Supply chain management system and method based on big data

Abstract

The invention discloses a supply chain management system and method based on big data, and belongs to the field of hotel supply chain management, and the system comprises an intelligent sensing module which is used for collecting multi-source public information and equipment operation data, and integrating and outputting structured data and a guest room state thermodynamic diagram; the decision optimization module is used for inputting multi-modal structured data, supplier parameters and inventory demand data, and outputting time-phased occupancy rate prediction through fusion of the structured data; generating a purchase negotiation strategy, and outputting a purchase plan; the execution control module is used for converting the purchase plan into a BPMN standardized process containing a distributed account book certificate; identifying an anomaly based on a dynamic threshold determination algorithm; and the iterative feedback module is used for collecting the purchase execution data and the exception handling data, and analyzing and generating an optimization report. The system can solve the industrial problems of demand lag, serious waste and the like of traditional hotel purchase.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of hotel supply chain management, and in particular to a supply chain management system and method based on big data. BACKGROUND

[0002] With the promotion of digital transformation of the hotel industry, material procurement management has been upgraded from manual experience judgment to data-driven prediction, but the existing technical solutions still have some shortcomings, which are difficult to meet the precise, real-time and compliant procurement needs of hotels. The existing hotel procurement prediction is mostly based on historical occupancy rate and material consumption data, ignoring front-end user demand data and not paying attention to real-time customer feedback, so the accuracy of material procurement demand prediction is limited by the timeliness of the data, and it is difficult to adapt to the small batch and high frequency procurement characteristics of hotels. Even if the front-end user demand data is concerned, it is only manually browsing OTA comments, and it is impossible to quantitatively analyze the text sentiment tendency and image material characteristics, so a large amount of unstructured data value is idle, and there is a lack of conversion path to convert unstructured data into quantitative demand indicators, resulting in a disconnection between customer feedback and procurement plans, and the inability to quickly respond to real-time customer demand for material quality and quantity. Some hotels use text analysis tools to extract material complaints from OTA comments, but they need to manually enter the analysis results into the procurement system; when logistics delays occur, they need to manually query alternative suppliers, which takes too long to respond; after monthly deviation analysis, they need to manually adjust the prediction model parameters, which is inefficient and prone to errors. Such a fragmented procurement process relies on manual connection at each link and cannot meet the real-time response and rapid adjustment needs of hotel procurement. SUMMARY

[0003] The present application aims to at least solve one of the technical problems existing in the prior art. To this end, the present application proposes a supply chain management system and method based on big data, which can solve the industry pain points of demand lag, serious waste and high compliance risk in traditional hotel procurement.

[0004] The embodiment of the present application provides a supply chain management system based on big data, which comprises an intelligent sensing module, a decision optimization module, an execution control module and an iterative feedback module. The intelligent sensing module is used for collecting text type multi-source public information, image type multi-source public information and equipment operation data, processing the public information through a layered desensitization algorithm, integrating the desensitized text type multi-source public information, the desensitized image type multi-source public information and the equipment operation data by using a multi-modal weighted fusion model, and outputting structured data containing emotional value, activity and energy consumption value characteristics and a room state heat map. The decision optimization module is used for inputting multi-modal structured data, supplier parameters and inventory demand data, realizing the fusion of industry knowledge and the structured data through a structured knowledge graph combined with a TF-IDF algorithm and a cosine similarity algorithm, introducing an attention mechanism to dynamically adjust data feature weights, and outputting a time period occupancy rate prediction. The multi-dimensional parameters include the supplier parameters and real-time market price fluctuation data. The execution control module is used for converting the procurement plan into a BPMN standardized process containing a distributed account evidence, realizing approval right through electronic signature, identifying abnormalities including logistics delay and inventory emergency based on a dynamic threshold determination algorithm, establishing a three-level response mechanism and generating an exception handling work order containing a time limit. The iterative feedback module is used for collecting procurement execution data and exception handling data, generating an optimization report through multi-dimensional deviation analysis, updating a knowledge graph and adjusting rule weights according to a threshold, and triggering a temporary optimization mechanism for a sudden scene.

[0005] The embodiment of the present application at least realizes the following beneficial effects: The OTA text, OTA image and equipment data are collected by the intelligent sensing module, the multi-modal weighted fusion model is used to generate quantitative indicators, the prediction data is expanded from historical static data to real-time dynamic data, the prediction lag pain point is solved, the knowledge graph, the attention mechanism and the C4.5 decision tree are used to generate the procurement plan, the execution control module realizes the procurement process automation, and the procurement process fragmentation pain point is solved. The OTA multi-modal data is introduced, the multi-modal data is accurately desensitized, dynamically fused and predicted, and the systematic technical combination of the whole-process closed loop is realized, the prediction lag, data idling, compliance conflict and process fragmentation defects of the prior art are solved, the hotel procurement is upgraded to real-time, accurate and compliant data-driven, and the technical gap between the front-end user demand and the back-end procurement decision in the hotel field is filled.

