Global industrial product hidden demand intention collaborative awareness and adaptive supply chain synapse optimization system based on federal brain-like learning

Abstract

The invention belongs to the fields of artificial intelligence, federal learning, brain-like calculation, supply chain optimization and international trade digitalization, and aims to solve the problems of demand information lag, high customer acquisition cost, slow market response, data islands, difficulty in privacy compliance, low supply and demand matching precision, lack of continuous learning mechanism and the like of Chinese industrial product enterprises. And proposing a implicit demand perception and adaptive supply chain optimization system based on federal brain learning. According to the system, privacy protection is embedded in perception, learning and transmission processes of federal edge nodes, and a safety aggregation mechanism of a central core, a brain-like heuristic learning method and a vertical large model analysis capability are combined, so that collaborative perception and intelligent optimization of hidden demand intentions of global industrial products are realized, and the problem of data privacy compliance is thoroughly solved; and the sea-going efficiency and success rate of Chinese enterprises are obviously improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical fields of artificial intelligence, machine learning, federated learning, multi-modal data processing, brain-inspired computing, supply chain optimization, and digital marketing for international trade. In particular, the present application relates to a method and system for real-time, privacy-protected collaborative perception of implicit demand intentions in the global industrial goods market, and dynamic optimization of matching Chinese suppliers with overseas market demand based on this, thereby generating adaptive overseas marketing strategies. BACKGROUND

[0002] With the deepening of globalization and the continuous transformation and upgrading of China's manufacturing industry, more and more Chinese enterprises, especially manufacturing enterprises such as materials, industrial products, and industrial finished products, are actively seeking opportunities in overseas markets. However, under the existing technology and business model, Chinese enterprises going overseas, especially in emerging markets, generally face the following pain points and limitations:

[0003] 1. Lagging and limited demand information acquisition: Traditional market research methods and existing general overseas platforms (such as large e-commerce platforms and social media advertising platforms) mainly rely on explicit, structured procurement information or historical transaction data. For emerging markets, the "implicit demand intentions" (such as weak signals of industry trends, potential technology upgrade willingness of enterprises, and consumer preferences in specific cultural backgrounds) that are fragmented, unstructured, or even not yet explicitly expressed, existing systems lack real-time, efficient, and accurate perception and mining capabilities.

[0004] 2. Low customer acquisition efficiency and high conversion cost: Most enterprises still use a broad marketing strategy or rely on traditional manual channels. Due to insufficient supply-demand matching accuracy, it leads to low customer acquisition efficiency, long conversion period, and low input-output ratio.

[0005] 3. Slow market response speed and strategy update lag: The global market environment is complex and variable, and existing solutions are difficult to achieve millisecond-level real-time perception of market signals and rapid adaptive adjustment of strategies.

[0006] 4. Data silos and privacy compliance challenges: Customer and market data scattered around the world, especially private data and sensitive business information in overseas local markets, form data silos due to data privacy regulations (such as GDPR and CCPA) and business sensitivity, which cannot be effectively integrated, severely restricting the ability of global intelligent analysis and model training.

[0007] 5. Lack of autonomous learning and continuous optimization mechanism: Most AI systems require periodic offline training and a large amount of manual intervention, making it difficult to continuously and autonomously learn and optimize in real business scenarios, thus failing to achieve true intelligent evolution.

[0008] Therefore, there is an urgent need for a new method and system that can overcome the above-mentioned defects, and realize more accurate, efficient, and privacy-protected intelligent matching and strategy optimization of global industrial product supply and demand. SUMMARY

[0009] The present application aims to solve the core problems in the prior art, such as insufficient global industrial product implicit demand perception, slow market response, data island, difficulty in ensuring privacy compliance, limited supply and demand matching accuracy, and lack of continuous learning ability. To this end, the present application proposes a global industrial product implicit demand intention collaborative perception and adaptive supply chain synapse optimization system based on federated brain-like learning. The system embeds data privacy protection capabilities in its distributed edge nodes and their interaction processes with the central core.

