and make corresponding predictions, we added corresponding functions on the basis of the aforementioned anti-counterfeiting labels, and considered the supporting terminal equipment for obtaining user information in the design, and used it for sales forecasting.

[0012] A two-way forecasting sales system including an anti-counterfeiting smart label includes a data collection and upload module, a data classification module, a data initialization module, a sales forecast module, an information feedback module, and a data presentation terminal; wherein the data collection and upload module includes a collection unit, a sample desensitization unit, and a data upload unit, and the data collection and upload module is arranged in a user device, a dealer device, and a manufacturer device, and collects user-side data by scanning the anti-counterfeiting smart label through the user's mobile phone device, and uploads the data after desensitization processing is completed in the consumer's mobile phone device, thereby protecting the consumer's privacy. Other modules can be arranged in a server.

[0013] The data classification module performs preliminary labeling on the data uploaded from the data acquisition module to prepare for the prediction and further processing of subsequent information data.

[0014] The data initialization module initializes the basic data based on the attributes of the enterprise's products so that the prediction framework does not deviate from the basic attributes of the goods. When the user purchases again and the data accumulates to a certain amount, the sales prediction module uses the sales prediction algorithm based on the data accumulated in the data classification module to make a specific analysis and prediction of the future sales of the goods sold by the enterprise. The sales prediction module includes a model storage unit and a data prediction unit.

[0015] The information feedback module collects specific data after making the initial forecast and feeds it back to the sales forecast module to provide support for adjusting the forecast model and improving accuracy.

[0016] The data presentation terminal displays and outputs sales forecast results to people, and includes the functions of displaying different information to different groups and collecting different information to different groups.

[0017] Significant effect: The present invention provides an anti-counterfeiting smart label and a two-way predictive sales system. The tear-type safety ring opening device arranged on the smart anti-counterfeiting label destroys the position of the chip, thereby disconnecting the chip from the antenna and reducing the possibility of malicious transfer and secondary use, thereby greatly improving its anti-counterfeiting performance. In addition, the use of the above-mentioned device in the setting of the corresponding prediction system solves the problem in the existing prediction technology that the factors considered are affected by the changes in the user's own situation and it is difficult to obtain relevant data, resulting in inaccurate prediction results, which provides strategic value for production enterprises to comprehensively coordinate and dispatch resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is an exploded structural diagram of an implementation method of an anti-counterfeiting smart label;

[0019] Figure 2 a is a half-section view of the main view when the anti-counterfeiting smart label is attached;

[0020] Figure 2b is a partial cross-sectional view of the top view when the anti-counterfeiting smart label is attached;

[0021] Figure 3 It is the appearance structure diagram and regional structure segmentation diagram of the anti-counterfeiting smart label;

[0022] Figure 4 Schematic diagram of the mounting method for anti-counterfeiting smart labels;

[0023] Figure 5 Schematic diagram of the opening process used for anti-counterfeiting smart labels;

[0024] Figure 6 It is a top view perspective view of each layer when the anti-counterfeiting smart label is attached;

[0025] Figure 7 A flow chart of a sales forecasting system based on a two-way forecasting concept disclosed in an embodiment of the present invention;

[0026] Figure 8 It is a flow chart of a data classification module in a sales forecasting system disclosed in an embodiment of the present invention;

[0027] Fig. 9 It is a flow chart of a data initialization module in the sales forecasting system disclosed in an embodiment of the present invention;

[0028] Fig.10 A flow chart of a sales forecasting module in a sales forecasting system disclosed in an embodiment of the present invention;

[0029] Fig.11 It is a functional flow chart of the supporting device in the sales forecasting system disclosed in the embodiment of the present invention;

[0030] Fig.12 This is the interface diagram for users to evaluate products;

[0031] Fig.13 This is an interface diagram for users to verify the authenticity of goods;

[0032] Fig.14 This is an interface diagram for dealers to view the overall sales status of goods;

[0033] Fig.15 This is an interface diagram for dealers to view the specific sales status of a certain product;

[0034] Fig.16 This is an interface diagram for manufacturers to view the overall sales status of goods;

[0035] Fig.17 This is the interface for manufacturers to view the sales status of each dealer. DETAILED DESCRIPTION

[0036] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] like Figure 1-Figure 17 As shown, an anti-counterfeiting smart label comprises a circular first flexible material layer 1, a traction layer 2 and a double-sided adhesive layer 3 arranged in sequence from top to bottom, and also comprises an NFC / RFID electronic smart label 4, the NFC / RFID electronic smart label 4 is provided with a chip 41, the chip 41 is connected to an antenna circuit 42, the chip 41 is placed on the upper surface of the first flexible material layer 1, the antenna circuit 42 is fixedly arranged on the upper surface of the first flexible material layer 1, and the first flexible material layer 1 is provided with a mechanical gap 43 on one side close to the chip 41; the traction layer 2 comprises a semicircular second flexible material layer 21, the second flexible material layer 21 is connected to a semicircular tear-type safety ring 22, the second flexible material layer 21 is fixedly connected to the first flexible material layer 1 and the double-sided adhesive layer 3, the tear-type safety ring 22 is connected to the chip 41 via a traction rope 23 passing through the mechanical gap 43, and the binding force between the traction rope 23, the tear-type safety ring 22 and the chip 41 is greater than the binding force between the chip 41 and the antenna circuit 42. The traction rope 23 is an inelastic rope that cannot be stretched and is in a relaxed state when not pulled. During the opening process, it moves with the tear-off safety ring 22 to reach the maximum stretching state until the chip 41 and the antenna circuit 42 are disconnected. The semicircular second flexible material layer 21 and the semicircular tear-off safety ring 22 together form a circular traction layer 2.