[0006] According to some embodiments of the present application, the text type multi-source public information comprises OTA platform scores, local activity public information and customer social media public comment texts; the image type multi-source public information comprises customer social media public comment pictures and OTA platform public room environment pictures; the equipment operation data comprises smart door lock switch frequency data, mini-bar weight change data and room energy consumption data; the sentiment value is extracted based on text comment semantics and image scene atmosphere, the activity level is calculated based on door lock switch frequency and room energy consumption, and the energy consumption value is derived from room energy consumption data; the intelligent sensing module comprises: an edge intelligent collection unit, configured to collect public information and perform layered desensitization; sensitive identifiers in the image type multi-source public information are identified by a YOLOv8 target detection algorithm, hotel LOGOs, room numbers and customer faces are identified, Gaussian filter blur processing is performed on customer faces, pixel erasure processing is performed on hotel LOGOs and room numbers, and gray pixel block covering processing is performed on license plate numbers that may be contained in the pictures; private information in the text type multi-source public information is shielded by character replacement.

[0007] According to some embodiments of the present application, the intelligent sensing module comprises: a multi-modal fusion unit, configured to perform a multi-modal weighted fusion model, and a calculation formula of the model is: ; wherein F is output data after fusion, n is the number of input data types, ωi is the weight of the i-th input data, and Di is the i-th input data; wherein the desensitized image type multi-source public information is extracted by CNN to obtain image feature data, the desensitized text type multi-source public information is converted into structured text data by natural language processing technology, and the equipment operation data is arranged in time sequence to obtain equipment time sequence data; each type of data participates in fusion according to a preset weight, and a three-dimensional feature matrix is finally output, which is structured data containing sentiment value, activity level and energy consumption value features.

[0008] According to some embodiments of the present application, the decision optimization module comprises: an industry knowledge enhanced prediction unit, configured to construct a structured knowledge graph, and business rule nodes of the knowledge graph comprise hotel linen replacement cycles, cleaning material consumption standards, consumable replenishment thresholds and room equipment maintenance specifications; multi-modal structured data is matched and associated with knowledge graph rules by a lightweight semantic retrieval algorithm; an attention mechanism is introduced to dynamically adjust the feature weight of each structured data, and a future time period occupancy rate prediction is output; and a calculation formula of the attention mechanism is: ; wherein, is the attention value of the j-th feature, are the weights of the j-th and k-th features respectively, Q is a feature importance query vector, and m is the total number of features.

[0009] According to some embodiments of the present application, the decision optimization module comprises: a category adaptive negotiation unit for inputting multi-dimensional parameters, processing supplier parameters in the multi-dimensional parameters, the supplier parameters including supplier rating, market benchmark price and procurement urgency; the supplier rating is divided into multiple levels according to historical fulfillment rate, different levels correspond to different cooperation priorities and procurement quantity allocation ratios; the market benchmark price sets a price interval according to the fluctuation range of the recent average price of the industry, and negotiation strategy adjustment is triggered when the price exceeds the interval; the procurement urgency is divided into multiple levels according to the demand time limit, different levels correspond to different negotiation response speeds and account period flexibility; different procurement categories are matched with exclusive negotiation strategies based on the C4.5 decision tree algorithm, and the generated supplier recommendation list contains supplier level, price interval and delivery cycle information.

[0010] According to some embodiments of the present application, the decision optimization module comprises: a dynamic rule engine unit for constructing a multi-level rule base, the multi-level rule base including a business priority level, a scenario adaptation level and an execution condition level; the business priority level is divided into three priorities of room operation materials, logistics support materials and administrative materials according to the use of materials, and the room operation materials have the highest priority; the scenario adaptation level contains three rule sets of peak season, off-season and emergency scenario, different scenarios correspond to different material demand coefficients; the execution condition level sets the safety stock threshold and the alert stock threshold of the real-time inventory level, and the arrival time warning threshold of the GPS trajectory data of the in-transit materials; the specific weight value of each priority material is calculated by the analytic hierarchy process, and the procurement plan generated in combination with the rule base contains the specific allocation ratio of batch procurement quantity and the selection basis of the supplier.

[0011] According to some embodiments of the present application, the execution control module comprises: an automatic procurement unit for converting the procurement plan into a BPMN standardized process containing distributed ledger evidence, the BPMN standardized process including a demand initiation link, a rule verification link, an approval link, an order placement link and a warehouse receipt link; the key node data including approver information, operation time information and operation content information is recorded through the distributed ledger.

[0012] According to some embodiments of the present application, the execution control module comprises: a scenario linkage abnormal response unit for identifying abnormalities based on a dynamic threshold determination algorithm, the abnormalities including logistics delay and inventory emergency; the dynamic threshold determination algorithm adjusts the abnormal determination threshold according to the supply chain operation scenario, the supply chain operation scenario including a peak season scenario and an off-season scenario, and the logistics delay secondary abnormal threshold in the peak season scenario is shorter than that in the off-season scenario, and the inventory emergency secondary abnormal threshold in the peak season scenario is higher than that in the off-season scenario, the threshold adjustment is automatically adapted according to historical scenario data; the calculation formula of the dynamic threshold determination algorithm includes: ; wherein, a logistics delay duration, an actual transportation time, a planned transportation time; an inventory emergency determination formula: ; wherein I is an inventory ratio, an actual inventory, a safety inventory; the scenario linkage abnormal response unit is further configured to establish a three-level response mechanism, a first-level abnormality corresponding to an automatic replenishment process, a second-level abnormality corresponding to an alternative supplier inquiry process, and a third-level abnormality corresponding to a cross-regional allocation process; and generate an abnormality processing work order with a time limit, the abnormality processing work order including responsibility department information and response time limit information.