[0010] The technical solution provided by the present application includes:

[0011] 1. Federated edge sensing nodes (Federated Edge Sensing Nodes): The node is mainly deployed in the form of lightweight software modules or containerized applications in local overseas markets or cloud environments controlled by overseas partners (for example, cloud servers hosting overseas partner websites and social media platforms, or integrated into partner-owned systems through SDK / API). The node is configured to:

[0012] o Real-time, low-power sensing and processing of local, multi-modal, heterogeneous, and unstructured spatio-temporal event streams. The spatio-temporal event stream includes but is not limited to text data stream (such as local social media comments, forum posts, instant messaging group chats, news reports, policy texts), behavior data stream (such as website click sequences, browsing duration, AI question and answer interaction patterns), image / video clips (such as visual features of specific industrial scenes or products), and structured / semi-structured data streams (such as desensitized local customs data segments, B2B platform price fluctuations).

[0013] o Encode the spatio-temporal event stream into a spatio-temporal event pulse sequence with timestamp, intensity, feature vector, and semantic label.

[0014] o Based on the spatio-temporal event pulse sequence and local privacy data, use a small neural network model inspired by the brain (such as a spiking neural network SNN, or a Transformer model with SNN characteristics such as event-driven and synaptic plasticity) for online learning to identify local-specific implicit demand intention patterns. The learning process utilizes synaptic plasticity mechanisms between neurons.

[0015] o Ensure that the local raw data does not leave the local premise, and apply privacy protection techniques (including but not limited to differential privacy, homomorphic encryption, etc.) to process the local model, generating encrypted and privacy-protected local model updates (rather than raw data).

[0016] o Transmit the local model updates to a central core through a secure communication protocol (e.g., TLS / SSL encrypted channel).

[0017] 2. Global Synapse Optimization Core: deployed in a central cloud system. This core is configured to:

[0018] o Receive and aggregate the local model updates transmitted from multiple global federated edge-aware nodes through a secure aggregation protocol (e.g., secure multi-party computation SMPc) to build and continuously optimize a global, dynamic, and adaptive "supply chain synapse network". This network represents Chinese suppliers (supply neurons) and global industrial demand intentions (demand intention neurons) in neurons, and the matching degree and business value in the strength and activity of synapse connections.

[0019] o Combine the vertical large model (VLM) in the supply chain field (e.g., large language model or multi-modal large model based on Transformer architecture) to conduct deep analysis, complex reasoning, and knowledge generation on the aggregated global knowledge and abstract event features, to mine more macro and strategic global implicit demand intentions and business values.

[0020] o Continuously optimize the strength and activity of synapses in the supply chain synapse network through global learning results and vertical model analysis, to achieve global synapse plasticity.

[0021] 3. Adaptive Supply Chain Matching & Outbound Strategy Generation Module: based on the dynamic "supply chain synapse network" formed by the global synapse optimization core, this module is used to:

[0022] o Real-time identification and activation of Chinese suppliers (supply neurons) that best match the demand intention when receiving high-intensity demand intention pulses or identifying new implicit demand.

[0023] o Adaptively generate and optimize a series of precise, multi-modal outbound strategies, including but not limited to: customized digital marketing content (such as intelligently generated text and video scripts, 3D display copywriting), precise advertising placement suggestions, high-priority buyer lead pushing, intelligent communication tactics and negotiation strategy suggestions.

[0024] o Collect the market feedback (such as click rate, inquiry volume, conversion rate, customer satisfaction) after the execution of the strategy, and return the feedback as new spatio-temporal event pulses to the federal edge perception node and the global synapse optimization core, driving continuous online learning and optimization of the entire system.

[0025] The present application realizes the collaborative perception and intelligent optimization of implicit demand intention of industrial products in the global range by embedding privacy protection capability in the perception, learning and transmission process of the federal edge perception node, combining the security aggregation mechanism of the central core, the learning method inspired by the brain and the powerful analysis ability of the vertical large model, and completely solving the problems of data privacy and compliance, thereby significantly improving the efficiency and success rate of Chinese enterprises going abroad.

[0026] The beneficial effects of the present application include:

[0027] 1. Revolutionary "implicit demand intention" discovery capability: breaking through the limitations of traditional explicit data analysis, the present application uses multi-modal data input and spatio-temporal pattern recognition capability of the brain-inspired model to infer the potential and unexpressed demand intention of industrial products from fuzzy and fragmented weak signals in real time and with high precision, bringing unprecedented market insight to enterprises and solving the pain points of current market opportunity perception lag.