[0038] The anti-counterfeiting smart label is adhered to the opening gap of the commercial box through the double-sided adhesive layer 3. When the user opens the product, the tear-off safety ring 22 is pulled to pull the traction rope 23, and the chip 41 is pulled by the traction rope 23. Since the binding force between the traction rope 23, the tear-off safety ring 22 and the chip 41 is greater than the binding force between the chip 41 and the antenna circuit 42, the chip 41 is pulled apart from the antenna circuit 42 by the traction rope 23, thereby destroying the anti-counterfeiting smart label and preventing it from being maliciously transferred for secondary use. Since the chip 41 is separated from the original antenna circuit 42 during this process, and the pins of the chip 41 may also be destroyed, the anti-counterfeiting label cannot be used for a second time.

[0039] One end of the traction rope 23 is fixedly connected to the inner side of the tear-off safety ring 22, and the other end is fixedly connected to the top surface of the chip 41. The center line of the double-sided adhesive layer 3 is a pressed die line 3a, and the lower surface of the double-sided adhesive layer 3 is divided into a layer with strong adhesion 3b and a layer with weak adhesion 3c by the die line 3a. The second flexible material layer 21 contacts and is fixedly connected to the upper surface of the double-sided adhesive layer 3 corresponding to the layer with weak adhesion 3c, and the tear-off safety ring 22 contacts the upper surface of the double-sided adhesive layer 3 corresponding to the layer with strong adhesion 3b. The bonding force between the layer with strong adhesion 3b and the product is greater than the bonding force between the upper surface of the double-sided adhesive layer 3 and the second flexible material layer 21, and the bonding force between the layer with weak adhesion 3c and the product.

[0040] like Figure 1 and Figure 5 As shown, before pasting, the release paper layer 7 needs to be torn off, and the die line 3a on the double-sided adhesive layer 3 is aligned with the opening gap of the product. The second flexible material layer 21 corresponds to the weaker adhesive layer 3c, the tear-type safety ring 22 corresponds to the stronger adhesive layer 3b, the die line 3a corresponds to the opening gap of the product box, the stronger adhesive layer 3b is adhered to the upper box cover, and the weaker adhesive layer 3c is adhered to the lower box cover. When the user pulls the tear-type safety ring 22, the chip 41 and the antenna circuit 42 are first pulled apart by the traction rope 23, and then the tear-type safety ring 22 is continued to be pulled, and the double-sided adhesive layer 3 is separated along the die line 3a, and the double-sided adhesive layer 3 corresponding to the weaker adhesive layer 3c is torn off together with the first flexible material layer 1 and the traction layer 2, and the double-sided adhesive layer 3 corresponding to the stronger adhesive layer 3b remains on the upper box cover.

[0041] like Figure 1 As shown, a circular gasket layer 5 is also adhered to the first flexible material layer 1, and a circular third flexible material layer 6 is also adhered to the gasket layer 5; the third flexible material layer 6, the gasket layer 5, and the first flexible material layer 1 together form a receiving space 8 for receiving the NFC / RFID electronic smart tag 4. Product trademarks, etc. can be set on the upper surface of the third flexible material layer 6. The thickness of the gasket layer 5 is slightly higher than that of the NFC / RFID electronic smart tag 4. The antenna circuit 42 is made of liquid metal or a printed metal antenna circuit; a release paper layer 7 is provided under the double-sided adhesive layer 3. The antenna circuit 42 is made of liquid metal or copper foil, but its bonding force with the chip 41 is much smaller than the bonding force between the traction rope 23 and the tear-type safety ring 22 and the chip 41.

[0042] like Figure 1 and Figure 2 As shown, one end of the tear-off safety ring 22 is fixedly connected to a corner of the second flexible material layer 21, and the other end is provided with a pull ring or handle 22a. The user pulls the pull ring or handle 22a to drive the tear-off safety ring 22 to rotate, thereby pulling the chip 41.

[0043] like Figure 1 and Figure 2 As shown, a plurality of gear teeth are arranged on the inner side of the tear-off safety ring 22, and one end of the traction rope 23 is fixed on a key gear tooth 22b in the middle. The smart tag uses liquid metal or printed metal to make the antenna circuit, and utilizes the destructibility of the liquid metal or printed metal antenna circuit to destroy the position of the chip 41 through the key gear teeth 22b of the tear-off safety ring 22 and the traction rope 23, thereby achieving the purpose of preventing the tag from being forged and transferred, and improving the anti-counterfeiting ability of the smart tag.

