A store digital operation management system based on cloud cooperation

Abstract

The application discloses a store digital operation management system based on cloud cooperation, comprising: an operation data acquisition module, which is used for collecting multi-source operation time series data and preprocessing to form a store standardized time series data set; a state representation generation module, which is used for inputting the store standardized time series data set into an improved TS2Vec model to generate a store operation state vector; a latent state energy modeling module, which is used for constructing a store operation latent state energy description structure to generate an operation phase transition driving force field; a strategy cooperation generation module, which is used for constructing a cross-strategy disturbance propagation tensor to generate a cloud cooperative operation strategy; a strategy issuing and executing module, which is used for forming store execution behavior time series data; and an execution feedback updating module, which is used for updating cloud cooperative operation strategy generation parameters. The application realizes closed-loop optimization of store operation state modeling, strategy generation and execution feedback by constructing a cloud cooperation and data driving mechanism.

Description

Technical Field

[0001] This invention relates to the field of cloud computing technology, and in particular to a cloud-based collaborative digital operation management system for stores. Background Technology

[0002] With the scaling up of chain retail, catering, and service stores, store operations management is gradually evolving towards digitalization and informatization. Existing store operations management systems typically collect and statistically analyze business data such as sales, inventory, personnel, and activities through information systems, and leverage cloud platforms for centralized data storage and unified display, thereby providing managers with business analysis and decision support. These systems have improved the accessibility and management efficiency of store data to a certain extent and have been widely used in actual business operations.

[0003] However, most existing technologies focus on static statistical analysis of store operation data or prediction of single indicators, typically centered on reports, rule engines, or simple models. Their characterization of store operation status remains discrete and fragmented, failing to reflect the inherent patterns of continuous evolution of store operation status over time. Cloud-based strategy generation relies heavily on human experience or preset rules, resulting in long strategy response cycles and difficulty in adapting to rapid changes in the store operating environment. Furthermore, in scenarios with multiple stores and multiple strategies operating in parallel, there is a lack of systematic modeling methods for the interrelationships between strategies.

[0004] Existing store operation management systems generally suffer from a disconnect between cloud-based strategies and actual store execution. After cloud-generated operational strategies are deployed to stores, they are often evaluated only after the fact based on outcome metrics, lacking continuous perception and feedback modeling of the execution process. This results in the strategy generation model being unable to dynamically adjust based on the actual store execution, thus affecting cross-store collaboration and strategy consistency.

[0005] Therefore, how to provide a cloud-based collaborative digital operation management system for stores is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a cloud-based collaborative digital operation management system for stores. This invention achieves continuous characterization and intelligent collaborative management of store operational status by introducing multi-source operational time-series data modeling, an improved time-series representation learning method, and a closed-loop mechanism for cloud-based collaborative strategy generation and execution feedback. This invention utilizes cloud computing technology and data-driven time-series modeling methods to deeply characterize and model potential states of store operational data, constructing an operational phase-change driving force field. Based on this, it realizes multi-strategy collaborative reasoning and strategy generation, combining actual store execution behavior for dynamic feedback and parameter updates, effectively improving the cloud-based strategy response speed and cross-store collaborative effect. This invention solves the problems of inaccurate characterization of store operational status, lack of targeted strategy generation, and insufficient execution feedback in existing technologies, possessing advantages such as high intelligence in strategy generation, strong collaborative consistency, and stable operational optimization effects.

[0007] A cloud-based collaborative digital operation management system for stores according to an embodiment of the present invention includes: The operational data acquisition module is used to collect multi-source operational time-series data generated during store operations and preprocess it to form a standardized time-series dataset for the store. The state representation generation module is used to upload standardized time-series datasets of stores to the cloud collaboration platform and input them into the improved TS2Vec model to generate store operation state vectors. The latent energy modeling module is used to construct the energy description structure of the potential state of store operation based on the store operation state vector, and generate the driving force field of operation phase transition. The strategy collaboration generation module is used to construct cross-strategy perturbation propagation tensors based on the store potential state parameters output by the operational phase change driving force field, and generate cloud-based collaborative operation strategies. The strategy delivery and execution module is used to deliver cloud-based collaborative operation strategies to the corresponding store execution terminals, collect actual execution behavior data, and form time-series data of store execution behavior. The execution feedback update module is used to build a two-domain association mapping mechanism between cloud collaborative operation strategies and store execution behaviors based on the time-series data of store execution behaviors, and to update the parameters generated by the cloud collaborative operation strategies.

[0008] Optionally, modules can be integrated using the following methods: Collect multi-source operational time-series data generated during store operations at each store, preprocess the multi-source operational time-series data, and form a standardized time-series dataset for each store. Upload the standardized time-series dataset of stores to the cloud collaboration platform and input it into the improved TS2Vec model. Perform multi-scale time pruning and hierarchical comparison learning on the standardized time-series dataset of stores to generate store operation status vectors. Based on the store operation state vector, a potential state energy description structure for store operation is constructed, and an operation phase transition driving force field is generated according to the continuous evolution relationship of the potential state of store operation over time. Based on the store potential state parameters output by the phase transition driving force field of operation, a cross-strategy perturbation propagation tensor is constructed in the cloud to model the relationship between store operation strategy, store potential state and the collaborative relationship between stores. Based on the cross-strategy perturbation propagation tensor, a cloud-based collaborative operation strategy is generated. The cloud-based collaborative operation strategy is distributed to the corresponding store execution terminal, and the actual execution behavior data of the store in response to the cloud-based collaborative operation strategy is collected to form time-series data of store execution behavior. Based on the time-series data of store execution behavior, a dual-domain association mapping mechanism is constructed in the cloud to represent the correspondence between cloud collaborative operation strategies and store execution behavior. By constraining the distribution consistency and correcting the association relationship between the characteristics of cloud collaborative operation strategies and the characteristics of store execution behavior, the generation parameters of cloud collaborative operation strategies are updated.