[0013] According to some embodiments of the present application, the iterative feedback module includes: a deviation analysis unit configured to calculate multi-dimensional deviation rates, the multi-dimensional deviation rates including a prediction deviation rate, a procurement plan deviation rate, and an abnormal response deviation rate; the prediction deviation rate is obtained by dividing the absolute value of the difference between a predicted value and an actual value by the actual value; the procurement plan deviation rate is obtained by dividing the absolute value of the difference between a planned procurement quantity and an actual procurement quantity by the actual procurement quantity; and the abnormal response deviation rate is obtained by dividing the absolute value of the difference between a planned response time and an actual response time by the planned response time; each deviation rate is respectively provided with a corresponding threshold value, and when the deviation rate exceeds the corresponding threshold value, a knowledge base updating operation, a rule weight adjusting operation, and a response process upgrading operation are respectively performed; a temporary optimization unit configured to identify a sudden scenario, the sudden scenario including a large-scale conference scenario and a holiday scenario; trigger a temporary optimization mechanism, the temporary optimization mechanism including supplementing similar scenario historical data for prediction model training, shortening a procurement plan generation period, and compressing an abnormal response time limit; and after the scenario ends, the system automatically restores regular parameters, the regular parameters including procurement process cycle parameters and abnormal response time limit parameters.

[0014] In another aspect, the embodiment of the present application provides a supply chain management method based on big data, comprising the following steps: S100, intelligent sensing processing, collecting text type multi-source public information, image type multi-source public information and equipment operation data, processing the multi-source public information by using a hierarchical desensitization algorithm, and then integrating the desensitized text type multi-source public information, the desensitized image type multi-source public information and the equipment operation data by using a multi-modal weighted fusion model to output structured data containing emotional value, activity and energy consumption value characteristics and a room state thermal map; S200, decision optimization processing, inputting the multi-modal structured data output by the step S100, supplier parameters and inventory demand data, realizing the fusion of industry knowledge and the structured data through a structured knowledge graph combined with a lightweight semantic retrieval containing a TF-IDF algorithm and a cosine similarity algorithm, introducing an attention mechanism to dynamically adjust data feature weights, and outputting a time period occupancy rate prediction; generating a procurement negotiation strategy by analyzing multi-dimensional parameters through a C4.5 decision tree algorithm, wherein the multi-dimensional parameters include the supplier parameters and real-time market price fluctuation data; outputting a procurement plan according to a multi-level rule base combined with a business priority weight calculated by an analytic hierarchy process; S300, execution control processing, converting the procurement plan output by the step S200 into a BPMN standardized process containing a distributed account evidence, and realizing approval right through electronic signature; identifying abnormalities including logistics delay and inventory emergency based on a dynamic threshold determination algorithm; establishing a three-level response mechanism and generating an exception handling work order containing a time limit; S400, iterative feedback processing, collecting procurement execution data and exception handling data in the step S300, generating an optimization report through multi-dimensional deviation analysis; updating the structured knowledge graph based on the optimization report according to a preset threshold, and adjusting the rule weight of the multi-level rule base; and triggering a temporary optimization mechanism to adapt to the scene demand for a sudden scene in the supply chain operation.

[0015] The embodiment of the present application at least realizes the following beneficial effects: the OTA text, OTA image and equipment data are collected by the intelligent sensing module, the multi-modal weighted fusion model is used to generate quantitative indicators, the prediction data is expanded from historical static data to real-time dynamic data, the prediction lag pain point is solved, the procurement plan is generated by the knowledge graph, the attention mechanism and the C4.5 decision tree, the procurement process automation is realized by the execution control module, and the procurement process fragmentation pain point is solved. The embodiment of the present application introduces OTA multi-modal data, accurately desensitizes the multi-modal data, dynamically fuses the prediction and realizes the systematic technical combination of the whole-process closed loop, specifically solves the prediction lag, data idling, compliance conflict and process fragmentation of the prior art, realizes the real-time, accurate and compliant data-driven of hotel procurement, and fills the technical gap of front-end user demand and back-end procurement decision in the hotel field.

[0016] Additional aspects and advantages of the application will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following and the attendant drawings or can be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0017] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, including the accompanying drawings, wherein:

[0018] Figure 1 Structure diagram of a system according to an embodiment of the present application.

[0019] Figure 2 Flow diagram of a method according to an embodiment of the present application.

[0020] REFERENCE NUMERALS

[0021] Intelligent sensing module 100, decision optimization module 200, execution control module 300, iterative feedback module 400. DETAILED DESCRIPTION

[0022] Embodiments of the present application are described in detail below with reference to the attached drawings, wherein the same or similar components are denoted by the same or similar reference numerals, and thus repeated description is omitted. The embodiments described below are merely exemplary, and are used only for the purpose of explaining the present application, and should not be construed as limiting the present application.

[0023] In the description of the present application, the meaning of one or more is one or more, the meaning of multiple is two or more, greater than, less than, more than, etc. are understood as not including the number, above, below, within, etc. are understood as including the number. If it is described as first, second, it is only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or implicitly indicating the order of indicated technical features.