[0044] like Figure 3 , Figure 4 , Figure 5 As shown, the technical solution of the present invention is suitable for products or product packaging boxes with smooth surface and tight packaging, such as jewelry, high-end tea and other valuables packaging boxes. It effectively prevents counterfeiting of traditional flip-top and twist-open packaging boxes and goods. First, when attaching the label, align the die line 3a with the gap at the opening of the packaging box or the product. Taking advantage of the flexible nature of liquid metal or printed metal circuits, when it is necessary to open an item, first pull the tear-type safety ring 22 through the pull handle 22a to open it. When the pull handle 22a rotates to open it, the key gear 22b follows the movement and drives the chip 41 to fall off the original antenna through the traction rope 23 through the arc-shaped mechanism gap 43, thereby achieving the purpose of destroying the electronic tag; since the gear of the tear-type safety ring 22 has only half a circumference, and after the pull handle 22a is pulled to open it, all the structures above the tear-type safety ring 22 are taken away by pulling, and at the same time, the double-sided adhesive layer 3 is separated from the knife die line 3a, and the weaker adhesive layer 3c is separated together with the tear-type safety ring 22, preventing the label from being transferred and forged while ensuring that the liquid metal or printed metal circuit will not be exposed, and the remaining strong adhesive layer 3b is retained on the product, at this time the opening gap of the product is exposed and can be directly opened. On the one hand, the existing sales forecasts are mostly based on the enterprise's historical sales data and historical event data for algorithmic forecasting, without considering the user-side information processing sales forecast; but it is worth noting that the user-side data can directly reflect the actual usage data of consumers, which cannot be ignored. To this end, we propose a two-way forecasting sales system including the anti-counterfeiting smart label.

[0045] A two-way predictive sales system including the anti-counterfeiting smart label includes a data initialization module, a data collection and upload module, a data classification module, a sales prediction module, an information feedback module and a data presentation terminal; wherein the data collection and upload module includes a collection unit, a sample desensitization unit and a data upload unit, which completes the desensitization processing of the data information in the consumer's mobile phone device and then uploads it to protect the consumer's privacy.

[0046] The data classification module performs preliminary labeling on the data uploaded from the data collection module to prepare for the prediction and further processing of subsequent information data. When the user purchases again and the data accumulates to a certain amount, the sales forecasting module uses the sales forecasting algorithm based on the data accumulated in the data classification module to make a specific analysis and forecast of the future sales of the company's products.

[0047] The data initialization module initializes the basic data based on the attributes of the enterprise's products so that the forecasting framework does not deviate from the basic attributes of the products. For example, the seasonal weights of clothing products are optimized, and different regions in my country are divided into different areas and coded accordingly based on the climate differences. After initialization, the data is stored in the model storage unit in the sales forecasting module and enters the sales forecasting module.

[0048] The sales forecasting module includes a model storage unit and a data forecasting unit.

[0049] The information feedback module collects specific data after making the initial forecast and feeds it back to the sales forecast module to provide support for adjusting the forecast model and improving accuracy.

[0050] The data presentation terminal displays and outputs sales forecast results to people, and includes the functions of displaying different information to different groups and collecting different information to different groups.

[0051] Specifically, the data collection and upload module collection method is as follows:

[0052] Collection unit: On the one hand, it can be used to obtain the user ID and a series of behavioral data associated with the ID, such as: the time and place when the user uses the device to check the authenticity after purchase, and the time and place when the product is opened; the number of times the user has purchased products in history and related information, the device information used by the user to check the authenticity, and the product information opened by the user; on the other hand, it can be used to obtain historical sales data of the enterprise, related product supply in the warehouse, product production (such as production date, production location, production batch) and other related information; in addition, the system also designs a method for obtaining information from dealers, and dealers can upload historical sales data in a package, and the missing parts of historical sales data can be supplemented by user-side data collection. Sample desensitization unit: Desensitize the information obtained by the collection unit locally on the user's device, such as converting the acquisition of the IP address into address information within a larger area, converting the physical address of the user's device into a virtual ID used only for identification features, etc. Data upload unit: Encrypt the desensitized data and store it, and wait for the device to be encrypted and uploaded to the server after connecting to the network.

[0053] Furthermore, the data classification module is used to perform preliminary processing on the data uploaded by the acquisition unit, label the information data, store it in categories, and select three to four important variables to obtain the corresponding weights. For example: based on the company's sales situation, during holidays and nearby periods, the abnormal values ​​of sales data are marked and processed; based on the time and number of times the user uses the device to verify the authenticity of the product, the user's active time period, active time interval, etc. can be analyzed; based on the time information of the user's repurchase, the repurchase ratio can be calculated; based on the address where the user opens the product and the solar term at that time, different regions can be divided to avoid the impact of different climates on subsequent predictions.

[0054] Furthermore, the sales forecasting module uses the data processed by the data classification module to make sales forecasts. Specifically, the sales forecasting module prediction method is as follows:

[0055] Model storage unit: connected with the data initialization module and the information feedback module, determines the weights of various influencing factors of the initial prediction model according to the basic attributes of the product, such as seasonality and regionality, and stores the prediction model.

[0056] Data prediction unit: In terms of users, the labeled data in the model storage unit and the behavioral information of users with similar labels after preliminary matching are used to predict the future repurchase ratio Y using a long-term and short-term neural network. p In terms of production enterprise data, the cubic exponential smoothing algorithm can be used to predict the future sales forecast E1 for seasonal products based on the correlation between historical sales data and events. For other products, different weights can be given to time factors such as season, usage demand, holidays, etc. in the model storage unit, and the bidirectional LSTM network can be used to predict future sales E2.