[0009] Optionally, the multi-source operational time-series data includes sales data, inventory data, customer flow data, staff scheduling data, equipment operation data, and activity execution data.

[0010] Optionally, the preprocessing of multi-source operational time-series data includes timestamp alignment, abnormal data removal, and unified field encoding of the multi-source operational time-series data.

[0011] Optionally, generating the store operation status vector includes: The standardized time series dataset of stores is divided into windows according to a preset time window length to obtain store window time series samples that correspond one-to-one with each time window. The store identifier and the start and end timestamps of the window are recorded for each store window time series sample. Multi-scale time clipping is performed on the time series sample of each store window. Within the time series sample of the same store window, multiple subsequences are extracted according to the preset clipping scale set. The clipping start time step index, clipping length and corresponding scale identifier are recorded for each subsequence. Multiple subsequences are input into the improved TS2Vec model, which includes a multi-source channel grouping coding unit, a trend-fluctuation decoupling unit, a hierarchical time series coding unit, and a store collaboration prototype memory unit, wherein: The multi-source channel grouping coding unit groups the feature channels according to the feature sources of the standardized time-series dataset of the stores and performs temporal convolutional coding on each group separately; The trend-fluctuation decoupling unit calculates the arithmetic mean of the encoded sequences of each group within a sliding window of a preset length as the trend component and uses the difference between the original sequence and the trend component as the fluctuation component. The hierarchical temporal coding unit performs multi-level temporal feature extraction on the trend component and the fluctuation component respectively and outputs the corresponding hierarchical representation sequence; The store collaboration prototype memory unit maintains a set of prototype entries associated with store identifiers in the cloud and outputs prototype retrieval results for time-series samples of each store window. Hierarchical contrastive learning is performed on the representation sequences at each level to construct the hierarchical contrastive learning objective: The same-level representations of the same store window time series samples obtained under different clipping scales are used as positive sample pairs, and the same-level representations corresponding to different store identifiers are used as negative sample pairs. The prototype representations retrieved by the store collaborative prototype memory unit are used as additional contrast constraint terms for the store window time series samples to participate in the hierarchical contrast learning process, thereby updating and improving the parameters of the TS2Vec model. Hierarchical aggregation and scale fusion processing are performed on the hierarchical representation sequences corresponding to each clipping scale to obtain the store operation status vector representing the comprehensive operation status of the store window time series samples.

[0012] Optionally, the generation of the operational phase transition driving force field includes: Obtain the store operation status vector, and form a store operation status vector sequence according to the store identifier and time window order. Each store operation status vector in the store operation status vector sequence is associated with a corresponding timestamp and time window identifier. Perform latent state mapping processing on the store operation state vector sequence, input the store operation state vector corresponding to each time window into the latent state mapping structure, output the corresponding store operation latent state vector, and perform normalization processing on the store operation latent state vector to obtain the store latent state parameters. A potential state energy description structure for store operations is constructed in the cloud. This structure includes a steady-state benchmark generation module, a hysteresis coupling potential energy calculation module, and a weight calibration and fusion module. The steady-state baseline generation module generates a store steady-state baseline vector based on the store operation state vector sequence within a preset historical window; The hysteresis coupling potential energy calculation module generates potential energy component parameters based on the degree of deviation of the store operation potential state vector from the store steady-state baseline vector in the current time window, the magnitude of change relative to the previous time window, and the degree of difference between the short-term deviation change sequence and the long-term deviation change sequence. The weight calibration and fusion module determines the potential energy fusion weight set based on the missing rate index and anomaly label index of the standardized time series dataset of the store, fuses the potential energy component parameters to obtain the potential state energy value of the store operation, and outputs the energy component parameters corresponding to the potential state energy value of the store operation. Based on the intensity of the change in the energy value of the potential state of store operation between adjacent time windows and the direction of the change in the vector of the potential state of store operation, operational phase change driving force parameters are generated, including driving force amplitude parameters and driving force direction parameters. The timestamp, store operation potential state vector, store operation potential state energy value, operation phase change driving force parameter and energy component parameter are organized and stored in chronological order to generate the operation phase change driving force field.

[0013] Optionally, the generation of cloud-based collaborative operation strategies based on cross-policy perturbation propagation tensors includes: Obtain the operational phase change driving force field, extract the set of potential state parameters of each store and each time window, and establish the potential state index relationship of stores according to the store identifier and time window identifier. A store collaboration relationship description structure is constructed in the cloud to quantitatively model the collaboration relationship between stores. The store collaboration relationship description structure generates store collaboration strength parameters based on the historical customer flow linkage degree, inventory transfer association frequency, activity execution synchronization degree and spatial distance information of stores in the standardized time series dataset. The store collaboration strength parameters are normalized to form a store collaboration relationship matrix. A cross-strategy perturbation propagation tensor structure is constructed in the cloud. The cross-strategy perturbation propagation tensor structure uses the store operation strategy dimension, the store potential state parameter dimension, and the store identification dimension as tensor axes. By statistically learning the correspondence between changes in store operation strategies and changes in store potential state parameters in historical time windows, perturbation propagation weights of each strategy dimension to each potential state parameter dimension are generated. Cross-store propagation calibration is performed on the perturbation propagation weights based on the store collaboration relationship matrix to form a strategy perturbation propagation structure with cross-store diffusion capabilities. Based on the cross-strategy perturbation propagation tensor structure and the set of potential state parameters of stores corresponding to the current time window, the perturbation response results of store operation strategies are calculated, and a strategy interference risk description vector is constructed. The strategy interference risk description vector is used to characterize the degree of mutual influence and conflict risk level of different store operation strategies within the current store and related stores. Based on the strategy intervention risk description vector, collaborative decision-making is generated for store operation strategies, and cloud-based collaborative operation strategies corresponding to each store and each time window are output.