[0028] 2. Extremely embedded data privacy and compliance protection: privacy protection capability is directly integrated into the data perception, local learning and model update transmission process of the federal edge perception node, and further protected by the security aggregation mechanism of the central core. This ensures that the original sensitive data does not leave the local, perfectly solves the pain points of cross-border data flow and privacy compliance, and greatly improves the trust and willingness of overseas partners.

[0029] 3. Globalization, collaborative evolution intelligence: through the federal learning mechanism to aggregate the local wisdom of federal edge nodes all over the world, the system can continuously and efficiently learn and adapt to the rapidly changing market dynamics of the world (especially emerging markets), realize real global collaborative evolution, and build a strong knowledge flywheel.

[0030] 4. High-precision, self-adaptive supply and demand matching and strategy optimization: the dynamic "supply chain synapse network" combined with the analysis ability of the vertical large model can continuously learn and optimize the supply and demand relationship like the human brain, achieving highly precise matching. At the same time, the strategy generation module can automatically adjust and optimize the overseas strategy according to real-time feedback, significantly improving transaction conversion rate and resource utilization efficiency.

[0031] 5. Lightweight, high-efficiency edge deployment: the federal edge perception node is deployed in software form, without additional hardware investment, with extremely low power consumption and high energy efficiency, easy to large-scale popularization and deployment, reducing the use threshold of enterprises.

[0032] 6. Constructing an insurmountable technical and data double moat: a complex technical architecture that combines federated learning, brain-inspired computing, multi-modal AI, and vertical large models, combined with a unique industry data flywheel effect, forms a high technical and data barrier to ensure long-term competitive advantage.

[0033] 7. Enhancing enterprise valuation and market leadership: transforming the company from a traditional service provider to a high-tech enterprise with a core AI brain and a global collaborative intelligent network, with huge market potential and higher valuation space, and expected to lead the new paradigm of industrial products going abroad. BRIEF DESCRIPTION OF DRAWINGS

[0034] · Figure 1 : The overall architecture of the system, clearly showing the federal edge perception node, global synapse optimization core, adaptive supply chain matching and export strategy generation module, and the communication and data flow between modules, and marking the privacy protection link.

[0035] · Figure 2 : Workflow diagram of the federal edge perception node, detailing multi-modal data perception, spatio-temporal event pulse coding, local learning and reasoning, and the generation and transmission process of privacy protection model updates.

[0036] · Figure 3 : Workflow diagram of the global synapse optimization core, detailing the process of federal aggregation, supply chain synapse network construction and optimization, and vertical large model analysis.

[0037] · Figure 4 : Conceptual diagram of the supply chain synapse network, showing the structure of supply neurons, demand intention neurons, and dynamic synapse connections. DETAILED DESCRIPTION

[0038] To describe the present invention in more detail, the following describes the present invention in conjunction with specific embodiments. The present invention is not limited to the examples listed below.

[0039] The "global industrial product implicit demand intention collaborative perception and adaptive supply chain synapse optimization system based on federal brain-inspired learning" proposed in the present invention is to build an intelligent network that can understand and respond to global industrial product supply and demand relationships through distributed perception, collaborative learning and dynamic connection, just like a biological brain.

[0040] I. System architecture and component details:

[0041] 1. Specific implementation of federal edge perception node (FL-ESN):

[0042] Deployment: FL-ESN is deployed as containerized applications (e.g., Docker containers) or microservices on servers provided by cloud service providers. These servers host the independent websites and social media accounts you build for free for overseas partners (e.g., independent website backends based on WordPress, Shopify, or self-developed platforms; services integrated with Facebook Business Manager, LinkedIn Pages, WhatsApp Business API). In addition, lightweight SDKs or API interfaces can be provided for overseas partners to integrate into their own local or cloud-based CRM, ERP, sales management software, etc. systems under explicit authorization, ensuring that core computing and learning are performed in a computing environment controlled or authorized by the partner.

[0043] Multi-modal data perception and spatio-temporal event pulse coding:

[0044] Text stream processing: Utilize natural language processing (NLP) techniques to perform word segmentation, named entity recognition, sentiment analysis, and topic modeling on text data such as social media comments, forum posts, and local news.

[0045] Behavior stream processing: Combine user behavior analysis tools and time series models to identify patterns in user click sequences, browsing duration, AI question and answer interactions, and download behavior, and build a user interest graph.

[0046] Image / video processing: Use computer vision (CV) techniques to perform object recognition, scene classification, and key feature extraction on images / videos such as industrial exhibition photos and product promotion videos.