[0057] Furthermore, the information feedback module records the difference between actual sales and future predicted sales, transmits it back to the sales forecast module, continuously adjusts the weights of different prediction influencing factors, further trains and optimizes the sales forecast model, and stores the better prediction model obtained after repeated iterations back into the model storage unit, saving prediction time for the next call of the model. The prediction values ​​that are difficult to fit and the results with a large difference between the predicted value and the true value are marked to analyze whether there are other influencing factors; and the influencing factors that are not significant to the model prediction results are pruned to make the prediction results fit the true value appropriately.

[0058] Furthermore, in the data presentation terminal, on the one hand, the enterprise can authorize dealers to view the local sales forecast results, so that dealers can refer to the forecast results to purchase and distribute goods. At the same time, the production enterprise can refer to the sales status of different dealers in different regions, and have a stronger control over the downstream. On the other hand, after registration, users can express their views and opinions on the product to other users through the data presentation terminal. Other users can perform operations such as likes, dislikes, comments, and reports, which are similar to interactive sharing communities. In particular, the sales forecasting system uses text semantic analysis algorithms, such as long short-term memory networks and LDA methods, to intelligently analyze the semantics of user-posted texts, and give points from negative to positive according to the emotional content in the text (1 point for negative, 9 points for positive). Other users' likes, dislikes, and other operations are given different weights accordingly to associate them with the main user. Similarly, other users' comment behaviors can also be semantically analyzed to establish a huge association network, thereby increasing our understanding of users and providing a reference for matching behaviors in the data classification module and subsequent further predictions.

[0059] Specifically, we provide three different interfaces for three different groups:

[0060] For registered users: Provide a community-like interactive interface. Users can enter the community and directly share their views on a certain product, their daily life, recommendations, etc., and can make corresponding star ratings on the product experience after verifying the authenticity and using it for a period of time. For dealers, we provide data visualization analysis to help dealers "understand" the data. On the one hand, the data can be used to predict future sales and give suggestions for the next purchase under our analysis. On the other hand, it can encourage dealers to retain and upload historical sales data to help make more comprehensive forecasting analysis. In addition, we provide different permission levels for dealers in different jurisdictions, so that they can view different sales data visualization analysis results.

[0061] For manufacturers, it can directly obtain sales conditions in all regions and visualize sales data analysis results, including the sales rankings of different dealers. At the same time, it can also make predictions about the overall future sales and provide suggestions for the next distribution and supply.

[0062] Additionally, dealers and manufacturers can register accounts in the user-side community. After we give them authentication, dealers and manufacturers can directly obtain real-time feedback from users. Based on actual needs, the system extracts the most frequently appearing keywords and feeds them back to dealers and manufacturers. On the one hand, the system shortens the distance between consumers, dealers, and manufacturers, and on the other hand, it builds the prototype of a rich and complete content ecosystem.

[0063] The benefits of the sales forecasting supporting system based on the idea of ​​two-way forecasting are: 1. In addition to considering the historical sales data on the enterprise side, the present invention also incorporates the user-side data as a reference factor into the product sales forecast, which can obtain more comprehensive forecasting results and reduce the noise in the forecast compared to simply considering the historical sales data of the enterprise. 2. The device design used in the present invention solves the time cost problem of needing to design a set of forecasting models by themselves for small and medium-sized enterprises, and the problem that the factors considered in the existing forecasting technology are affected by the changes in the user's own situation and it is difficult to obtain relevant data, resulting in inaccurate forecasting results, and provides strategic value for the comprehensive coordination and scheduling of resources for enterprises. 3. Regarding the relationship between enterprises and distributors, the supporting device proposed in the present invention enables enterprises to have a clearer understanding of the downstream sales situation, and can refer to the sales situation of different distributors in the region where the resources are reasonably allocated to downstream distributors, thereby enhancing the enterprise's control over the downstream.

[0064] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0065] Reference Figure 7-Figure 17This embodiment discloses a sales forecasting system based on the idea of ​​two-way forecasting, including the following modules: Data collection and upload module: mainly collects part of the user's data through the sales forecasting supporting device, and encrypts and uploads the data in a way that only retains the feature distribution without affecting the user's privacy. The main collected information includes: the time, location and number of times the user accesses the system, the user's device, and the user's unique feature identification code; the manufacturer and distributor provide a data entry interface, which can batch input the product production area, production time, production batch, product sales quantity, sales time, sales batch, etc. as needed, and generate a batch number containing the above information.

[0066] Data classification module: After the product has been sold for a period of time, the data collection module will collect enough data and then sort and classify the data. The specific steps are as follows:

[0067] Step A1: Single user information collation: The time when a single user feature identification code accesses the system is collated to generate new features, such as user active time period, user active time interval, etc.;

[0068] That is, a single user set contains multiple feature variables, and single user data is stored in a single user vector;

[0069] Generate new feature variables according to time. For example, the user active time period is divided into early morning (x=0, 6:00-8:00), morning (x=1, 8:00-11:00), noon (x=2, 11:00-13:00), afternoon (x=3, 13:00-17:00), evening (x=4, 17:00-19:00), night (x=5, 19:00-23:00), and midnight (x=6, 23:00-6:00) according to the time when the user enters the system, where x represents the value assigned to the new feature variable after the system is sorted out; the user active time interval is divided into less than one hour (x=0), one to two hours (x=1), two to three hours (x=2), and more than three hours (x=3) according to the time interval between the user entering the system and exiting the system, where x represents the value assigned to the new feature variable after the system is sorted out.