[0024] Referring to Figure 1 An embodiment of the present application proposes a supply chain management system based on big data, comprising:

[0025] The intelligent perception module 100 is configured to collect text-based multi-source public information, image-based multi-source public information and equipment operation data, the text-based multi-source public information includes OTA platform scores, local activity public information and customer social media public comment texts, the image-based multi-source public information includes customer social media public comment pictures and OTA platform public room environment pictures, and the equipment operation data includes smart door lock switching frequency data, mini-bar weight change data and room energy consumption data; the public information is processed through a layered desensitization algorithm; a multi-modal weighted fusion model is used to integrate the desensitized text-based multi-source public information, the desensitized image-based multi-source public information and the equipment operation data, and output structured data containing sentiment values, activity levels and energy consumption values and a room state heat map; the sentiment values are extracted based on text comment semantics and image scene atmosphere, the activity levels are calculated based on door lock switching frequencies and room energy consumption, and the energy consumption values are derived from room energy consumption data.

[0026] The decision optimization module 200 is configured to input multi-modal structured data, supplier parameters and inventory demand data, realize the fusion of industry knowledge and structured data through a structured knowledge graph combined with a TF-IDF algorithm and a cosine similarity algorithm, introduce an attention mechanism to dynamically adjust data feature weights, and output a time-period-based occupancy rate prediction; generate a procurement negotiation strategy by analyzing multi-dimensional parameters through a C4.5 decision tree algorithm, the multi-dimensional parameters include supplier parameters and real-time market price fluctuation data; output a procurement plan according to a multi-level rule base combined with business priority weights calculated by an analytic hierarchy process.

[0027] The execution control module 300 is configured to convert the procurement plan into a BPMN standardized process containing distributed ledger evidence, realize approval right through electronic signature; identify abnormalities including logistics delays and inventory emergencies based on a dynamic threshold determination algorithm; establish a three-level response mechanism and generate an exception handling work order containing time limits.

[0028] The iterative feedback module 400 is configured to collect procurement execution data and exception handling data, generate an optimization report through multi-dimensional deviation analysis, update a knowledge graph and adjust rule weights according to thresholds, and trigger a temporary optimization mechanism for sudden scenarios.

[0029] In one specific embodiment, a three-star city hotel with 150 guest rooms needs to predict the procurement quantity of guest room consumables (such as towels, toiletries, and mini-bar snacks) 7 days before the May Day holiday to avoid shortages or waste. The hotel can collect the text of the past 30 days of ratings on the OTA platform "Qunar" and "Meituan" through the system, extract keywords such as "towel thin" and "toiletries insufficient"; collect the local tourism bureau's public "May City Exhibition Information" text on the public information platform; collect customer-published room environment review images through customer social media "Xiao Hong Shu" and "Douyin", some of which contain "mini-bar empty bottle" and "towel stains"; at the same time, collect public room photos on the OTA platform; and collect device data including smart lock switch frequency (1.8 times per room per day), mini-bar weight change (0.5 kg per room per day), and room energy consumption (12 degrees per room per day). The system desensitizes the collected public information in layers, blurs customer faces in review images using Gaussian filtering, pixel erases room numbers (such as "302") after YOLOv8 recognition, covers parking lot license plates with gray pixel blocks, and replaces customer nicknames in text with "customer XXX". Then, through a multi-modal weighted fusion model (weights: text 0.3, image 0.2, device data 0.5), the data is integrated, and structured data (sentiment value: -0.2, reflecting "partial negative reviews"; activity level: 0.8, reflecting "high occupancy expectations"; energy consumption value: 12 degrees per room) and room status heat map (1-3 floor room heat value 85%, 4-5 floor 70%) are output. According to the output data, the hotel predicts that the occupancy rate during the May Day holiday will reach 92%, the accuracy of consumable procurement has increased by 35% compared to the same period last year, and toiletry waste has been reduced by 18%.

[0030] In another specific embodiment, a business hotel with 200 guest rooms receives a notice of a city IT exhibition, which is expected to bring 150 room orders per day, and the procurement plan for linen, toiletries, and conference refreshment materials needs to be generated within 5 days. The system imports rules such as hotel linen replacement standards (once every 3 days) and conference refreshment per capita consumption (2 bottles of mineral water and 3 snacks per person) through the structured knowledge graph; through the TF-IDF algorithm, multi-modal structured data is extracted, including the text features of the exhibition scale of 500 people and the historical same period exhibition occupancy rate of 95%, combined with the cosine similarity matching of the knowledge graph in the large conference material demand rules. Through the attention mechanism, the feature weights of the exhibition scale and the historical exhibition occupancy rate are dynamically enhanced, and the output is the time period occupancy rate prediction: 80% one day before the exhibition, 98% during the exhibition, and 60% after the exhibition. Through the C4.5 decision tree input multi-dimensional parameters, including the supplier A's performance rate of 98%, the market benchmark price of 12 yuan per set of toiletries, and the procurement urgency of high, the negotiation strategy is generated: the account period is shortened to 15 days, and a 5% discount is sought, and the procurement plan is generated: linen is additionally purchased 50 sets, toiletries are 300 sets, and conference refreshment materials are 1500. The procurement plan meets the material demand during the exhibition without out-of-stock, and the procurement cost is reduced by 8% through negotiation.