[0047] Structured data processing: Clean, standardize, and perform correlation analysis on customs data and B2B platform data.

[0048] Pulse coding: Convert the above multi-modal raw data streams into "spatio-temporal event pulse sequences". Each pulse is a discrete event with a clear timestamp, intensity value (e.g., based on popularity, relevance, and influence calculation), feature vector, and semantic label (e.g., geographic location, industry, specific product category, emotional tendency, and procurement stage). For example, a news article about "a certain country's government announcing subsidies for environmentally friendly building materials" will be coded as a high-intensity pulse with labels such as "policy benefit", "environmentally friendly building materials", and "specific country".

[0049] Local implicit demand intention learning and reasoning:

[0050] *FL-ESN runs a lightweight neural network model inside, which is inspired by the principles of brain-like biology, with a strong emphasis on handling spatio-temporal information sequences and online learning capabilities. Specifically, a Spiking Neural Network (SNN) or an attention mechanism-based Transformer model can be adopted, but with lightweight processing to adapt to edge deployment environments.

[0051] *The model performs online learning and fine-tuning of model parameters on local private data by simulating synaptic plasticity between neurons (e.g., based on variants of Spike-Timing-Dependent Plasticity, STDP).

[0052] *By learning complex patterns, correlations, and temporal dependencies in these spatio-temporal spike sequences, FL-ESN can infer "implicit demand intention patterns" in local markets in real-time with extremely low power consumption. For example, by analyzing the series of events "a buyer frequently browses water treatment equipment websites -> downloads a white paper on membrane separation technology -> then follows experts in the wastewater treatment field on LinkedIn", FL-ESN can infer that it "is evaluating or planning a large industrial wastewater treatment project and has potential demand for membrane separation technology" implicit intention.

[0053] *Privacy-protected model updates:

[0054] *FL-ESN strictly ensures that raw sensitive data such as user IP, complete chat records, and specific transaction amounts do not leave the local area.

[0055] *When generating model updates, apply differential privacy (Differential Privacy) mechanism, add carefully calculated random noise to model parameters to blur the impact of individual training samples on the final model update, thus providing mathematically provable privacy protection.

[0056] *After encryption, model updates are transmitted to GSOCore through secure communication protocols such as TLS / SSL, ensuring confidentiality and integrity during transmission.

[0057] 2. Specific implementation of the Global Synaptic Optimization Core (GSOCore):

[0058] *Deployment: Deployed on a highly scalable cloud computing platform, using GPUs, TPUs, or future dedicated AI accelerator clusters

[0059] to provide powerful computing power.

[0060] *Federal aggregation and knowledge fusion: GSOCore receives encrypted and privacy-protected local model updates uploaded from FL-ESN around the world.

[0061] * Apply federated learning aggregation algorithms (e.g., FedAvg, FedProx, FedOpt, or more sophisticated distillation-based federated learning algorithms) to these distributed model updates under secure aggregation protocols (like secure multi-party computation, SMPC) to perform a weighted average or fusion of these model updates, generating a more robust and comprehensive global model. This process ensures that GSO Core cannot access individual edge node's raw model updates during the aggregation process.

[0062] * "Supply Chain Synaptic Network" Construction and Optimization: At the core of GSOCore is a dynamic Graph Neural Network (GNN) or large-scale knowledge graph, whose structure and behavior are inspired by principles of the brain.

[0063] * Neuron Layers: Nodes of the GNN represent supply neurons (Chinese SMEs and their products / technologies) and demand intent neurons (various explicit / implicit industrial product demand intents globally).

[0064] * Synaptic Connection Layers: Edges of the GNN represent "synaptic connections" between neurons, whose weights and activities are dynamically adjusted through the following mechanisms:

[0065] * Online Learning and Synaptic Plasticity: Combining the global learning outcomes of federated aggregation, as well as the central core's own analysis of macro data (such as global trade reports, industry research, economic indices), the system continuously adjusts synaptic weights. For example, if a specific implicit demand intent from a certain region is successfully matched to a certain SME multiple times and generates orders, the synaptic connection between that SME and that intent will be strengthened.

[0066] * Multi-factor Weights: Synaptic weights not only consider matching degree, but also include historical conversion rates, customer satisfaction, market competition intensity, SME cooperation degree, and other dynamic factors.