[0070] The location where a single user feature identification code accesses the system is obtained and associated by the manufacturer entering the product batch number, bound to the corresponding regional code, and assigned to the corresponding feature new variable. For example: different regions in my country are divided into different regions (Beijing, Shanghai, Shenzhen, Chongqing, etc.) according to the climate differences and corresponding codes (arranged in a specific dictionary in a predetermined order in advance), and the corresponding batch number is coded. A batch number contains a sequence to prove which dealer it belongs to and which region it is sold to. After the user verifies the authenticity, the system automatically obtains the corresponding data based on the product serial number. If the user turns on the positioning when verifying the authenticity and finds that the positioning is different from the information contained in the batch number (for example, across provinces), the data will be marked and included in the newly added abnormal coding feature variable of the actual positioning area.

[0071] Optionally, if a user registers an account and logs in, a single user feature identification code will be used to bind an account, and the interaction information between the user and other users will be recorded. The probability that the two are acquaintances will be analyzed based on the semantic analysis algorithm and included in the feature variable; optionally, when processing an IP address, if it is a company's IP address, the corresponding company information (time, location) will be searched and the user will be counted in the feature variable of "working time use" (working time use assigns the variable a value of 1, otherwise it assigns a value of 0); optionally, according to the location corresponding to the batch number, relevant weather, temperature, season and other factors can also be associated;

[0072] Step A2: Labeling of single user information: The obtained and sorted user information is stored in the corresponding row matrix in the order of the given labels. Different users correspond to different row matrices. Assuming there are n label dimensions and m users, an mxn matrix is ​​obtained.

[0073] Step A3: Determine users with similar behaviors and establish preliminary associations. To meet specific behaviors, extract each row in the matrix separately to form an n-dimensional vector. To determine the degree of association between users, a "match" is performed here, that is, the similarity between two vectors is determined. The following formula is used here:

[0074]

[0075] Where D(x i , x j ) represents two n-dimensional vectors x i and x j The similarity between -1Represents the inverse matrix of the covariance matrix of two vectors. Because this method has certain requirements on the system computing power, it is chosen to be executed in the data classification module. After the data upload module uploads a group of data, it will be processed by the classification module and matched with the n-dimensional vectors represented by other user data. The top three vectors with similarity are selected regularly for strong association, that is, they are given higher weights. Specifically, for D(x i , x j ) After calculating the first three values ​​and normalizing them, the corresponding weights of the first three vectors are obtained. Data initialization module: Initialize the basic data based on the attributes of the production enterprise's products. After initialization, store the data in the model storage unit in the sales forecast module and enter the sales forecast module.

[0076] Specifically, according to Fig. 9 ,right Fig. 9 The steps are described as follows:

[0077] Step D1: Given the user's repurchase history, obtain the user's repurchase time t and the user's repurchase ratio V for the product.

[0078] Step D2: Set the basic repurchase probability, that is, each user tag must be associated with a specific location. According to the known current user repurchase ratio in the area, set the initial basic repurchase probability p k .

[0079] Step D3: Based on the weights given by the classification module, the expected value p of the probability of repurchase of a single user is obtained Δ , the calculation formula is as follows:

[0080] p Δ =D1P1+D2P2+D3P3;

[0081] Step D4: The overall repurchase probability of the region is obtained using the following calculation formula:

[0082]

[0083] Step D5: Use the random forest algorithm to extract m (usually three to four) important variables from the feature variable set and initialize the data.

[0084] Using the random forest algorithm, we select the feature variables that have a greater impact on the repurchase ratio to participate in subsequent predictions. Specifically, we have the following steps: Step R1: Data normalization. Because different data have different dimensions, the data directly input into the calculation is quite different, so we use the following formula to normalize the data:

[0085]

[0086] Among them, x′ is the data obtained after normalization, x is the data before normalization, λmax is the maximum value of the series of data, and λmin is the minimum value of the series of data.

[0087] Step R2: ① Assume that there are k trees in the random forest, where the characteristic variable of the kth tree is x 1k …x nk ; Usually, considering the system performance and effect, it is more reasonable to set k to 100;

[0088] ② For each tree, a certain scale of data is randomly selected with replacement from the feature data set R (including all collectible feature variables, such as product seasonality, product periodicity, product life, user satisfaction with the product, etc.) through random resampling to form the sample training subset Ri, and the unselected data constitutes b Out-of-Bag data;

[0089] ③ Repeat steps a) to c) in Ri, and in each cycle, grow the decision tree as much as possible without pruning it to obtain a decision tree.

[0090] a) Randomly extract m attributes from the input feature attribute set as the attribute set for the current decision tree split. The value of m is determined by the number of feature variables M collected by the system, usually

[0091] m=sqrt(M)

[0092] Where sqrt() is the root function; or,

[0093] m=log2(M)+1

[0094] Configured according to the specific needs of different systems.

[0095] b) Select the best variable F and split point S from the m feature variables to get θ i (F, s).

[0096] Specifically, the CART regression method is used to determine the optimal split point and the best variable, and the optimal variable F and split point S corresponding to the minimum sum of squared errors are selected to obtain θ i (F, s). Correspondingly, we need to solve the following problem to find the most suitable θ that meets the conditions. i (F, s):

[0097]

[0098] where θ1(F, s) = {x|x(F)≤s}, θ2(F, s) = {x|x(F)>s}, is the corresponding data set y i The mean of y iRepresents the characteristic variable in R i The normalized value in .

