AI-assisted new energy data element intelligent mining method and system
The AI-assisted intelligent mining system for new energy data elements, utilizing multimodal perception and hierarchical federated governance modules, solves the data silo problem in new energy data mining, achieves efficient data quality assessment and dynamic benefit distribution, and improves the accuracy and efficiency of new energy data mining.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DONGFANG GREEN ENERGY (HEBEI) CO LTD
- Filing Date
- 2025-08-07
- Publication Date
- 2026-04-24
AI Technical Summary
Currently, new energy data mining faces serious data silos, with power grid, meteorological, and equipment data belonging to different entities, and traditional federated learning ignores the problem of data quality differences.
An AI-assisted intelligent mining system for new energy data elements is adopted, including a multimodal perception module, a hierarchical federated governance module, a physical constraint causal mining module, and a dynamic optimization decision-making module. By deploying satellite remote sensing terminals, UAV infrared imaging units, and acoustic sensor arrays, a spatiotemporal fusion graph neural network is constructed, a dynamic adaptive weight allocation mechanism is designed, and a lightweight local model and blockchain rights confirmation are combined to achieve privacy-preserving aggregation and global collaborative training of cross-domain model parameters.
It solves the problems of data silos and missing mechanisms, optimizes the efficiency of global model aggregation, improves the generalization ability of small samples, dynamically allocates benefits according to data quality and contribution, forms a trustworthy federated intelligent agent, and provides a high-precision multi-source heterogeneous perception foundation.
Smart Images
Figure CN121920494A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of new energy data mining technology, and more specifically, to an AI-assisted intelligent mining method and system for new energy data elements. Background Technology
[0002] AI-assisted intelligent mining of new energy data elements is using cutting-edge technologies such as multimodal fusion, causal reasoning, and federated learning to extract high-value information from massive heterogeneous data, driving the new energy industry to leap from "experience-based decision-making" to "data-driven" approaches. New energy data mining is a key technology that uses machine learning, deep learning, and other techniques to extract patterns, predict trends, and optimize decisions from massive amounts of new energy data. Its core applications include three major areas: smart grid management, new energy vehicle R&D and operation and maintenance, and wind and solar power plant efficiency improvement.
[0003] Currently, new energy data mining faces a technical bottleneck: severe data silos. Data from power grids, meteorology, and equipment belong to different entities, and traditional federated learning with equal weights ignores differences in data quality.
[0004] Therefore, an AI-assisted intelligent mining method and system for new energy data elements is proposed to solve the above problems. Summary of the Invention
[0005] The purpose of this invention is to address the shortcomings of existing technologies: current new energy data mining faces a technical bottleneck, namely, severe data silos, with power grid, meteorological, and equipment data belonging to different entities, and traditional federated learning and equal weight aggregation ignoring differences in data quality. Therefore, this invention proposes an AI-assisted intelligent mining method and system for new energy data elements.
[0006] The specific technical solution is: an AI-assisted intelligent mining system for new energy data elements, including:
[0007] 1) Multimodal sensing module
[0008] Deploy satellite remote sensing terminals, UAV infrared imaging units, and acoustic sensor arrays to collect three-dimensional turbulence maps of wind farms, images of microcracks in photovoltaic modules, and acoustic signals of equipment operation;
[0009] 2) Hierarchical Federal Governance Module
[0010] The hierarchical federated governance module constructs a heterogeneous data alignment layer based on a spatiotemporal fusion graph neural network. It uses spatiotemporal embedding, including representation learning, to uniformly calibrate the timestamps and spatial coordinate systems of multi-source heterogeneous data from satellites, UAVs, and ground sensors. At the federated architecture layer, a dynamic adaptive weight allocation mechanism is designed. Lightweight local models are distributed and deployed in power data centers in various provinces. Relying on a lightweight encrypted channel, privacy-preserving aggregation of cross-domain model parameters and global collaborative training are achieved, forming a trustworthy federated intelligent agent that balances spatiotemporal consistency, data sovereignty compliance, and computational efficiency.
[0011] 3) Physical Constraint Causal Mining Module
[0012] Hard-constrained units for the dynamic equations are injected into the LSTM loss function:
[0013]
[0014] A multi-level causal inferencer executes PC algorithms, counterfactual quantization, and physical GNN embeddings.
[0015] 4) Dynamic optimization decision-making module
[0016] The reinforcement learning agent generates energy storage charging and discharging strategies, and combines digital twin simulation to verify peak-valley arbitrage profits; the blockchain-based rights confirmation unit records the contribution of data elements and distributes profits based on smart contracts.