[0014] Optionally, the formation of time-series data of store execution behavior includes: Obtain the cloud-based collaborative operation strategy and encapsulate the cloud-based collaborative operation strategy into a strategy task data package. The strategy task data package includes a store identifier, a time window identifier, a strategy version number, a strategy effective start and end timestamp, a strategy content payload, and a strategy distribution control flag. The cloud-based collaborative platform determines the strategy delivery path based on the store identifier and the store's online status, and establishes a target delivery queue for the strategy task data packet based on the terminal identifier of the store execution end. The strategy task data packet is then written into the target delivery queue according to the effective start timestamp corresponding to the time window identifier. The strategy task data package is pushed from the target delivery queue to the corresponding store execution terminal, and the integrity verification and time window verification are performed on the strategy task data package at the store execution terminal. At the store execution end, idempotency control and conflict control are performed based on the policy version number of the policy task data packet. Idempotency control includes suppressing repeated execution of policy task data packets with the same policy version number. Conflict control includes prioritizing multiple policy task data packets that are simultaneously in the effective window and determining the execution sequence. The actual execution behavior data of the store in response to the cloud-based collaborative operation strategy is collected at the store execution end. The actual execution behavior data includes strategy reception timestamp, strategy confirmation timestamp, strategy execution start timestamp, strategy execution end timestamp, execution status identifier, execution action record, and execution result indicator. The actual execution behavior data is then organized in chronological order to form store execution behavior time sequence data.

[0015] Optionally, the step of updating the cloud-based collaborative operation strategy generation parameters by applying distribution consistency constraints and correlation corrections to the characteristics of cloud-based collaborative operation strategies and the characteristics of store execution behaviors includes: Obtain the cloud-based collaborative operation strategy, extract the strategy feature information corresponding to the cloud-based collaborative operation strategy based on the store identifier and time window identifier, encode the strategy feature information according to the preset feature encoding rules, and form the cloud-based collaborative operation strategy feature vector. Acquire time-series data of store execution behavior, align the time-series data of store execution behavior according to the time sequence of strategy reception, strategy confirmation, strategy execution start, strategy execution end and execution result, and encode the execution behavior according to the preset event coding rules to form a feature vector of store execution behavior; A dual-domain association mapping mechanism is constructed in the cloud to represent the correspondence between cloud collaborative operation strategy features and store execution behavior features. The dual-domain association mapping mechanism maps the cloud collaborative operation strategy feature vector and the store execution behavior feature vector respectively, transforming them into a unified potential representation space to form a strategy domain potential representation and an execution behavior domain potential representation. Based on the consistency constraint processing of the distribution of potential representations in the strategy domain and potential representations in the execution behavior domain, the consistency constraint results are generated by measuring the overall distribution characteristics of the potential representations in the strategy domain and potential representations in the execution behavior domain corresponding to the same store identifier and the same time window identifier. Based on the distribution consistency constraint results, the association relationship in the dual-domain association mapping mechanism is corrected to reduce the matching degree between potential representations corresponding to different store identifiers or different time window identifiers. Based on the corrected association relationship, strategy parameter update information is generated to update the cloud collaborative operation strategy generation parameters.

[0016] The beneficial effects of this invention are: This invention achieves a continuous and dynamic depiction of store operational status by uniformly collecting and deeply characterizing multi-source operational time-series data. Compared with existing technologies that rely primarily on discrete indicators or static reports, this invention reflects the inherent trends in store operational status from a time-evolution perspective, effectively reducing decision-making biases caused by fragmented data or information lag, improving the accuracy and stability of store operational status identification, and providing a reliable data foundation for cloud-based collaborative management.

[0017] This invention introduces an operational phase change driving force field and a cross-strategy perturbation propagation modeling mechanism in the cloud, enabling the systematic modeling and evaluation of the mutual influence relationships between different store operation strategies. The generation of cloud-based collaborative operation strategies no longer depends on fixed rules or a single strategy dimension, but can comprehensively consider the potential state of stores and the collaborative relationships between stores, realize multi-strategy collaborative reasoning and dynamic adjustment, and improve the timeliness of cloud strategy response and cross-store linkage effect.

[0018] This invention establishes a mapping mechanism between cloud-based collaborative operation strategies and actual store execution, introducing a dynamic closed-loop optimization process based on execution feedback. This allows cloud-based strategy-generated parameters to be continuously adjusted according to actual store performance. This closed-loop mechanism effectively reduces the discrepancy between cloud-based strategies and store execution, improving the consistency and sustainable optimization capabilities of strategy execution, thereby enhancing the robustness and practical application value of the overall store digital operation management system. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of a cloud-based collaborative digital operation management system for stores proposed in this invention; Figure 2 This is a flowchart illustrating a cloud-based collaborative digital operation management method for stores proposed in this invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0021] refer to Figure 1 A cloud-based collaborative digital operation management system for stores, comprising: The operational data acquisition module is used to collect multi-source operational time-series data generated during store operations and preprocess it to form a standardized time-series dataset for the store. The state representation generation module is used to upload standardized time-series datasets of stores to the cloud collaboration platform and input them into the improved TS2Vec model to generate store operation state vectors. The latent energy modeling module is used to construct the energy description structure of the potential state of store operation based on the store operation state vector, and generate the driving force field of operation phase transition. The strategy collaboration generation module is used to construct cross-strategy perturbation propagation tensors based on the store potential state parameters output by the operational phase change driving force field, and generate cloud-based collaborative operation strategies. The strategy delivery and execution module is used to deliver cloud-based collaborative operation strategies to the corresponding store execution terminals, collect actual execution behavior data, and form time-series data of store execution behavior. The execution feedback update module is used to build a two-domain association mapping mechanism between cloud collaborative operation strategies and store execution behaviors based on the time-series data of store execution behaviors, and to update the parameters generated by the cloud collaborative operation strategies.