[0099] c) The node is adjusted according to θ i (F, s) is split into two child nodes θ i1 (F, s) and θ i2 (F, s), divide the attribute values ​​less than s into one node, and the attribute values ​​greater than s into another node, and the node output value is the node mean.

[0100] ④ Randomly change the feature variables x in the Out-of-Bag data sample i The value of is used to generate new Out-of-Bag data test samples, and the new Out-of-Bag data is voted by the following formula;

[0101]

[0102] Here counter x i Represents the final classification x of each tree i Number of occurrences.

[0103] Get the score matrix corresponding to row A and column xb, where each row represents a change in x i H after data i , a total of A changes.

[0104] ⑤ For the characteristic variable x i Assign weight;

[0105] The specific formula is as follows:

[0106]

[0107] Among them, H F and H iF Respectively represent the Out-of-Bag error rate of the i-th tree before and after the variable changes; w xi Represents the contribution of each feature vector to the classification process, which is specifically converted into the corresponding weight.

[0108] Step D6: Save the generated model to the model storage unit and submit it for subsequent use. In addition, each time the user repurchases, the data is stored in the cache unit. The method of extracting the data update probability at a fixed time point every month is adopted to re-execute steps D1 to D6.

[0109] The sales forecasting module forecasting steps are as follows:

[0110] Step 1: Call the model storage unit connected to the data initialization unit and transfer the stored prediction model to the data prediction unit. Step 2: Enter the data prediction unit and make sales predictions in two aspects:

[0111] Step S1: Enter the data prediction unit. In terms of users, use the labeled data in the model storage unit and the behavioral information of users with similar labels after preliminary matching to use the long-term and short-term neural network to predict the future repurchase ratio Y p ; Specifically, it is divided into the following steps:

[0112] B1: Combine the important feature vectors selected by the random forest algorithm with the historical sales data as features to establish the prediction feature matrix x L =(x1, x2, ..., x n ,t,P Re , y) T , where x1, x2, ..., x n represents the corresponding characteristic variable, t represents the commodity usage cycle, P Re represents the overall repurchase probability of the product, and y represents the repurchase ratio of the product within the corresponding time.

[0113] B2: The feature set x L As the input layer, the middle layer is the hidden layer, and the final output is the predicted value of the future repurchase ratio Y p ;

[0114] Using the long-term and short-term network model, we build a corresponding calculation model for prediction. The specific steps are as follows:

[0115] C1: Constructing long-term and short-term network input gate i t ;

[0116] i t =δ(w ix x t +w ih h i-1 +b i )

[0117] C2: Constructing the long-term and short-term network output gate o t ;

[0118] o t =δ(w ox x t +w oh h t-1 +b o )

[0119] C3: Constructing long-term and short-term network forget gate f t ;

[0120] f t =δ(w fx x t +w fh h t-1 +bf )

[0121] C4: Constructing the intermediate output node g of the long-term and short-term network t and state unit S t ;

[0122] g t =φ(w gx x t +w gh h t-1 +b g )

[0123] S t =g t Θi t +S t-1 Θf t

[0124] h t =φ(S t )Θo t

[0125] δ and φ represent the sigmoid function and tanh function respectively; h t represents the intermediate input node; w ix , w ih , w ox , w oh , w fx , w fh , w gx and w gh Respectively represent the input x t and intermediate output h t-1 The matrix weight given by the random forest algorithm when multiplying with the corresponding gate; Θ represents the bitwise multiplication of the elements in the two vectors; b i , b o , b f and b g Represent the correction vectors of the corresponding gates respectively.

[0126] Through continuous iteration, repeat steps C1 to C4 and continuously update until the loss function value is less than the set error value ε, and output Y p .

[0127] The activation function φ and loss function selected by the model are:

[0128]

[0129] L((Y p ), f(X L ))=(Y p -f(X L )) 2

[0130] Step S2: In terms of production enterprise data, the future sales forecast E1 can be predicted for seasonal products using the triple exponential smoothing algorithm based on the correlation between historical sales data and events. Different weights are given to other products based on time factors such as season, usage demand, holidays, etc. in the model storage unit, and the future sales E2 is predicted using a bidirectional LSTM network.

[0131] Specifically, the triple exponential smoothing method is used to predict seasonal historical sales data within multiple time nodes. The specific recursive prediction formula is:

[0132] S i =α(x i -P i-k )+(1-α)(s i-1 +T i-1 )

[0133] T i =β(s i -s i-1 )+(1-β)T i-1

[0134] P i =γ(x i -s i )+(1-γ)p i-k

[0135] x i+h =S i +hTi+p i-k+(h mod k)

[0136] Among them, k is the period; h is the prediction period step; S i , S i-1 is the specific historical sales value corresponding to time nodes i and i-1; α, β, γ are smoothing exponents, which are between [0, 1] and retain one decimal place; T i is the smoothed trend corresponding to time node i; P i The seasonal information is retained, which is the seasonal trend after smoothing at the corresponding time node i; the initial value of the corresponding value is usually taken as S0=X0, T0=X1-X0, P0=0, and the smoothing exponents α, β, and γ are usually taken as (0.1, 0.9). In addition, according to the sales situation of the manufacturing enterprise, the abnormal values ​​of the sales data are marked during holidays and nearby periods.