[0017] The detailed technical solution involves the multimodal perception module integrating a high-resolution satellite remote sensing terminal, an unmanned aerial vehicle (UAV) inspection system equipped with an infrared thermal imaging unit, and an array-type acoustic signature sensor network. This allows for the three-dimensional spatial turbulence distribution map of the wind farm area, images of microcracks and hot spot defects inside photovoltaic modules, and acoustic signature signal characteristics generated during the operation of various key equipment. This constructs a comprehensive perception data system covering multiple dimensions (air, space, and ground) and multiple physical fields (light, heat, and sound), providing a high-precision and multi-source heterogeneous perception foundation for subsequent intelligent diagnosis and analysis.
[0018] The detailed technical solution involves designing a dynamic adaptive weight allocation mechanism within the hierarchical federated governance module, based on real-time calculation of contributions according to node data quality, model accuracy, and network status.
[0019] In the detailed technical solution, within the hierarchical federated governance module, the dynamic weight calculation unit generates aggregate weights according to the following formula:
[0020]
[0021] Homomorphic encryption transmission module.
[0022] In detail, the Federated Learning Framework (FedEE) employs a dynamic weight allocation mechanism within the hierarchical federated governance module, specifically including:
[0023] Calculate the data quality score for each local model:
[0024]
[0025] The aggregation weights are dynamically adjusted based on $Q_i$, allowing high-quality data nodes to dominate global model updates.
[0026] The detailed technical solution includes a physical constraint causal mining module that integrates the DoWhy causal discovery framework to analyze the fault chain of new energy equipment and output the causal effect value β and significance p value; it also constructs a cross-modal Transformer model, integrates acoustic fingerprint, vibration, and infrared image data, and identifies hidden cracks in the blades.
[0027] The detailed technical solution is that the dynamic optimization decision module is deployed with a reinforcement learning agent based on the near-end policy optimization algorithm. By deeply sensing the real-time electricity price fluctuations, load demand and renewable energy output of the power grid, it dynamically generates multi-timescale charging and discharging strategies for the energy storage system.
[0028] Simultaneously connect to a high-fidelity digital twin simulation platform to verify the peak-valley arbitrage economic benefits and equipment lifespan loss of the strategy in a virtual environment, and achieve closed-loop optimization of the strategy.
[0029] Simultaneously, it integrates a blockchain-based trusted incentive engine, which uses smart contracts to automatically parse the contribution of multimodal sensing data, including satellite remote sensing data quality and federated model parameter accuracy, to construct a data element and economic benefit mapping chain for accurate weighting and automated allocation of revenue, forming a sustainable energy economic closed loop of strategy optimization, simulation verification, and contribution confirmation.
[0030] Another objective of this invention is to provide an AI-assisted intelligent mining method for new energy data elements, applied to the aforementioned AI-assisted intelligent mining system for new energy data elements, comprising the following steps:
[0031] Step 1: Multi-source data perception and edge preprocessing
[0032] A 50m×50m grid turbulence intensity map is generated by LiDAR scanning to mark high-risk areas of sudden wind speed changes; the infrared imaging unit of the UAV identifies hot spots on the photovoltaic module, and the IV curve diagnostic module locates the failed battery cell.
[0033] Step 2: Cross-domain knowledge fusion under privacy protection
[0034] Construct a four-dimensional spatiotemporal knowledge graph of resources, equipment, power grid, and carbon emissions, and use GNN to encode node relationships:
[0035]
[0036] Step 3: Causal-Driven Factor Value Mining
[0037] The PC algorithm is applied to identify the causal structure of variables from SCADA data and output a DAG graph; the fault propagation time sequence is determined based on the Granger causality test.
[0038] Step 4: Dynamic Decision-Making and Blockchain Evidence Storage
[0039] Deep Q-Network (DQN) generates a green electricity trading strategy: $Q(s,a)\leftarrow Q(s,a)+\alpha[r+\gamma\max{a'}Q(s',a')-Q(s,a)]$;
[0040] Write data usage records to the blockchain and distribute revenue according to smart contracts:
[0041]
[0042] In the detailed technical solution, step three, based on the Granger causality test, determines the fault propagation sequence, including bearing temperature and vibration anomalies.
[0043] In the detailed technical solution, step three, causal-driven element value mining, includes causal discovery and value quantification, which encompass:
[0044] The DoWhy framework was used to calculate the average treatment effect (ATE) of photovoltaic penetration on voltage fluctuations:
[0045]
[0046] Where $X$ represents the photovoltaic penetration rate tier and $Y$ represents the voltage deviation rate; when ATE> the threshold $\delta$ and p<0.05, a power grid renovation early warning is triggered.
