A method and system for hotspot trend prediction and computing power sharing for multi-source heterogeneous data
By combining API gateways, Z-Score standardization, BERT-wwm and ResNet-50 models, ST-GNN, federated learning, and blockchain smart contracts, the problems of multi-source heterogeneous data fusion and cross-organizational computing power sharing are solved, enabling efficient hotspot trend prediction and resource optimization.
Patent Information
- Application Number
- CN202610031886.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-12
- Publication Date
- 2026-06-02
AI Technical Summary
Traditional ETL tools cannot efficiently handle the semantic relationships between unstructured and structured data; single-modal prediction models struggle to capture the spatiotemporal relationships of multi-dimensional data; and cloud computing centers suffer from low GPU utilization and lack dynamic scheduling mechanisms to achieve cross-organizational computing power sharing.
The system aggregates multi-source data using an API gateway, extracts features through Z-Score standardization and BERT-wwm and ResNet-50 models, uses ST-GNN for cross-modal fusion, constructs a federated learning prediction model, and employs blockchain smart contracts for shared computing power scheduling, combined with genetic algorithms to optimize resource allocation.
It achieves efficient fusion of multi-source heterogeneous data and real-time accurate prediction of hot trends, improves the utilization efficiency of computing resources, breaks down data silos, supports cross-organizational sharing, and has high scalability and data security.
Smart Images
Figure CN122133051A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of data processing and artificial intelligence technology, specifically providing a method and system for predicting hotspot trends and sharing computing power for multi-source heterogeneous data. Background Technology
[0002] Traditional ETL tools cannot handle the semantic relationships between unstructured data (such as video frames and natural language) and structured data, resulting in a fusion efficiency of less than 60%. Single-modal prediction models (such as ARIMA) have difficulty capturing the spatiotemporal relationships of multi-dimensional data, and the prediction delay can exceed 3 hours in sudden event scenarios.
[0003] The utilization rate of GPUs in cloud computing centers is often below 30%, and there is a lack of dynamic scheduling mechanisms to achieve cross-organizational sharing of computing power. Summary of the Invention
[0004] This invention addresses the shortcomings of the prior art by providing a highly practical method for predicting hotspot trends and sharing computing power for multi-source heterogeneous data.
[0005] A further technical objective of this invention is to provide a reasonably designed, safe, and applicable hotspot trend prediction and computing power sharing system for multi-source heterogeneous data.
[0006] The technical solution adopted by this invention to solve its technical problem is: A method for hotspot trend prediction and computing power sharing for multi-source heterogeneous data has the following steps: S1. Multi-source data acquisition; S2, Intelligent Preprocessing; S3, cross-modal fusion; S4, Hotspot Prediction; S5, computing power sharing and scheduling.
[0007] Furthermore, in step S1, social media APIs, industrial IoT MQTT streams, and database JDBC interfaces are aggregated through an API gateway to perform Kafka message queue buffering.
[0008] Furthermore, in step S2, Z-Score normalization and one-hot encoding are applied to the structured data; For unstructured data, the BERT-wwm model is used to extract text semantic vectors, and ResNet-50 is used to extract image feature hashes. For streaming data, window aggregation and outlier detection are implemented based on Flink.
[0009] Furthermore, in step S3, a spatiotemporal graph neural network ST-GNN is used, where nodes represent entities and events, edges represent temporal associations, and an attention mechanism is used to weight multimodal features, outputting a 300-dimensional fusion vector.
[0010] Furthermore, in step S4, a federated learning prediction model is constructed, including LSTM sub-models trained by each data owner and parameters aggregated by a central server, and the Secure Aggregation protocol is used to protect privacy. Output a heatmap showing the event's popularity index and propagation probability.
[0011] Furthermore, in step S5, based on the blockchain smart contract, the node declares CPU or GPU computing power and storage resources, uses a genetic algorithm to optimize task-resource allocation, and performs automatic payment and audit tracking based on ERC-20 tokens.
[0012] A device for predicting hotspot trends and sharing computing power for multi-source heterogeneous data firstly collects multi-source data, aggregates social media APIs, industrial IoT MQTT streams and database JDBC interfaces through an API gateway, and buffers Kafka message queues. Then, intelligent preprocessing, cross-modal fusion and hotspot prediction are performed, and finally, computing power sharing and scheduling are carried out.
[0013] Furthermore, during intelligent preprocessing, Z-Score normalization and one-hot encoding are used for structured data; For unstructured data, the BERT-wwm model is used to extract text semantic vectors, and ResNet-50 is used to extract image feature hashes. For streaming data, window aggregation and outlier detection are implemented based on Flink.