[0137] After setting the initial value and iterating repeatedly, the future sales forecast E1 is obtained.

[0138] The method of building a two-way long-term and short-term network model is adopted to make future sales forecasts for general commodities.

[0139] The bidirectional LSTM network model is similar to the ordinary LSTM network described above. The difference is that the bidirectional LSTM network is a recurrent neural network that calculates the hidden vector from front to back. Another recurrent neural network calculates the latent vector from back to front Combine the following formula to get the future sales E2:

[0140]

[0141] In the actual application of the system, the bidirectional LSTM network includes an input layer, a multi-layer bidirectional LSTM part, a multi-layer fully connected layer part, and an output layer. Usually, there is no need to set a Dropout layer to meet the system prediction requirements. In system construction, the various network parameters are usually determined based on the particle swarm optimization algorithm, or the optimal parameters are determined through active attempts.

[0142] Step S3: Output the result E3 to the data presentation terminal, and display different contents according to different authorizations.

[0143] Information feedback module: According to the structure described above, the prediction model is in a state of dynamic adjustment and continuous updating. Therefore, according to the real data in the external real information (such as the data updated by the dealer in the data acquisition module or the latest user data obtained by our label), the difference between the two is recorded and transmitted back to the sales prediction module, and the weights of different prediction influencing factors are continuously adjusted to further train and optimize the sales prediction model. It is necessary to store the better prediction model obtained after repeated iterations back into the model storage unit to save prediction time for the next call of the model. It is difficult to fit the prediction value, and the result with a large difference between the prediction value and the true value is marked. In this module, it can be analyzed whether there are other influencing factors; at the same time, in this step, subjective judgment can be made and the influencing factors that are not meaningful to the model prediction results can be eliminated, so that the model continues to be accurate and the prediction results fit the true value appropriately.

[0144] In addition, to match the above system, we propose a sales forecast terminal device based on the idea of ​​two-way forecasting, which has the following features: it can match the data collection and upload module described in the above system, check the number and time of the authenticity of the goods, can be opened together with the opening of the goods, obtain the time when the user opens the goods, can bind the user's identity ID through the user's mobile device, obtain the user's region, device information, and re-purchase information. Exemplarily, the NFC / RFID circuit anti-counterfeiting label has the above functional characteristics.

[0145] Finally, it should be noted that the above examples are only specific implementation examples of the present invention. Of course, those skilled in the art can make changes and modifications to the present invention. If these modifications and variations fall within the scope of the claims of the present invention and their equivalent technologies, they should be considered to be within the scope of protection of the present invention.

Claims

1. An anti-counterfeiting smart label, characterized in that: The invention comprises a circular first flexible material layer (1), a traction layer (2) and a double-sided adhesive layer (3) which are arranged in sequence from top to bottom, and also comprises an NFC / RFID electronic smart tag (4), wherein the NFC / RFID electronic smart tag (4) is provided with a chip (41), the chip (41) is connected to an antenna circuit (42), the chip (41) is placed on the upper surface of the first flexible material layer (1), the antenna circuit (42) is fixedly arranged on the upper surface of the first flexible material layer (1), and the first flexible material layer (1) has a mechanism on one side close to the chip (41). The traction layer (2) comprises a semicircular second flexible material layer (21), the second flexible material layer (21) is connected to a semicircular tear-type safety ring (22), the second flexible material layer (21) is fixedly connected to the first flexible material layer (1) and the double-sided adhesive layer (3), the tear-type safety ring (22) is connected to the chip (41) via a traction rope (23) passing through the mechanical slit (43), and the binding force between the traction rope (23), the tear-type safety ring (22) and the chip (41) is greater than the binding force between the chip (41) and the antenna circuit (42); One end of the tear-off safety ring (22) is fixedly connected to a corner of the second flexible material layer (21), and the other end is provided with a pull ring or a pull handle (22a); A plurality of gear teeth are arranged on the inner side of the tear-off type safety ring (22), and one end of the traction rope (23) is fixed on a key gear tooth (22b) in the middle.

2. The anti-counterfeiting smart label according to claim 1, characterized in that: The center line of the double-sided adhesive layer (3) is a pressed die line (3a), and the lower surface of the double-sided adhesive layer (3) is divided into a layer with stronger adhesion (3b) and a layer with weaker adhesion (3c) by the die line (3a); the second flexible material layer (21) contacts and is fixedly connected to the upper surface of the double-sided adhesive layer (3) corresponding to the layer with weaker adhesion (3c); and the tear-off safety ring (22) contacts the upper surface of the double-sided adhesive layer (3) corresponding to the layer with stronger adhesion (3b).

3. The anti-counterfeiting smart label according to claim 1, characterized in that: A circular gasket layer (5) is adhered to the first flexible material layer (1), and a circular third flexible material layer (6) is adhered to the gasket layer (5); the third flexible material layer (6), the gasket layer (5) and the first flexible material layer (1) together form a receiving space (8) for receiving the NFC / RFID electronic smart tag (4).

4. The anti-counterfeiting smart label according to claim 1, characterized in that: The antenna circuit (42) is made of liquid metal or is a printed metal antenna circuit; a release paper layer (7) is provided below the double-sided adhesive layer (3).