[0047] Compared with the prior art, the present invention can achieve the following:
[0048] 1. A dual-engine architecture integrating hierarchical federated governance and physical mechanism embedding addresses the issues of data silos and missing mechanisms; dynamically optimizes the decision-making module to quantitatively evaluate node data quality and optimizes the global model aggregation efficiency; uses equipment physical equations as regularization terms to constrain AI training, improving small-sample generalization ability; and dynamically allocates benefits based on data quality and contribution to activate factor flow.
[0049] 2. It integrates three major technology stacks: spatiotemporal AI, federated learning, and privacy computing. It not only clarifies how to achieve ST-GNN alignment, dynamic weights, and encrypted aggregation, but also defines a trusted federated intelligent agent, forming a closed-loop technology narrative.
[0050] 3. Constructing a comprehensive sensing data system covering multiple dimensions including air, space, and ground, as well as multiple physical fields such as light, heat, and sound, clearly highlights the core value of this module. It integrates different spatial levels and different physical information to form a powerful comprehensive data foundation; it provides a high-precision and multi-source heterogeneous sensing foundation for subsequent intelligent diagnosis and analysis, clearly indicates that this module serves upper-level applications, and emphasizes the characteristics of the data. Attached Figure Description
[0051] Figure 1 This is a block diagram of the AI-assisted intelligent mining system for new energy data elements of the present invention. Detailed Implementation
[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.
[0053] In this embodiment of the invention, the AI-assisted intelligent mining system for new energy data elements includes:
[0054] 1) Multimodal sensing module
[0055] Deploy satellite remote sensing terminals, UAV infrared imaging units, and acoustic sensor arrays to collect three-dimensional turbulence maps of wind farms, images of microcracks in photovoltaic modules, and acoustic signals of equipment operation;
[0056] 2) Hierarchical Federal Governance Module
[0057] The hierarchical federated governance module constructs a heterogeneous data alignment layer based on a spatiotemporal fusion graph neural network. It uses spatiotemporal embedding, including representation learning, to uniformly calibrate the timestamps and spatial coordinate systems of multi-source heterogeneous data from satellites, UAVs, and ground sensors. At the federated architecture layer, a dynamic adaptive weight allocation mechanism is designed. Lightweight local models are distributed and deployed in power data centers in various provinces. Relying on a lightweight encrypted channel, privacy-preserving aggregation of cross-domain model parameters and global collaborative training are achieved, forming a trustworthy federated intelligent agent that balances spatiotemporal consistency, data sovereignty compliance, and computational efficiency.
[0058] By integrating three major technology stacks—spatiotemporal AI, federated learning, and privacy computing—it not only clarifies how to achieve ST-GNN alignment, dynamic weights, and encrypted aggregation, but also defines a trusted federated intelligent agent, forming a closed-loop technology narrative.
[0059] 3) Physical Constraint Causal Mining Module
[0060] Hard-constrained units for the dynamic equations are injected into the LSTM loss function:
[0061]
[0062] A multi-level causal inferencer executes PC algorithms, counterfactual quantization, and physical GNN embeddings.
[0063] 4) Dynamic optimization decision-making module
[0064] The reinforcement learning agent generates energy storage charging and discharging strategies, and combines digital twin simulation to verify peak-valley arbitrage profits; the blockchain-based rights confirmation unit records the contribution of data elements and distributes profits based on smart contracts.
[0065] Therefore, a dual-engine architecture embedding hierarchical federation governance and physical mechanisms addresses the issues of data silos and missing mechanisms; dynamically optimizes the decision-making module to quantitatively evaluate node data quality and optimizes the global model aggregation efficiency; uses equipment physical equations as regularization terms to constrain AI training and improve small-sample generalization ability; and dynamically allocates benefits based on data quality and contribution to activate the flow of factors.
[0066] In this embodiment of the invention, the multimodal perception module integrates and deploys a high-resolution satellite remote sensing terminal, a drone inspection system equipped with an infrared thermal imaging unit, and an array-type acoustic fingerprint sensor network to collect three-dimensional spatial turbulence distribution maps of wind farm areas, images of microcracks and hot spot defects inside photovoltaic modules, and acoustic fingerprint signal features generated during the operation of various key equipment. This constructs a comprehensive perception data system covering multiple dimensions of air, space, and ground, as well as multiple physical fields of light, heat, and sound, providing a high-precision and multi-source heterogeneous perception foundation for subsequent intelligent diagnosis and analysis.