[0014] Furthermore, during cross-modal fusion, a spatiotemporal graph neural network ST-GNN is used, with nodes representing entities and events and edges representing temporal associations. An attention mechanism is employed to weight multimodal features, outputting a 300-dimensional fusion vector. When predicting hotspots, a federated learning prediction model is constructed, which includes LSTM sub-models trained by each data owner and parameters aggregated by a central server, and the Secure Aggregation protocol is used to protect privacy. Output a heatmap showing the event's popularity index and propagation probability.
[0015] Furthermore, during the sharing and scheduling of computing power, smart contracts based on blockchain are used to declare CPU or GPU computing power and storage resources for nodes, genetic algorithms are used to optimize task-resource allocation, and automatic payment and audit tracking are carried out based on ERC-20 tokens.
[0016] Compared with existing technologies, the hotspot trend prediction and computing power sharing method and system for multi-source heterogeneous data of the present invention have the following outstanding advantages: This invention achieves efficient fusion of multi-source heterogeneous data through cross-modal semantic alignment technology, effectively breaking down data silos and promoting deep integration and value mining of structured and unstructured data. The combination of a federated learning framework and a spatiotemporal graph neural network enables real-time and accurate prediction of hotspot trends, capturing the spatiotemporal correlations of multi-dimensional data and generating instructive prediction results, providing decision support for scenarios such as smart cities and financial risk control.
[0017] This blockchain-based smart contract-based computing power sharing platform significantly improves the utilization efficiency of computing resources through dynamic scheduling and optimized configuration. It supports cross-organizational and cross-platform sharing and trading of computing resources, reducing idle resource costs. The system adopts a modular architecture design, possessing high scalability and deployment flexibility, capable of adapting to data processing and prediction needs in different scenarios. Simultaneously, it ensures data security and compliance through privacy protection mechanisms, demonstrating broad technical applicability and promising industry application prospects. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a schematic diagram of a framework for a method of predicting hotspot trends and sharing computing power for multi-source heterogeneous data. Detailed Implementation
[0020] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0021] The following is a preferred embodiment: like Figure 1 As shown in the figure, a method for hotspot trend prediction and computing power sharing for multi-source heterogeneous data in this embodiment has the following steps: S1. Multi-source data acquisition; The API gateway aggregates 12 types of data sources, including social media APIs, industrial IoT MQTT streams, and database JDBC interfaces, and supports Kafka message queue buffering.
[0022] S2, Intelligent Preprocessing; Structured data: Z-Score normalization and one-hot encoding are used; Unstructured data: Text semantic vectors were extracted using the BERT-wwm model, and image feature hashes were extracted using ResNet-50; Streaming data: Implementing window aggregation and outlier detection based on Flink.
[0023] S3, cross-modal fusion; Design a space-time graph neural network (ST-GNN): Nodes: Entities (such as users, devices) and events; Edge: Temporal correlation (interaction frequency within time window Δt); Feature fusion: Multimodal features are weighted using an attention mechanism to output a 300-dimensional fusion vector.
[0024] S4, Hotspot Prediction; Building a federated learning prediction model: Local model: Each data owner trains an LSTM sub-model; Global model: The central server aggregates parameters and uses the Secure Aggregation protocol to protect privacy; Output: Event popularity index (0-100 points) and propagation probability heatmap.
[0025] S5, shared computing power scheduling; Implementation of smart contracts based on blockchain: Computing power registration: Nodes declare their CPU / GPU computing power and storage resources; Task matching: A genetic algorithm is used to optimize task-resource allocation, taking into account network latency and cost; Transaction settlement: Automated payments and audit trail based on ERC-20 tokens.
[0026] Based on the above method, the hotspot trend prediction and computing power sharing device for multi-source heterogeneous data in this embodiment firstly collects multi-source data, and aggregates social media APIs, industrial IoT MQTT streams and database JDBC interfaces through an API gateway, and performs Kafka message queue buffering. Then, intelligent preprocessing, cross-modal fusion and hotspot prediction are performed, and finally, computing power sharing and scheduling are carried out.
[0027] Among them, during intelligent preprocessing, Z-Score normalization and one-hot encoding are used for structured data; For unstructured data, the BERT-wwm model is used to extract text semantic vectors, and ResNet-50 is used to extract image feature hashes. For streaming data, window aggregation and outlier detection are implemented based on Flink.