[0022] refer to Figure 2 A cloud-based collaborative digital operation management method for stores includes: Collect multi-source operational time-series data generated during store operations at each store, preprocess the multi-source operational time-series data, and form a standardized time-series dataset for each store. Upload the standardized time-series dataset of stores to the cloud collaboration platform and input it into the improved TS2Vec model. Perform multi-scale time pruning and hierarchical comparison learning on the standardized time-series dataset of stores to generate store operation status vectors. Based on the store operation state vector, a potential state energy description structure for store operation is constructed, and an operation phase transition driving force field is generated according to the continuous evolution relationship of the potential state of store operation over time. Based on the store potential state parameters output by the phase transition driving force field of operation, a cross-strategy perturbation propagation tensor is constructed in the cloud to model the relationship between store operation strategy, store potential state and the collaborative relationship between stores. Based on the cross-strategy perturbation propagation tensor, a cloud-based collaborative operation strategy is generated. The cloud-based collaborative operation strategy is distributed to the corresponding store execution terminal, and the actual execution behavior data of the store in response to the cloud-based collaborative operation strategy is collected to form time-series data of store execution behavior. Based on the time-series data of store execution behavior, a dual-domain association mapping mechanism is constructed in the cloud to represent the correspondence between cloud collaborative operation strategies and store execution behavior. By constraining the distribution consistency and correcting the association relationship between the characteristics of cloud collaborative operation strategies and the characteristics of store execution behavior, the generation parameters of cloud collaborative operation strategies are updated.

[0023] In this embodiment, the multi-source operational time-series data includes sales data, inventory data, customer flow data, staff scheduling data, equipment operation data, and activity execution data.

[0024] In this embodiment, the preprocessing of multi-source operational time-series data includes timestamp alignment, abnormal data removal, and unified field encoding of the multi-source operational time-series data.

[0025] In this embodiment, generating the store operation status vector includes: The standardized time series dataset of stores is divided into windows according to a preset time window length to obtain store window time series samples that correspond one-to-one with each time window. The store identifier and the start and end timestamps of the window are recorded for each store window time series sample. The preset time window length is 15 minutes to 2 hours. Multi-scale time clipping is performed on the time series sample of each store window. Within the time series sample of the same store window, multiple subsequences are extracted according to the preset clipping scale set. The clipping start time step index, clipping length and corresponding scale identifier are recorded for each subsequence. Multiple subsequences are input into the improved TS2Vec model, which includes a multi-source channel grouping coding unit, a trend-fluctuation decoupling unit, a hierarchical time series coding unit, and a store collaboration prototype memory unit, wherein: The multi-source channel grouping coding unit groups the feature channels according to the feature sources of the standardized time-series dataset of the stores and performs temporal convolutional coding on each group separately; The trend-fluctuation decoupling unit calculates the arithmetic mean of the encoded sequences of each group within a sliding window of a preset length as the trend component and uses the difference between the original sequence and the trend component as the fluctuation component. The hierarchical temporal coding unit performs multi-level temporal feature extraction on the trend component and the fluctuation component respectively and outputs the corresponding hierarchical representation sequence; The store collaboration prototype memory unit maintains a set of prototype entries associated with store identifiers in the cloud and outputs prototype retrieval results for time-series samples of each store window. Hierarchical contrastive learning is performed on the representation sequences at each level to construct the hierarchical contrastive learning objective: The same-level representations of the same store window time series samples obtained under different clipping scales are used as positive sample pairs, and the same-level representations corresponding to different store identifiers are used as negative sample pairs. The prototype representations retrieved by the store collaborative prototype memory unit are used as additional contrast constraint terms for the store window time series samples to participate in the hierarchical contrast learning process, thereby updating and improving the parameters of the TS2Vec model. Hierarchical aggregation and scale fusion processing are performed on the hierarchical representation sequences corresponding to each clipping scale to obtain the store operation status vector representing the comprehensive operation status of the store window time series samples.

[0026] In this embodiment, the generation of the operational phase change driving force field includes: Obtain the store operation status vector, and form a store operation status vector sequence according to the store identifier and time window order. Each store operation status vector in the store operation status vector sequence is associated with a corresponding timestamp and time window identifier. Perform latent state mapping processing on the store operation state vector sequence, input the store operation state vector corresponding to each time window into the latent state mapping structure, output the corresponding store operation latent state vector, and perform normalization processing on the store operation latent state vector to obtain the store latent state parameters. A potential state energy description structure for store operations is constructed in the cloud. This structure includes a steady-state benchmark generation module, a hysteresis coupling potential energy calculation module, and a weight calibration and fusion module. The steady-state baseline generation module generates a store steady-state baseline vector based on the store operation state vector sequence within a preset historical window, specifically as follows: Select consecutive store operation status vectors in chronological order within a preset historical window; Anomaly vector removal processing is performed on the selected store operation status vector sequence. The anomaly vector removal processing excludes store operation status vectors whose deviation from the overall distribution of the sequence exceeds a preset threshold based on the similarity threshold between vectors. The store operation status vector sequence after removing anomalies is aggregated by time-weighted average to generate a store steady-state benchmark vector to characterize the steady-state operation level of the store. The hysteresis coupling potential energy calculation module generates potential energy component parameters based on the degree of deviation of the store operation potential state vector from the store steady-state baseline vector in the current time window, the magnitude of change relative to the previous time window, and the degree of difference between the short-term deviation change sequence and the long-term deviation change sequence. The weight calibration and fusion module determines the potential energy fusion weight set based on the missing rate index and anomaly label index of the standardized time series dataset of the store, fuses the potential energy component parameters to obtain the potential state energy value of the store operation, and outputs the energy component parameters corresponding to the potential state energy value of the store operation. Based on the intensity of the change in the energy value of the potential state of store operation between adjacent time windows and the direction of the change in the vector of the potential state of store operation, operational phase change driving force parameters are generated, including driving force amplitude parameters and driving force direction parameters. The timestamp, store operation potential state vector, store operation potential state energy value, operation phase change driving force parameter and energy component parameter are organized and stored in chronological order to generate the operation phase change driving force field.