[0067] Therefore, constructing a comprehensive sensing data system covering multiple dimensions including air, space, and ground, as well as multiple physical fields such as light, heat, and sound, clearly highlights the core value of this module. It integrates different spatial levels (satellite-space, UAV-air, ground sensor-ground) and different physical information (light-satellite image / visible light defect, heat-infrared thermal image, sound-acoustic pattern) to form a powerful comprehensive data foundation. This provides a high-precision and multi-source heterogeneous sensing foundation for subsequent intelligent diagnosis and analysis, clearly indicating that this module serves upper-level applications (intelligent diagnosis and analysis), and emphasizing the characteristics of the data (high precision, multi-source, heterogeneous).
[0068] In this embodiment of the invention, a dynamic adaptive weight allocation mechanism is designed in the hierarchical federated governance module based on the contribution of node data quality, model accuracy and network status in real time.
[0069] In the detailed technical solution, within the hierarchical federated governance module, the dynamic weight calculation unit generates aggregate weights according to the following formula:
[0070]
[0071] Homomorphic encryption transmission module.
[0072] In this embodiment of the invention, the Federated Learning Framework (FedEE) employs a dynamic weight allocation mechanism in the hierarchical federated governance module, specifically including:
[0073] Calculate the data quality score for each local model:
[0074]
[0075] The aggregation weights are dynamically adjusted based on $Q_i$, allowing high-quality data nodes to dominate global model updates.
[0076] In this embodiment of the invention, the physical constraint causal mining module integrates the DoWhy causal discovery framework to analyze the fault chain of new energy equipment and output the causal effect value β and the significance p value; it constructs a cross-modal Transformer model, integrates acoustic print, vibration and infrared image data, and identifies hidden cracks in the blades.
[0077] In this embodiment of the invention, the dynamic optimization decision module deploys a reinforcement learning agent based on a near-end policy optimization algorithm, which dynamically generates multi-timescale charging and discharging strategies for the energy storage system by deeply sensing real-time electricity price fluctuations, load demand, and renewable energy output.
[0078] Simultaneously connect to a high-fidelity digital twin simulation platform to verify the peak-valley arbitrage economic benefits and equipment lifespan loss of the strategy in a virtual environment, and achieve closed-loop optimization of the strategy.
[0079] Simultaneously, it integrates a blockchain-based trusted incentive engine, which uses smart contracts to automatically parse the contribution of multimodal sensing data, including satellite remote sensing data quality and federated model parameter accuracy, to construct a data element and economic benefit mapping chain for accurate weighting and automated allocation of revenue, forming a sustainable energy economic closed loop of strategy optimization, simulation verification, and contribution confirmation.
[0080] In this embodiment of the invention, an AI-assisted intelligent mining method for new energy data elements, applied to the aforementioned AI-assisted intelligent mining system for new energy data elements, includes the following steps:
[0081] Step 1: Multi-source data perception and edge preprocessing
[0082] A 50m×50m grid turbulence intensity map is generated by LiDAR scanning to mark high-risk areas of sudden wind speed changes; the infrared imaging unit of the UAV identifies hot spots on the photovoltaic module, and the IV curve diagnostic module locates the failed battery cell.
[0083] Step 2: Cross-domain knowledge fusion under privacy protection
[0084] Construct a four-dimensional spatiotemporal knowledge graph of resources, equipment, power grid, and carbon emissions, and use GNN to encode node relationships:
[0085]
[0086] Step 3: Causal-Driven Factor Value Mining
[0087] The PC algorithm is applied to identify the causal structure of variables from SCADA data and output a DAG graph; the fault propagation time sequence is determined based on the Granger causality test.
[0088] Step 4: Dynamic Decision-Making and Blockchain Evidence Storage
[0089] Deep Q-Network (DQN) generates a green electricity trading strategy: $Q(s,a)\leftarrow Q(s,a)+\alpha[r+\gamma\max{a'}Q(s',a')-Q(s,a)]$;
[0090] Write data usage records to the blockchain and distribute revenue according to smart contracts:
[0091]
[0092] In this embodiment of the invention, in step three, the fault propagation sequence is determined based on the Granger causality test, including abnormal bearing temperature vibration.
[0093] In this embodiment of the invention, in the causal-driven element value mining of step three, causal discovery and value quantification include:
[0094] The DoWhy framework was used to calculate the average treatment effect (ATE) of photovoltaic penetration on voltage fluctuations:
[0095]
[0096] Where $X$ represents the photovoltaic penetration rate tier and $Y$ represents the voltage deviation rate; when ATE> the threshold $\delta$ and p<0.05, a power grid renovation early warning is triggered.