[0031] In some embodiments, the intelligent perception module 100 includes an edge intelligent collection unit for collecting public information and performing hierarchical desensitization; for sensitive identifiers in image-based multi-source public information, the hotel LOGO, room number, and customer face are identified through the YOLOv8 target detection algorithm, the customer face is subjected to Gaussian filter blur processing, the hotel LOGO and room number are subjected to pixel erasure processing, and the license plate number that may be contained in the picture is subjected to gray pixel block covering processing; for private information in text-based multi-source public information, customer personal information is shielded through character replacement.

[0032] In some embodiments, the intelligent perception module 100 includes:

[0033] A multi-modal fusion unit for executing a multi-modal weighted fusion model, the calculation formula of the model being:

[0034] ;

[0035] Wherein F is the output data after fusion, n is the number of input data types, ωi is the weight of the i-th input data, and Di is the i-th input data; the desensitized image-based multi-source public information is extracted through CNN to obtain image feature data, the desensitized text-based multi-source public information is converted into structured text data through natural language processing technology, and the device operation data is arranged in time sequence as device time sequence data; each type of data participates in fusion according to a preset weight, and a three-dimensional feature matrix is finally output, which is structured data containing emotional value, activity level, and energy consumption value features.

[0036] In some embodiments, the decision optimization module 200 comprises an industry knowledge enhanced prediction unit for constructing a structured knowledge graph, the business rule nodes of the knowledge graph including hotel linen replacement cycle, cleaning material consumption standard, consumable replenishment threshold and guest room equipment maintenance specification; matching and associating multi-modal structured data with knowledge graph rules through a lightweight semantic retrieval algorithm; introducing an attention mechanism to dynamically adjust the feature weight of each structured data and output future sub-period occupancy rate prediction; the calculation formula of the attention mechanism is:

[0037] ;

[0038] In the formula, is the attention value of the jth feature, are the weights of the jth and kth features respectively, Q is a feature importance query vector, and m is the total number of features.

[0039] In some embodiments, the decision optimization module 200 comprises a category adaptive negotiation unit for inputting multi-dimensional parameters, processing supplier parameters in the multi-dimensional parameters, and the supplier parameters including supplier rating, market benchmark price and procurement urgency; the supplier rating is divided into multiple grades according to historical performance rate, and different grades correspond to different cooperation priorities and procurement quantity distribution proportions; the market benchmark price sets a price interval according to the fluctuation range of the recent average price in the industry, and when the interval is exceeded, negotiation strategy adjustment is triggered; the procurement urgency is divided into multiple levels according to the demand time limit, and different levels correspond to different negotiation response speeds and account period flexibility; based on the C4.5 decision tree algorithm, exclusive negotiation strategies are matched for different procurement categories, and the generated supplier recommendation list includes supplier grade, price interval and delivery cycle information.

[0040] In some embodiments, the decision optimization module 200 comprises a dynamic rule engine unit for constructing a multi-level rule base, the multi-level rule base including a business priority level, a scene adaptation level and an execution condition level; the business priority level is divided into three priorities of guest room operation materials, logistics support materials and administrative materials according to the purpose of the materials, and the priority of the guest room operation materials is the highest; the scene adaptation level includes three rule sets of peak season, off-season and sudden scene, and different scenes correspond to different material demand coefficients; the execution condition layer sets the safety stock threshold and the alarm stock threshold of the real-time inventory level, and the arrival time warning threshold of the in-transit material GPS trajectory data; the specific weight value of each priority material is calculated by the analytic hierarchy process, and the procurement plan generated in combination with the rule base includes the specific allocation proportion of the batch procurement quantity and the selection basis of the supplier.

[0041] In some embodiments, the execution control module 300 comprises an automatic purchase unit for converting a purchase plan into a BPMN standardized process containing distributed ledger evidence, the BPMN standardized process comprising a demand initiation link, a rule verification link, an approval link, an order placement link, and an inspection and warehousing link; and recording process key node data in a distributed ledger, the key node data comprising approver information, operation time information, and operation content information.

[0042] In some embodiments, the execution control module 300 comprises a scenario linkage exception response unit for identifying exceptions based on a dynamic threshold determination algorithm, the exceptions including logistics delays and inventory emergencies; the dynamic threshold determination algorithm adjusts exception determination thresholds according to supply chain operation scenarios, the supply chain operation scenarios including peak season scenarios and off-season scenarios, and the second-level exception threshold for logistics delays in the peak season scenario is shorter than that in the off-season scenario, and the second-level exception threshold for inventory emergencies in the peak season scenario is higher than that in the off-season scenario, the threshold adjustment being automatically adapted according to historical scenario data; and the calculation formula of the dynamic threshold determination algorithm comprises:

[0043] Logistics delay determination formula: ;

[0044] wherein, is the logistics delay duration, is the actual transportation time, is the planned transportation time;

[0045] Inventory emergency determination formula: ;

[0046] wherein, I is the inventory ratio, is the actual inventory, is the safety inventory;

[0047] The scenario linkage exception response unit is further configured to establish a three-level response mechanism, a first-level exception corresponding to an automatic replenishment process, a second-level exception corresponding to an alternative supplier inquiry process, and a third-level exception corresponding to a cross-regional allocation process; and generate an exception handling work order containing a time limit, the exception handling work order comprising responsibility department information and response time limit information.