[0027] In this embodiment, the generation of cloud-based collaborative operation strategies based on cross-strategy perturbation propagation tensors includes: Obtain the operational phase change driving force field, extract the set of potential state parameters of each store and each time window, and establish the potential state index relationship of stores according to the store identifier and time window identifier. A store collaboration relationship description structure is constructed in the cloud to quantitatively model the collaboration relationship between stores. The store collaboration relationship description structure generates store collaboration strength parameters based on the historical customer flow linkage degree, inventory transfer association frequency, activity execution synchronization degree and spatial distance information of stores in the standardized time series dataset. The store collaboration strength parameters are normalized to form a store collaboration relationship matrix. A cross-strategy perturbation propagation tensor structure is constructed in the cloud. The cross-strategy perturbation propagation tensor structure uses the store operation strategy dimension, the store potential state parameter dimension, and the store identification dimension as tensor axes. By statistically learning the correspondence between changes in store operation strategies and changes in store potential state parameters in historical time windows, perturbation propagation weights of each strategy dimension to each potential state parameter dimension are generated. Cross-store propagation calibration is performed on the perturbation propagation weights based on the store collaboration relationship matrix to form a strategy perturbation propagation structure with cross-store diffusion capabilities. Based on the cross-strategy perturbation propagation tensor structure and the set of potential store state parameters corresponding to the current time window, the perturbation response results of store operation strategies are calculated, and a strategy interference risk description vector is constructed. This vector characterizes the degree of mutual influence and conflict risk level of different store operation strategies within the current store and related stores. Specifically, the calculation of the perturbation response results of store operation strategies involves: Based on the set of potential state parameters of stores corresponding to the current time window, extract the strategy action vector corresponding to the operation strategy of each store. Using the cross-policy perturbation propagation tensor structure, the policy action vector is mapped to the store potential state parameter space to obtain the perturbation impact of each store operation policy on the current store potential state. Based on the description structure of store collaboration relationships, the disturbance impact results are propagated and weighted within the scope of related stores to form a store operation strategy disturbance response result that reflects the comprehensive disturbance impact of each store's operation strategy within the current store and related stores. Based on the strategy intervention risk description vector, collaborative decision-making is generated for store operation strategies, and cloud-based collaborative operation strategies corresponding to each store and each time window are output.

[0028] In this embodiment, the step of forming time-series data of store execution behavior includes: Obtain the cloud-based collaborative operation strategy and encapsulate the cloud-based collaborative operation strategy into a strategy task data package. The strategy task data package includes a store identifier, a time window identifier, a strategy version number, a strategy effective start and end timestamp, a strategy content payload, and a strategy distribution control flag. The cloud-based collaborative platform determines the strategy delivery path based on the store identifier and the store's online status, and establishes a target delivery queue for the strategy task data packet based on the terminal identifier of the store execution end. The strategy task data packet is then written into the target delivery queue according to the effective start timestamp corresponding to the time window identifier. The strategy task data package is pushed from the target delivery queue to the corresponding store execution terminal, and the integrity verification and time window verification are performed on the strategy task data package at the store execution terminal. At the store execution end, idempotency control and conflict control are performed based on the policy version number of the policy task data packet. Idempotency control includes suppressing repeated execution of policy task data packets with the same policy version number. Conflict control includes prioritizing multiple policy task data packets that are simultaneously in the effective window and determining the execution sequence. The actual execution behavior data of the store in response to the cloud-based collaborative operation strategy is collected at the store execution end. The actual execution behavior data includes strategy reception timestamp, strategy confirmation timestamp, strategy execution start timestamp, strategy execution end timestamp, execution status identifier, execution action record, and execution result indicator. The actual execution behavior data is then organized in chronological order to form store execution behavior time sequence data.

[0029] In this embodiment, updating the cloud-based collaborative operation strategy generation parameters by applying distribution consistency constraints and correlation corrections to the characteristics of cloud-based collaborative operation strategies and the characteristics of store execution behaviors includes: The cloud-based collaborative operation strategy is obtained. Based on the store identifier and time window identifier, the strategy feature information corresponding to the cloud-based collaborative operation strategy is extracted. The strategy feature information is then encoded according to a preset feature encoding rule to form a cloud-based collaborative operation strategy feature vector. The preset feature encoding rule is as follows: The strategy feature information is grouped and coded according to the structured fields of the store operation strategy. The structured fields include strategy type identifier, strategy target identifier, strategy effective time parameter, and strategy strength parameter. The strategy type identifier and the strategy target identifier are converted into a fixed-dimensional category encoding vector using a discrete category mapping method; The strategy effective time parameter and strategy strength parameter are converted into continuous numerical encoding vectors using numerical normalization. The category encoding vector and the continuous numerical encoding vector are then concatenated according to the preset field order to form the feature vector of the cloud collaborative operation strategy. Acquire time-series data of store execution behavior, align the time-series data of store execution behavior according to the time sequence of strategy reception, strategy confirmation, strategy execution start, strategy execution end and execution result, and encode the execution behavior according to the preset event coding rules to form a feature vector of store execution behavior; A dual-domain association mapping mechanism is constructed in the cloud to represent the correspondence between cloud-based collaborative operation strategy features and store execution behavior features. This mechanism maps both the cloud-based collaborative operation strategy feature vector and the store execution behavior feature vector to a unified latent representation space, forming a strategy domain latent representation and an execution behavior domain latent representation. The dual-domain association mapping mechanism refers to: In the cloud, a strategy domain mapping substructure and an execution behavior domain mapping substructure are constructed respectively. The strategy domain mapping substructure is used to perform feature transformation processing on the feature vector of cloud collaborative operation strategy, and the execution behavior domain mapping substructure is used to perform feature transformation processing on the feature vector of store execution behavior. By using the strategy domain mapping substructure and the execution behavior domain mapping substructure, the feature vectors of cloud collaborative operation strategy and the feature vectors of store execution behavior are mapped to a unified potential representation space with consistent dimensions and semantic alignment, forming the potential representation of strategy domain and the potential representation of execution behavior domain. The potential representations of the strategy domain and the potential representations of the execution behavior domain are paired and stored according to the store identifier and the time window identifier; Based on the consistency constraint processing of the distribution of potential representations in the strategy domain and potential representations in the execution behavior domain, the consistency constraint results are generated by measuring the overall distribution characteristics of the potential representations in the strategy domain and potential representations in the execution behavior domain corresponding to the same store identifier and the same time window identifier. Based on the distribution consistency constraint results, the association relationship in the dual-domain association mapping mechanism is corrected to reduce the matching degree between potential representations corresponding to different store identifiers or different time window identifiers. Based on the corrected association relationship, strategy parameter update information is generated to update the cloud collaborative operation strategy generation parameters.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a chain convenience store chain. This chain has 18 stores in the area, with store sizes ranging from 60 to 110 square meters. The main product categories are ready-to-eat foods, beverages, daily necessities, and seasonal promotional items. The geographical locations and customer demographics of each store vary. Some stores are located in areas with concentrated office buildings, experiencing significant weekday lunch and evening peak hours, while others are located near residential areas, with relatively concentrated customer traffic on weekends and evenings. Before deploying this invention, regional operations mainly relied on store staff experience and periodic reports for replenishment and promotional planning, which generally resulted in problems such as delayed identification of store operational status, slow response to cloud-based strategies, and insufficient consistency in strategy execution.