[0067] * Vertical Large Model (VLM) Integration for Supply Chain Domain: GSOCore integrates and runs a Vertical Large Model (VLM) for the industrial product supply chain domain. This VLM has:

[0068] * Deep understanding of industry-specific terminology, technical specifications, supply chain processes, and international trade rules.

[0069] * Receives global model parameters after federated aggregation and abstracted event features uploaded by FL-ESN.

[0070] *Deeper semantic understanding, complex reasoning and knowledge generation on the aggregated global knowledge (e.g. infer a country is forming a new industry chain opportunity for "special alloy materials" from seemingly unrelated events). *Assist in more refined supply chain matching optimization, e.g. recommend the most suitable supplier from multiple qualified SMEs that best fit the "implicit preferences" of a specific buyer (e.g. higher requirements for customization, lead time, specific certifications).

[0071] 3. Implementation of the adaptive supply chain matching and go-to-sea strategy generation module:

[0072] *Input: Receive "high-intensity demand intent pulses" (and their associated SMEs) from GSOCore output, as well as the latest optimized supply chain synapse network state.

[0073] *Matching decision: The system intelligently identifies the most suitable Chinese SME for a specific demand intent based on real-time triggered synapse connection strength, VLM deep analysis results, and pre-set business rules.

[0074] *Strategy generation: Generate a highly personalized and real-time multi-modal go-to-sea strategy package for the matched SME.

[0075] *Content automatic generation: Use the VLM's generation capabilities to automatically generate multi-language, multi-modal marketing content (such as intelligently adjusted independent station product descriptions, social media short video scripts, AR / VR product display scripts, email drafts) based on SME product characteristics and target buyer implicit intent.

[0076] *Channel and advertising optimization: Intelligently analyze market trends and buyer behavior provided by GSOCore to recommend the most effective promotion channels (such as specific industry vertical social media groups, emerging market local B2B platforms), and dynamically adjust advertising platforms, budget allocation, target audience and keywords.

[0077] *Lead pushing and insights: Identify and push high-priority, high-conversion potential overseas buyer leads, along with the buyer's

[0078] implicit intent analysis, behavior pattern report and recommended communication entry points.

[0079] Intelligent communication strategy: based on the context understanding ability of VLM, provide SMEs with intelligent outreach scripts, negotiation strategy suggestions, and professional knowledge points and dialogue processes preset in AI question and answer robots for this buyer and demand intention. Strategy optimization and closed-loop feedback: all market feedback after strategy execution (such as ad click rate, website dwell time, inquiry quantity, inquiry quality, conversion rate, customer satisfaction, order amount) as new "spatiotemporal event pulse", perceived by FL-ESN and returned to GSOCore, driving the continuous online learning and optimization of the whole system, realizing the closed-loop evolution that never stops.

[0080] Three, workflow:

[0081] 1. Multi-modal perception and local learning: FL-ESN in various parts of the world perceive local multi-modal spatiotemporal event streams in real time, encode them into pulse sequences, and use brain-inspired models on local privacy data for online learning to identify local implicit demand intention patterns.

[0082] 2. Privacy protection transmission: FL-ESN applies privacy protection technology to generate and encrypt local model updates.

[0083] 3. Federated aggregation and global optimization: GSOCore receives and securely aggregates local model updates, combines vertical large models to optimize the global "supply chain synapse network", and conducts macroscopic implicit demand intention reasoning.

[0084] 4. Matching and strategy generation: The adaptive supply chain matching and outbound strategy generation module accurately matches Chinese SMEs with overseas implicit demand intentions based on the optimized synapse network and global reasoning results, and adaptively generates customized outbound strategies.

[0085] 5. Strategy execution and feedback: SMEs execute outbound marketing activities according to the strategy, and market feedback is returned to FL-ESN and GSOCore as new spatiotemporal event pulses, driving the whole system to continuously learn and optimize online, realizing a closed loop.