[0048] In some embodiments, the iterative feedback module 400 comprises a deviation analysis unit for calculating multi-dimensional deviation rates, the multi-dimensional deviation rates comprising a prediction deviation rate, a purchase plan deviation rate, and an exception response deviation rate; the prediction deviation rate being obtained by dividing the absolute value of the difference between a predicted value and an actual value by the actual value; the purchase plan deviation rate being obtained by dividing the absolute value of the difference between a planned purchase quantity and an actual purchase quantity by the actual purchase quantity; and the exception response deviation rate being obtained by dividing the absolute value of the difference between a planned response time and an actual response time by the planned response time; each deviation rate being provided with a corresponding threshold value, and when the deviation rate exceeds the corresponding threshold value, a knowledge base updating operation, a rule weight adjusting operation, and a response process upgrading operation are respectively performed.

[0049] In some embodiments, the iterative feedback module 400 comprises: a temporary optimization unit for identifying burst scenarios, the burst scenarios comprising large conference scenarios and holiday scenarios; triggering a temporary optimization mechanism, the temporary optimization mechanism comprising supplementing the same kind of scenario historical data for the prediction model training, shortening the procurement plan generation cycle and compressing the abnormal response time limit; after the end of the scenario, the system automatically restores the regular parameters, the regular parameters comprising the procurement process cycle parameters and the abnormal response time limit parameters.

[0050] Referring Figure 2 The embodiment of the present application proposes a supply chain management method based on big data, comprising the following steps:

[0051] S100, intelligent sensing processing, collecting text type multi-source public information, image type multi-source public information and equipment operation data, using a hierarchical desensitization algorithm to process the multi-source public information, and then using a multi-modal weighted fusion model to integrate the multi-source public information and the equipment operation data, outputting structured data containing sentiment value and activity degree characteristics and a room state heat map.

[0052] S200, decision optimization processing, inputting the multi-modal structured data output by step S100, supplier parameters and inventory demand data, fusing industry knowledge and input data through a structured knowledge graph combined with a lightweight semantic retrieval, outputting a time period occupancy rate prediction; generating a procurement negotiation strategy through algorithm analysis of multi-dimensional parameters; calculating according to a multi-level rule base and a business priority, and outputting a procurement plan.

[0053] S300, execution control processing, converting the procurement plan output by step S200 into a BPMN standardized process containing distributed ledger evidence, using electronic signatures to authenticate the approval links in the process; identifying abnormalities in the supply chain operation based on a dynamic threshold determination algorithm, the identified abnormalities including logistics delays and inventory emergencies; establishing a three-level response mechanism to dispose of the abnormalities and generating an abnormality handling work order containing a response time limit.

[0054] S400, iterative feedback processing, collecting procurement execution data and abnormality handling data in step S300, generating an optimization report through multi-dimensional deviation analysis; updating the structured knowledge graph based on the optimization report according to a preset threshold, and adjusting the rule weights of the multi-level rule base; for burst scenarios in the supply chain operation, triggering a temporary optimization mechanism to adapt to the scenario requirements.

[0055] While specific embodiments are described herein, one of ordinary skill in the art will appreciate that many other modifications or alternative embodiments are within the scope of the present disclosure. For example, any of the functions and / or process capabilities described in connection with a particular device or component can be performed by any other device or component. Additionally, while various illustrative implementations and architectures have been described in accordance with embodiments of the present disclosure, one of ordinary skill in the art will appreciate that many other modifications to the illustrative implementations and architectures described herein are within the scope of the present disclosure.

[0056] Certain aspects of the present disclosure are described above with reference to block and flow diagrams of systems, methods, apparatuses, and / or computer program products according to example embodiments. It will be understood that one or more blocks of the block diagrams and flow diagrams, and combinations of blocks in the block diagrams and flow diagrams, can be implemented by

[0057] Thus, the blocks in the block diagrams and flow diagrams support combinations of means for performing the specified functions, combinations of elements or steps for performing the specified functions, and program instruction means for performing the specified functions. It will also be understood that each block of the block diagrams and flow diagrams, and combinations of blocks in the block diagrams and flow diagrams, can be implemented by dedicated hardware-based computer systems which perform the specified functions, or combinations of dedicated hardware and computer instructions.

[0058] The program modules, applications, and the like described herein can include one or more software components, including, for example, software objects, methods, data structures, and the like. Each such software component can include computer-executable instructions that, in response to execution, cause at least a portion of the functionality described herein (e.g., one or more operations of the illustrative methods described herein) to be performed.

[0059] Software components can be coded in any of a variety of programming languages. One illustrative programming language can be a low-level programming language such as an assembly language associated with a specific hardware architecture and / or operating system platform. Software components comprising assembly language instructions can need to be translated via an assembler before execution by the hardware architecture and / or platform. Another illustrative programming language can be a higher-level programming language that can be portable to multiple architectures. Software components comprising a higher-level programming language can need to be translated to an intermediate representation before execution via an interpreter or compiler. Other examples of programming languages include, but are not limited to, a macro language, a shell or command language, a job control language, a script language, a database query or search language, or a report writing language. In one or more illustrative embodiments, a software component containing instructions in one of the above examples of programming languages can be executed directly by an operating system or other software component without first being converted to another form.

[0060] Software components can be stored as files or other data storage constructs. Software components of a similar type or related function can be stored together such as in a particular directory, folder, or library. Software components can be static (e.g., preset or fixed) or dynamic (e.g., created or modified at execution time).