[0028] For cross-modal fusion, a spatiotemporal graph neural network ST-GNN is used, with nodes representing entities and events and edges representing temporal associations. An attention mechanism is used to weight multimodal features, and a 300-dimensional fusion vector is output. When predicting hotspots, a federated learning prediction model is constructed, which includes LSTM sub-models trained by each data owner and parameters aggregated by a central server, and the Secure Aggregation protocol is used to protect privacy. Output a heatmap showing the event's popularity index and propagation probability.
[0029] During the shared scheduling of computing power, smart contracts based on blockchain are used to declare CPU or GPU computing power and storage resources for nodes, and genetic algorithms are used to optimize task-resource allocation. Automatic payment and audit tracking are carried out based on ERC-20 tokens.
[0030] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for hotspot trend prediction and computing power sharing for multi-source heterogeneous data, characterized in that, It has the following steps: S1. Multi-source data acquisition; S2, Intelligent Preprocessing; S3, cross-modal fusion; S4, Hotspot Prediction; S5, computing power sharing and scheduling.
2. The method for hotspot trend prediction and computing power sharing for multi-source heterogeneous data according to claim 1, characterized in that, In step S1, social media APIs, industrial IoT MQTT streams, and database JDBC interfaces are aggregated through an API gateway to perform Kafka message queue buffering.
3. The method for hotspot trend prediction and computing power sharing for multi-source heterogeneous data according to claim 2, characterized in that, In step S2, Z-Score normalization and one-hot encoding are used for structured data; For unstructured data, the BERT-wwm model is used to extract text semantic vectors, and ResNet-50 is used to extract image feature hashes. For streaming data, window aggregation and outlier detection are implemented based on Flink.
4. The method for hotspot trend prediction and computing power sharing for multi-source heterogeneous data according to claim 3, characterized in that, In step S3, a spatiotemporal graph neural network ST-GNN is used, with nodes representing entities and events and edges representing temporal associations. An attention mechanism is used to weight multimodal features, and a 300-dimensional fusion vector is output.
5. A method for hotspot trend prediction and computing power sharing for multi-source heterogeneous data according to claim 4, characterized in that, In step S4, a federated learning prediction model is constructed, including LSTM sub-models trained by each data owner and parameters aggregated by a central server, and the Secure Aggregation protocol is used to protect privacy. Output a heatmap showing the event's popularity index and propagation probability.
6. The method for hotspot trend prediction and computing power sharing for multi-source heterogeneous data according to claim 5, characterized in that, In step S5, based on the blockchain smart contract, the node declares CPU or GPU computing power and storage resources, uses a genetic algorithm to optimize task-resource allocation, and performs automatic payment and audit tracking based on ERC-20 tokens.
7. A device for hotspot trend prediction and computing power sharing for multi-source heterogeneous data, characterized in that, First, multi-source data is collected, and social media APIs, industrial IoT MQTT streams, and database JDBC interfaces are aggregated through an API gateway for Kafka message queue buffering. Then, intelligent preprocessing, cross-modal fusion and hotspot prediction are performed, and finally, computing power sharing and scheduling are carried out.
8. A hotspot trend prediction and computing power sharing device for multi-source heterogeneous data according to claim 7, characterized in that, When performing intelligent preprocessing, Z-Score normalization and one-hot encoding are used for structured data; For unstructured data, the BERT-wwm model is used to extract text semantic vectors, and ResNet-50 is used to extract image feature hashes. For streaming data, window aggregation and outlier detection are implemented based on Flink.
9. A hotspot trend prediction and computing power sharing device for multi-source heterogeneous data according to claim 8, characterized in that, For cross-modal fusion, a spatiotemporal graph neural network ST-GNN is used, with nodes representing entities and events and edges representing temporal associations. An attention mechanism is used to weight multimodal features, and a 300-dimensional fusion vector is output. When predicting hotspots, a federated learning prediction model is constructed, which includes LSTM sub-models trained by each data owner and parameters aggregated by a central server, and the Secure Aggregation protocol is used to protect privacy. Output a heatmap showing the event's popularity index and propagation probability.
10. A hotspot trend prediction and computing power sharing device for multi-source heterogeneous data according to claim 9, characterized in that, During the shared scheduling of computing power, smart contracts based on blockchain are used to declare CPU or GPU computing power and storage resources for nodes, and genetic algorithms are used to optimize task-resource allocation. Automatic payment and audit tracking are carried out based on ERC-20 tokens.