5. A two-way predictive sales system comprising the anti-counterfeiting smart label according to claim 1, characterized in that: It includes data collection and upload module, data classification module, data initialization module, sales forecast module, information feedback module and data presentation terminal; The data collection and uploading module includes a collection unit, a sample desensitization unit and a data uploading unit, which are arranged in the user device, the dealer device and the manufacturer device, and collects data by scanning the anti-counterfeiting smart label of claim 1 through the user device, and uploads the data after desensitization processing of the data information is completed in the user device; The data classification module is set in the server, and performs preliminary labeling processing on the information uploaded from the data collection and upload module; The data initialization module initializes the basic data based on the attributes of the enterprise's products, so that the prediction framework does not deviate from the basic attributes of the products; The sales forecasting module includes a model storage unit and a data forecasting unit. When the user makes a purchase again and the data is accumulated to a certain amount, the sales forecasting module uses the sales forecasting algorithm based on the data accumulated in the data classification module to make a specific analysis and forecast of the future sales of the goods sold by the enterprise. The information feedback module collects specific data after making the initial forecast and feeds it back to the sales forecast module to provide support for adjusting the forecast model to improve accuracy; The data presentation terminal displays and outputs sales forecast results to people, and includes the functions of displaying different information to different groups and collecting different information to different groups.

6. A control method for a two-way predictive sales system, characterized in that: The data collection and upload module collection method is as follows: Collection unit: on the one hand, it is used to obtain the user ID and a series of behavioral data associated with the ID; including: the time and place when the user uses the consumer mobile device to check the authenticity after purchase, and the time and place when the product is opened; the number of times the user has purchased the product in the past and related information, the device information used by the user to check the authenticity, and the product information opened by the user; on the other hand, it can be used to obtain the company's historical sales data, the warehouse supply of related products, and product production information. The product production information includes the production date, production location, and production batch information; in addition, a method for obtaining information from dealers is also designed. Dealers can package and upload historical sales data, and the missing parts of the historical sales data can be supplemented through user-side data collection; Sample desensitization unit: desensitizes the information obtained by the collection unit locally on the user device, including converting the IP address of the user device into address information within a larger area and converting the physical address of the user device into a virtual ID used only for identification features; Data upload unit: encrypts and stores the desensitized data, and then encrypts and uploads it to the server after the user device is connected to the Internet; The data classification module is used to perform preliminary processing on the data uploaded by the acquisition unit, label the information data, and store them in categories. This includes: analyzing the user's active time period and active time interval based on the time and number of times the user's device verifies the authenticity of the product, calculating the repurchase ratio based on the user's repurchase time information, and dividing different regions based on the user's opening address and the solar term at the time to avoid the impact of different climates on subsequent predictions; and marking and processing abnormal sales data during holidays and nearby periods based on the company's sales situation. The sales forecast module uses the data processed by the data classification module and the data initialization module to make sales forecasts. The sales forecast module prediction method is as follows: Model storage unit: connected with the data initialization module and the information feedback module, determines the weights of various influencing factors of the initial forecast model according to the basic attributes of the product, including seasonality and regionality, and stores the forecast model; Data prediction unit: In terms of users, the labeled data and the behavioral information of users with similar labels after matching are used, combined with the weights of season, usage demand, and holiday time factors in the model storage unit, to preliminarily predict the stickiness of enterprise products and future sales A; the enterprise side can use big data methods to preliminarily predict future sales B based on the correlation between historical sales data and events; by referring to the two sales forecast values, the data is used to make sales forecasts based on the idea of ​​two-way forecasting, and the future sales C of the predicted product is obtained; The information feedback module records the difference between actual sales and future sales C, transmits it back to the sales forecasting module, continuously adjusts the weights of the forecasting influencing factors, further trains and optimizes the sales forecasting model, and stores the better forecasting model obtained after repeated iterations back into the model storage unit, saving forecasting time for the next call of the model; marks the forecast values ​​that are difficult to fit and the results with a large difference between the forecast value and the true value, and analyzes whether there are other influencing factors; and prunes the influencing factors that have little significance to the model prediction results, so that the prediction results properly fit the true values; In the data presentation terminal, on the one hand, manufacturers can authorize dealers to view local sales forecast results; on the other hand, users can register through the data presentation terminal; Three different interfaces are provided for three different groups: For registered users: Provide a community-like interactive interface where users can enter the community and directly share their opinions on a product, their daily life, and recommendations. They can also make corresponding star ratings on the product experience after verifying the authenticity and using it for a period of time. Provide visual analysis of data for dealers. On the one hand, the data can be used to predict future sales and give suggestions for the next purchase period under visual analysis. On the other hand, it can encourage dealers to retain and upload historical sales data. In addition, different permission levels are provided for dealers in different jurisdictions, allowing them to view different sales data visualization analysis results; For manufacturers, it can directly obtain the sales situation of all regions and the visual analysis results of sales data, including the ranking of sales of different dealers. At the same time, it can also make predictions about the overall future sales situation and give suggestions for the next distribution of goods. Additionally, dealers and manufacturers can register accounts in the user-side community. After being authenticated by the user side, dealers and manufacturers can directly obtain real-time feedback from users and, based on actual needs, extract the most frequently appearing keywords and provide feedback to dealers and manufacturers.

Images