[0061] The above embodiments of the present application have been described in detail but not limited thereto or thereby. Various modifications can be made to these embodiments without departing from the spirit of the application.

Claims

1. A supply chain management system based on big data, characterized in that, include: The intelligent sensing module is used to collect multi-source public information in the form of text, multi-source public information in the form of images, and equipment operation data; Public information is processed using a layered desensitization algorithm; A multimodal weighted fusion model is used to integrate desensitized text-based multi-source public information, desensitized image-based multi-source public information, and equipment operation data, and output structured data and guest room status heatmaps containing sentiment values, activity levels, and energy consumption values. The decision optimization module takes multimodal structured data, supplier parameters, and inventory demand data as input. It integrates industry knowledge with the structured data using a structured knowledge graph combined with lightweight semantic retrieval incorporating TF-IDF and cosine similarity algorithms. An attention mechanism is introduced to dynamically adjust data feature weights, outputting time-segmented occupancy rate predictions. A C4.5 decision tree algorithm is used to analyze multi-dimensional parameters, including supplier parameters and real-time market price fluctuation data, to generate procurement negotiation strategies. Based on a multi-level rule base and business priority weights calculated using the analytic hierarchy process, a procurement plan is output. The execution control module is used to transform the procurement plan into a BPMN standardized process with distributed ledger notarization, and to realize approval and authorization through electronic signatures; it identifies anomalies based on a dynamic threshold judgment algorithm, including logistics delays and inventory shortages. Establish a three-level response mechanism and generate time-limited exception handling work orders; The iterative feedback module is used to collect procurement execution data and anomaly handling data, generate optimization reports through multi-dimensional deviation analysis, update the knowledge graph and adjust rule weights based on thresholds, and trigger temporary optimization mechanisms for unexpected scenarios.

2. The supply chain management system based on big data according to claim 1, characterized in that, The text-based multi-source public information includes OTA platform ratings, public information about local events, and public comments from customers on social media; the image-based multi-source public information includes public comments from customers on social media and public images of the guest room environment from OTA platforms; the device operation data includes smart door lock opening and closing frequency data, minibar weight change data, and guest room energy consumption data. The sentiment value is extracted based on the semantics of the text comments and the atmosphere of the image scene; the activity level is calculated based on the frequency of door lock opening and closing and the energy consumption of the guest room; and the energy consumption value is derived from the energy consumption data of the guest room. The intelligent sensing module includes: The edge intelligent acquisition unit is used to collect publicly available information and perform layered desensitization. This includes identifying sensitive identifiers in multi-source publicly available image information using the YOLOv8 object detection algorithm to recognize hotel logos, room numbers, and customer faces; performing Gaussian filtering blurring on customer faces; performing pixel erasure on hotel logos and room numbers; and performing gray pixel block masking on license plates that may be present in the image. For privacy information in multi-source publicly available text information, customer personal information is masked through character replacement.

3. The supply chain management system based on big data according to claim 1, characterized in that, The intelligent sensing module includes: A multimodal fusion unit is used to execute a multimodal weighted fusion model, the calculation formula of which is: ； Where F represents the fused output data, and n represents the number of input data types. Let Di be the weight of the i-th type of input data; wherein the desensitized image-type multi-source public information is processed by CNN to extract image feature data, the desensitized text-type multi-source public information is transformed into structured text data through natural language processing technology, and the device operation data is organized into device time series data according to time series; each type of data participates in fusion according to preset weights, and the final output is a three-dimensional feature matrix, wherein the three-dimensional feature matrix is ​​the structured data containing sentiment value, activity level and energy consumption value features.

4. A supply chain management system based on big data according to claim 1, characterized in that, The decision optimization module includes: An industry knowledge-enhanced prediction unit is used to construct a structured knowledge graph. The business rule nodes of this knowledge graph include hotel linen replacement cycles, cleaning supply consumption standards, consumable replenishment thresholds, and guest room equipment maintenance specifications. A lightweight semantic retrieval algorithm is used to match and associate multimodal structured data with the knowledge graph rules. An attention mechanism is introduced to dynamically adjust the weights of each structured data feature, outputting future occupancy rate predictions for different time periods. The calculation formula for the attention mechanism is as follows: ； In the formula, Let j be the attention value for the j-th feature. where are the weights of the j-th and k-th features respectively, Q is the feature importance query vector, and m is the total number of features.

5. A supply chain management system based on big data according to claim 1, characterized in that, The decision optimization module includes: The category-adaptive negotiation unit is used to input multi-dimensional parameters and process supplier parameters, including supplier rating, market benchmark price, and procurement urgency. Supplier ratings are divided into multiple levels based on historical fulfillment rates, with different levels corresponding to different cooperation priorities and procurement volume allocation ratios. The market benchmark price is set within a price range based on the recent average price fluctuations in the industry; exceeding this range triggers adjustments to the negotiation strategy. The procurement urgency is divided into multiple levels based on demand timeframes, with different levels corresponding to different negotiation response speeds and payment term flexibility. Based on the C4.5 decision tree algorithm, a dedicated negotiation strategy is matched for different procurement categories, and the generated supplier recommendation list includes supplier rating, price range, and delivery cycle information.