[0031] In this scenario, after applying the cloud-based collaborative digital operation management method for stores proposed in this invention, each store continuously collects sales data, inventory change data, customer flow data, staff scheduling data, and activity execution data through existing POS systems, inventory management systems, and scheduling systems, forming multi-source operational time-series data. The system first performs timestamp alignment and field unification processing on the collected multi-source operational time-series data to generate a standardized time-series dataset for each store, and uploads it to the cloud collaborative platform in real time.

[0032] The cloud-based collaborative platform inputs standardized time-series datasets from stores into an improved TS2Vec model. It performs multi-scale time pruning and hierarchical comparative learning on the multi-source time-series data from stores, generating store operation state vectors that reflect the comprehensive operational status of each store under different time windows. Based on these store operation state vectors, the system constructs a potential energy description structure for store operations and generates an operational phase transition driving force field based on the continuous evolution relationship of the potential store operation states over time, used to characterize the changing trends of store operation states.

[0033] The cloud platform utilizes the store's potential state parameters output from the operational phase transition driving force field to construct a cross-strategy perturbation propagation model. This model correlates store operational strategies with store potential states and the collaborative relationships between stores, thereby generating cloud-based collaborative operational strategies. These strategies are then distributed to the corresponding store execution terminals in time windows. During execution, stores automatically record information such as strategy reception time, execution start and end times, and execution results, forming time-series data of store execution behavior, which is then transmitted back to the cloud.

[0034] Based on the time-series data of store execution behavior, the cloud constructs a dual-domain association mapping mechanism to represent the correspondence between cloud collaborative operation strategies and store execution behavior. By constraining the distribution consistency and correcting the association relationship between the characteristics of cloud collaborative operation strategies and the characteristics of store execution behavior, the cloud collaborative operation strategy generation parameters are continuously updated. The system gradually reduces the deviation between cloud strategies and actual store execution, making strategy generation more in line with the real operation of stores.

[0035] After the system ran continuously in the region for eight weeks, a comparison of the operational data before and after deployment revealed that the overall operational stability of the stores and the consistency of strategy execution were significantly improved, without placing an additional burden on the existing store systems, demonstrating good engineering feasibility.

[0036] Table 1. Comparison of Regional Store Operation Indicators Before and After Deployment of This Invention

[0037] As can be seen from the data in Table 1, after deploying the cloud-based collaborative digital operation management method for stores described in this invention, the overall operating performance and operational stability of regional stores have improved. The average daily sales per store increased from RMB 12,860 before deployment to RMB 13,540, with the increase remaining within a reasonable range. This indicates that the system did not drive short-term sales through aggressive promotional methods, but rather achieved relatively stable and sustainable sales growth based on optimizing replenishment rhythm and operational collaboration.

[0038] Regarding operational assurance metrics, the stockout rate during peak hours decreased from 5.2% to 3.9%, and inventory turnover days decreased from 8.3 days to 7.4 days. This indicates that the system can more accurately identify changes in store operational status and generate and distribute collaborative operational strategies in advance during critical periods, effectively alleviating stockouts and overstocking caused by delayed replenishment or unreasonable inventory allocation. This change reflects the practical effectiveness of this invention in store operational status modeling and strategy generation.

[0039] From a management and execution efficiency perspective, the average response time for cloud-based strategies has been reduced from 36 hours to 9 hours, and the on-time execution rate has increased from 81% to 92%. Simultaneously, the number of strategy conflicts or rollbacks has decreased from an average of 6.1 times per week to 2.3 times, indicating that the generation and adjustment of cloud-based collaborative strategies are more timely, and the matching degree between strategy content and actual store execution conditions has improved. By introducing a dynamic closed-loop optimization mechanism based on execution behavior feedback, cloud-based strategies can be continuously revised according to the actual execution situation in stores, effectively reducing strategy conflicts and improving the consistency and reliability of overall operational management.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A cloud-based collaborative digital operation management system for stores, characterized in that, include: The operational data acquisition module is used to collect multi-source operational time-series data generated during store operations and preprocess it to form a standardized time-series dataset for the store. The state representation generation module is used to upload standardized time-series datasets of stores to the cloud collaboration platform and input them into the improved TS2Vec model to generate store operation state vectors. The latent energy modeling module is used to construct the energy description structure of the potential state of store operation based on the store operation state vector, and generate the driving force field of operation phase transition. The strategy collaboration generation module is used to construct cross-strategy perturbation propagation tensors based on the store potential state parameters output by the operational phase change driving force field, and generate cloud-based collaborative operation strategies. The strategy delivery and execution module is used to deliver cloud-based collaborative operation strategies to the corresponding store execution terminals, collect actual execution behavior data, and form time-series data of store execution behavior. The execution feedback update module is used to build a two-domain association mapping mechanism between cloud collaborative operation strategies and store execution behaviors based on the time-series data of store execution behaviors, and to update the parameters generated by the cloud collaborative operation strategies.