Claims

1. A collaborative perception of implicit demand intentions for global industrial products and an adaptive supply chain synaptic optimization system based on federated brain-like learning, characterized by: include: a. Federated edge perception nodes, deployed as lightweight software modules or containerized applications in local overseas markets or in cloud environments controlled by overseas partners, with the following configurations: i. Used to perceive and receive local, multimodal, heterogeneous, and unstructured spatiotemporal event streams in real time; ii. for encoding the spatiotemporal event stream into a spatiotemporal event pulse sequence having a timestamp, intensity, feature vector and semantic label; iii. for performing online learning using a small neural network model inspired by brain-inspired techniques based on the spatiotemporal event pulse sequence and local private data to identify local-specific implicit demand intention patterns; as well as iv. Under the premise of ensuring that the local original data does not leave the local area, applying privacy protection technology to generate a local model update and transmitting the local model update via a secure communication protocol; b. Global synaptic optimization core, deployed in the central cloud system, configured as follows: i. Receive and aggregate local model updates from multiple federated edge-sensing nodes via a secure aggregation protocol to build and continuously optimize a global, dynamically adaptive "supply chain synaptic network," comprising supply neurons representing Chinese suppliers, demand intention neurons representing global industrial product demand intentions, and dynamic synaptic connections representing the degree of match and opportunity value between the supply neurons and the demand intention neurons. ii. Combined with a large vertical model for the supply chain, this platform conducts in-depth analysis of aggregated global knowledge and abstracted event features to uncover global implicit demand intentions and commercial value. c. Adaptive supply chain matching and overseas expansion strategy generation module, configured as follows: i. for identifying and activating the most matching supply neurons in real time based on the supply chain synaptic network formed by the global supply chain synaptic optimization core and the implicit demand intention; as well as ii. for adaptively generating and optimizing a multimodal outbound strategy corresponding to the activated supply neurons.

2. The system according to claim 1, wherein: The small brain-inspired neural network model inside the federated edge perception node utilizes synaptic plasticity mechanism for online learning.

3. The system according to claim 1, wherein: The spatiotemporal event stream includes but is not limited to text data stream, behavior data stream, image / video clips, and structured / semi-structured data stream.

4. The system according to claim 1, wherein: The privacy protection technologies applied by the federated edge perception node include but are not limited to differential privacy and homomorphic encryption.

5. The system according to claim 1, wherein: The secure communication protocol includes but is not limited to TLS / SSL encrypted channels.

6. The system according to claim 1, wherein: The secure aggregation protocols used by the global synaptic optimization core include but are not limited to secure multi-party computation.

7. The system according to claim 1, wherein: The adaptive overseas strategy includes but is not limited to customized digital marketing content, precise advertising recommendations, high-priority buyer lead push, and intelligent communication strategies.

8. The system according to claim 1, wherein: The adaptive supply chain matching and overseas strategy generation module is also configured to collect market feedback after the strategy is executed, and transmit the feedback as a new spatiotemporal event pulse back to the federated edge perception node and the global synaptic optimization core to drive continuous online learning and optimization of the entire system.

9. The system according to claim 1, wherein: The strength and activity of the dynamic synaptic connection are dynamically adjusted based on real-time market feedback, historical transaction success rates, and customer satisfaction.

10. A method for collaborative perception of implicit demand intentions of global industrial products and adaptive supply chain synaptic optimization based on federated brain-like learning, characterized by: The following steps are involved: a. Through federated edge perception nodes deployed in local or overseas partner cloud environments, real-time perception and reception of local, multimodal, heterogeneous, and unstructured spatiotemporal event streams; b. encoding the spatiotemporal event stream into a spatiotemporal event pulse sequence having a timestamp, intensity, feature vector, and semantic label; c. Based on the spatiotemporal event pulse sequence and local private data, a small neural network model inspired by brain-inspired learning is used for online learning to identify local, implicit demand intention patterns; d. Generating local model updates using privacy-preserving techniques while ensuring that the local raw data does not leave the local data store, and transmitting the local model updates to the central core via a secure communication protocol; e. Receive and aggregate local model updates from multiple federated edge perception nodes through a central global synaptic optimization core via a secure aggregation protocol. In combination with a large vertical model for the supply chain domain, the aggregated global knowledge is deeply analyzed to uncover global implicit demand intent and business value. f. Based on the supply chain synaptic network formed by the global synaptic optimization core and the implicit demand intention, identify and activate the most suitable Chinese supplier in real time; as well as g. Adaptively generate and optimize a multimodal global expansion strategy corresponding to the activated suppliers.

11. The method according to claim 10, characterized in that The privacy protection technology applied in step d includes but is not limited to differential privacy and homomorphic encryption.

12. The method according to claim 10, characterized in that The secure aggregation protocol in step e includes but is not limited to secure multi-party computation.

Images