6. A supply chain management system based on big data according to claim 1, characterized in that, The decision optimization module includes: The dynamic rule engine unit is used to construct a multi-level rule base, which includes a business priority layer, a scenario adaptation layer, and an execution condition layer. The business priority layer is divided into three priorities based on the use of materials: guest room operation materials, logistics support materials, and administrative materials, with guest room operation materials having the highest priority. The scenario adaptation layer contains three sets of rules for peak season, off-season, and emergency scenarios, with different material demand coefficients corresponding to different scenarios. The execution condition layer sets safety stock thresholds and warning stock thresholds for real-time inventory levels, as well as arrival time warning thresholds based on GPS trajectory data of materials in transit. The analytic hierarchy process (AHP) is used to calculate the specific weight values ​​of each priority material, and the procurement plan generated by combining the rule base includes the specific allocation ratio of batch procurement quantities and the selection criteria for suppliers.

7. A supply chain management system based on big data according to claim 1, characterized in that, The execution control module includes: An automated procurement unit is used to transform procurement plans into a BPMN standardized process with distributed ledger storage. The BPMN standardized process includes a demand initiation stage, a rule verification stage, an approval stage, an order placement stage, and an acceptance and warehousing stage. The distributed ledger records key node data of the process, including approver information, operation time information, and operation content information.

8. A supply chain management system based on big data according to claim 1, characterized in that, The execution control module includes: A scenario-linked anomaly response unit is used to identify anomalies based on a dynamic threshold determination algorithm. This algorithm adjusts the anomaly determination threshold according to the supply chain operation scenario, which includes peak and off-peak seasons. In peak seasons, the threshold for secondary anomalies related to logistics delays is lower than in off-peak seasons, while the threshold for secondary anomalies related to inventory shortages is higher. The threshold adjustment is automatically adapted based on historical scenario data. The calculation formula for the dynamic threshold determination algorithm includes: Formula for determining logistics delays: ; in, Due to logistics delays, This refers to the actual transportation time. For planned transportation time; Formula for determining low inventory: ; Where I is the inventory ratio. This is the actual inventory. For safety stock; The scenario-linked anomaly response unit is also used to establish a three-level response mechanism: level one anomalies correspond to an automatic replenishment process, level two anomalies correspond to a candidate supplier inquiry process, and level three anomalies correspond to a cross-regional transfer process; and to generate an anomaly handling work order with a time limit, wherein the anomaly handling work order includes information on the responsible department and the response time limit.

9. A supply chain management system based on big data according to claim 1, characterized in that, The iterative feedback module includes: The deviation analysis unit is used to calculate multi-dimensional deviation rates, including prediction deviation rate, procurement plan deviation rate, and abnormal response deviation rate. The prediction deviation rate is obtained by dividing the absolute value of the difference between the predicted value and the actual value by the actual value. The procurement plan deviation rate is obtained by dividing the absolute value of the difference between the planned procurement quantity and the actual procurement quantity by the actual procurement quantity. The abnormal response deviation rate is obtained by dividing the absolute value of the difference between the planned response time and the actual response time by the planned response time. Each deviation rate has a corresponding threshold. When the deviation rate exceeds the corresponding threshold, the knowledge base update operation, rule weight adjustment operation, and response process upgrade operation are executed respectively. A temporary optimization unit is used to identify sudden scenarios, including large-scale conference scenarios and holiday scenarios; trigger a temporary optimization mechanism, which includes supplementing historical data of similar scenarios for predictive model training, shortening the procurement plan generation cycle, and compressing the anomaly response time limit; after the scenario ends, the system automatically restores normal parameters, including procurement process cycle parameters and anomaly response time limit parameters.

10. A big data-based supply chain management method, used in the system as described in any one of claims 1 to 9, characterized in that, The method includes the following steps: S100 intelligent sensing and processing collects multi-source public information in text and image formats, as well as equipment operation data. It uses a hierarchical desensitization algorithm to process the multi-source public information, and then uses a multimodal weighted fusion model to integrate the desensitized multi-source public information in text and image formats, as well as the equipment operation data. It outputs structured data and a room status heatmap containing features such as sentiment value, activity level, and energy consumption value. S200, Decision Optimization Processing: Inputting the multimodal structured data, supplier parameters, and inventory demand data output from step S100, the system integrates industry knowledge with the structured data through a structured knowledge graph combined with lightweight semantic retrieval using TF-IDF and cosine similarity algorithms. An attention mechanism is introduced to dynamically adjust data feature weights, outputting time-segmented occupancy rate predictions. A procurement negotiation strategy is generated by analyzing multi-dimensional parameters using the C4.5 decision tree algorithm, including supplier parameters and real-time market price fluctuation data. Based on a multi-level rule base and business priority weights calculated using the analytic hierarchy process, a procurement plan is output. S300: Execute control processing, transform the procurement plan output from step S200 into a BPMN standardized process with distributed ledger notarization, and realize approval and authorization through electronic signature; identify anomalies based on dynamic threshold judgment algorithm, including logistics delays and inventory shortages; establish a three-level response mechanism and generate anomaly handling work orders with time limits; S400 Iterative feedback processing: Collect procurement execution data and anomaly handling data from step S300, generate an optimization report through multi-dimensional deviation analysis; update the structured knowledge graph based on the optimization report according to preset thresholds, and adjust the rule weights of the multi-level rule base; trigger a temporary optimization mechanism to adapt to the scenario requirements for sudden scenarios in the supply chain operation.

Images