2. A cloud-based collaborative store digital operation management method, applied to the cloud-based collaborative store digital operation management system described in claim 1, characterized in that, include: Collect multi-source operational time-series data generated during store operations at each store, preprocess the multi-source operational time-series data, and form a standardized time-series dataset for each store. Upload the standardized time-series dataset of stores to the cloud collaboration platform and input it into the improved TS2Vec model. Perform multi-scale time pruning and hierarchical comparison learning on the standardized time-series dataset of stores to generate store operation status vectors. Based on the store operation state vector, a potential state energy description structure for store operation is constructed, and an operation phase transition driving force field is generated according to the continuous evolution relationship of the potential state of store operation over time. Based on the store potential state parameters output by the phase transition driving force field of operation, a cross-strategy perturbation propagation tensor is constructed in the cloud to model the relationship between store operation strategy, store potential state and the collaborative relationship between stores. Based on the cross-strategy perturbation propagation tensor, a cloud-based collaborative operation strategy is generated. The cloud-based collaborative operation strategy is distributed to the corresponding store execution terminal, and the actual execution behavior data of the store in response to the cloud-based collaborative operation strategy is collected to form time-series data of store execution behavior. Based on the time-series data of store execution behavior, a dual-domain association mapping mechanism is constructed in the cloud to represent the correspondence between cloud collaborative operation strategies and store execution behavior. By constraining the distribution consistency and correcting the association relationship between the characteristics of cloud collaborative operation strategies and the characteristics of store execution behavior, the generation parameters of cloud collaborative operation strategies are updated.

3. The cloud-based collaborative digital operation management method for stores according to claim 2, characterized in that, The multi-source operational time-series data includes sales data, inventory data, customer flow data, staff scheduling data, equipment operation data, and event execution data.

4. The cloud-based collaborative digital operation management method for stores according to claim 2, characterized in that, The preprocessing of multi-source operational time-series data includes timestamp alignment, abnormal data removal, and unified field encoding.

5. The cloud-based collaborative digital operation management method for stores according to claim 2, characterized in that, The generated store operation status vector includes: The standardized time series dataset of stores is divided into windows according to a preset time window length to obtain store window time series samples that correspond one-to-one with each time window. The store identifier and the start and end timestamps of the window are recorded for each store window time series sample. Multi-scale time clipping is performed on the time series sample of each store window. Within the time series sample of the same store window, multiple subsequences are extracted according to the preset clipping scale set. The clipping start time step index, clipping length and corresponding scale identifier are recorded for each subsequence. Multiple subsequences are input into the improved TS2Vec model, which includes a multi-source channel grouping coding unit, a trend-fluctuation decoupling unit, a hierarchical time series coding unit, and a store collaboration prototype memory unit, wherein: The multi-source channel grouping coding unit groups the feature channels according to the feature sources of the standardized time-series dataset of the stores and performs temporal convolutional coding on each group separately; The trend-fluctuation decoupling unit calculates the arithmetic mean of the encoded sequences of each group within a sliding window of a preset length as the trend component and uses the difference between the original sequence and the trend component as the fluctuation component. The hierarchical temporal coding unit performs multi-level temporal feature extraction on the trend component and the fluctuation component respectively and outputs the corresponding hierarchical representation sequence; The store collaboration prototype memory unit maintains a set of prototype entries associated with store identifiers in the cloud and outputs prototype retrieval results for time-series samples of each store window. Hierarchical contrastive learning is performed on the representation sequences at each level to construct the hierarchical contrastive learning objective: The same-level representations of the same store window time series samples obtained under different clipping scales are used as positive sample pairs, and the same-level representations corresponding to different store identifiers are used as negative sample pairs. The prototype representations retrieved by the store collaborative prototype memory unit are used as additional contrast constraint terms for the store window time series samples to participate in the hierarchical contrast learning process, thereby updating and improving the parameters of the TS2Vec model. Hierarchical aggregation and scale fusion processing are performed on the hierarchical representation sequences corresponding to each clipping scale to obtain the store operation status vector representing the comprehensive operation status of the time series samples of the store window.

6. The cloud-based collaborative digital operation management method for stores according to claim 2, characterized in that, The generation and operation of the phase transition driving force field includes: Obtain the store operation status vector, and form a store operation status vector sequence according to the store identifier and time window order. Each store operation status vector in the store operation status vector sequence is associated with a corresponding timestamp and time window identifier. Perform latent state mapping processing on the store operation state vector sequence, input the store operation state vector corresponding to each time window into the latent state mapping structure, output the corresponding store operation latent state vector, and perform normalization processing on the store operation latent state vector to obtain the store latent state parameters. A potential state energy description structure for store operations is constructed in the cloud. This structure includes a steady-state benchmark generation module, a hysteresis coupling potential energy calculation module, and a weight calibration and fusion module. The steady-state baseline generation module generates a store steady-state baseline vector based on the store operation state vector sequence within a preset historical window; The hysteresis coupling potential energy calculation module generates potential energy component parameters based on the degree of deviation of the store operation potential state vector from the store steady-state baseline vector in the current time window, the magnitude of change relative to the previous time window, and the degree of difference between the short-term deviation change sequence and the long-term deviation change sequence. The weight calibration and fusion module determines the potential energy fusion weight set based on the missing rate index and anomaly label index of the standardized time series dataset of the store, fuses the potential energy component parameters to obtain the potential state energy value of the store operation, and outputs the energy component parameters corresponding to the potential state energy value of the store operation. Based on the intensity of the change in the energy value of the potential state of store operation between adjacent time windows and the direction of the change in the vector of the potential state of store operation, operational phase change driving force parameters are generated, including driving force amplitude parameters and driving force direction parameters. The timestamp, store operation potential state vector, store operation potential state energy value, operation phase change driving force parameter and energy component parameter are organized and stored in chronological order to generate the operation phase change driving force field.

7. The cloud-based collaborative digital operation management method for stores according to claim 2, characterized in that, The cloud-based collaborative operation strategy generated based on cross-strategy perturbation propagation tensors includes: Obtain the operational phase change driving force field, extract the set of potential state parameters of each store and each time window, and establish the potential state index relationship of stores according to the store identifier and time window identifier. A store collaboration relationship description structure is constructed in the cloud to quantitatively model the collaboration relationship between stores. The store collaboration relationship description structure generates store collaboration strength parameters based on the historical customer flow linkage degree, inventory transfer association frequency, activity execution synchronization degree and spatial distance information of stores in the standardized time series dataset. The store collaboration strength parameters are normalized to form a store collaboration relationship matrix. A cross-strategy perturbation propagation tensor structure is constructed in the cloud. The cross-strategy perturbation propagation tensor structure uses the store operation strategy dimension, the store potential state parameter dimension, and the store identification dimension as tensor axes. By statistically learning the correspondence between changes in store operation strategies and changes in store potential state parameters in historical time windows, perturbation propagation weights of each strategy dimension to each potential state parameter dimension are generated. Cross-store propagation calibration is performed on the perturbation propagation weights based on the store collaboration relationship matrix to form a strategy perturbation propagation structure with cross-store diffusion capabilities. Based on the cross-strategy perturbation propagation tensor structure and the set of potential state parameters of stores corresponding to the current time window, the perturbation response results of store operation strategies are calculated, and a strategy interference risk description vector is constructed. The strategy interference risk description vector is used to characterize the degree of mutual influence and conflict risk level of different store operation strategies within the current store and related stores. Based on the strategy intervention risk description vector, collaborative decision-making is generated for store operation strategies, and cloud-based collaborative operation strategies corresponding to each store and each time window are output.

8. The cloud-based collaborative digital operation management method for stores according to claim 2, characterized in that, The time-series data for forming store execution behavior includes: Obtain the cloud-based collaborative operation strategy and encapsulate the cloud-based collaborative operation strategy into a strategy task data package. The strategy task data package includes a store identifier, a time window identifier, a strategy version number, a strategy effective start and end timestamp, a strategy content payload, and a strategy distribution control flag. The cloud-based collaborative platform determines the strategy delivery path based on the store identifier and the store's online status, and establishes a target delivery queue for the strategy task data packet based on the terminal identifier of the store execution end. The strategy task data packet is then written into the target delivery queue according to the effective start timestamp corresponding to the time window identifier. The strategy task data package is pushed from the target delivery queue to the corresponding store execution terminal, and the integrity verification and time window verification are performed on the strategy task data package at the store execution terminal. At the store execution end, idempotency control and conflict control are performed based on the policy version number of the policy task data packet. Idempotency control includes suppressing repeated execution of policy task data packets with the same policy version number. Conflict control includes prioritizing multiple policy task data packets that are simultaneously in the effective window and determining the execution sequence. The actual execution behavior data of the store in response to the cloud-based collaborative operation strategy is collected at the store execution end. The actual execution behavior data includes strategy reception timestamp, strategy confirmation timestamp, strategy execution start timestamp, strategy execution end timestamp, execution status identifier, execution action record, and execution result indicator. The actual execution behavior data is then organized in chronological order to form store execution behavior time sequence data.

9. A cloud-based collaborative digital operation management method for stores according to claim 2, characterized in that, The process involves updating the cloud-based collaborative operation strategy generation parameters by applying distribution consistency constraints and correlation corrections to the characteristics of cloud-based collaborative operation strategies and the characteristics of store execution behaviors. Obtain the cloud-based collaborative operation strategy, extract the strategy feature information corresponding to the cloud-based collaborative operation strategy based on the store identifier and time window identifier, encode the strategy feature information according to the preset feature encoding rules, and form the cloud-based collaborative operation strategy feature vector. Acquire time-series data of store execution behavior, align the time-series data of store execution behavior according to the time sequence of strategy reception, strategy confirmation, strategy execution start, strategy execution end and execution result, and encode the execution behavior according to the preset event coding rules to form a feature vector of store execution behavior; A dual-domain association mapping mechanism is constructed in the cloud to represent the correspondence between cloud collaborative operation strategy features and store execution behavior features. The dual-domain association mapping mechanism maps the cloud collaborative operation strategy feature vector and the store execution behavior feature vector respectively, transforming them into a unified potential representation space to form a strategy domain potential representation and an execution behavior domain potential representation. Based on the consistency constraint processing of the distribution of potential representations in the strategy domain and potential representations in the execution behavior domain, the consistency constraint results are generated by measuring the overall distribution characteristics of the potential representations in the strategy domain and potential representations in the execution behavior domain corresponding to the same store identifier and the same time window identifier. Based on the distribution consistency constraint results, the association relationship in the dual-domain association mapping mechanism is corrected to reduce the matching degree between potential representations corresponding to different store identifiers or different time window identifiers. Based on the corrected association relationship, strategy parameter update information is generated to update the cloud collaborative operation strategy generation parameters.

Images