[0097] In this embodiment of the invention, the value of the PPO algorithm is emphasized by its ability to handle multi-timescale strategies (adapting to the needs of second-level control and hour-level arbitrage); the role of digital twins is highlighted by verifying the dual objectives (economic benefits + equipment lifespan loss), emphasizing the closed-loop optimization mechanism; the blockchain upgrade evolves the "rights confirmation unit" into a trusted incentive engine, clarifying the input sources of smart contracts (satellite data quality, model parameter accuracy); and the construction of a data element-economic benefit mapping chain directly addresses how blockchain can transform technological contributions (data / models) into economic value.
[0098] The sustainable energy economy closed loop is extracted, and the entire value chain of "strategy generation → simulation verification → contribution confirmation → revenue distribution" is summarized.
[0099] In this embodiment of the invention, "Spatiotemporal Fusion Graph Neural Network (ST-GNN)" and "Embedded Representation Learning" are explicitly used to illustrate that spatiotemporal correlation features are automatically learned through deep learning (replacing simple timestamp / coordinate matching); the evaluation dimensions (data quality, model accuracy, network status) are refined, emphasizing the dynamic nature of "real-time computation contribution"; a "heterogeneous data alignment layer" and a "federated architecture layer" are divided to reflect modular design; and the local model is limited to "lightweight" to adapt to edge-side resource constraints.
[0100] In this embodiment of the invention, the satellite remote sensing terminal emphasizes "high resolution" and points out the fusion of "optical and SAR imaging" to explain its capabilities (optical imaging can see the surface, while SAR can penetrate clouds and monitor deformation, etc.).
[0101] The drone inspection system is clearly defined as a "system" and its core unit is "infrared thermal imaging," highlighting its ability to detect "hot spot defects" (which is key to photovoltaic inspection).
[0102] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. An AI-assisted intelligent data element mining system for new energy, characterized in that, include: 1) Multimodal sensing module, Deploy satellite remote sensing terminals, UAV infrared imaging units, and acoustic sensor arrays to collect three-dimensional turbulence maps of wind farms, images of microcracks in photovoltaic modules, and acoustic signals of equipment operation; 2) Hierarchical Federal Governance Module The hierarchical federated governance module constructs a heterogeneous data alignment layer based on a spatiotemporal fusion graph neural network. It uses spatiotemporal embedding, including representation learning, to uniformly calibrate the timestamps and spatial coordinate systems of multi-source heterogeneous data from satellites, drones, and ground sensors. At the federated architecture layer, a dynamic adaptive weight allocation mechanism is designed, and lightweight local models are distributed and deployed in power data centers in various provinces. Relying on lightweight encrypted channels, privacy-preserving aggregation of cross-domain model parameters and global collaborative training are achieved, forming a trustworthy federated intelligent agent that takes into account spatiotemporal consistency, data sovereignty compliance and computational efficiency. 3) Physical Constraint Causal Mining Module Hard-constrained units for the dynamic equations are injected into the LSTM loss function: A multi-level causal inferencer executes PC algorithms, counterfactual quantization, and physical GNN embeddings. 4) Dynamic optimization decision-making module Reinforcement learning agents generate energy storage charging and discharging strategies, and combine digital twin simulations to verify peak-valley arbitrage profits; The blockchain-based rights confirmation unit records the contribution of data elements and distributes benefits based on smart contracts.
2. The AI-assisted intelligent mining system for new energy data elements according to claim 1, characterized in that, The multimodal perception module integrates and deploys a high-resolution satellite remote sensing terminal, an unmanned aerial vehicle (UAV) inspection system equipped with an infrared thermal imaging unit, and an array-type acoustic fingerprint sensor network. It collects three-dimensional spatial turbulence distribution maps of wind farm areas, images of microcracks and hot spot defects inside photovoltaic modules, and acoustic fingerprint signal characteristics generated during the operation of various key equipment. This constructs a comprehensive perception data system covering multiple dimensions of air, space, and ground, as well as multiple physical fields of light, heat, and sound, providing a high-precision and multi-source heterogeneous perception foundation for subsequent intelligent diagnosis and analysis.
3. The AI-assisted intelligent mining system for new energy data elements according to claim 1, characterized in that, In the hierarchical federated governance module, a dynamic adaptive weight allocation mechanism is designed based on the contribution of node data quality, model accuracy, and network status to calculate in real time.
4. The AI-assisted intelligent mining system for new energy data elements according to claim 1, characterized in that, In the hierarchical federated governance module, the dynamic weight calculation unit generates aggregate weights according to the formula: Homomorphic encryption transmission module.
5. The AI-assisted intelligent mining system for new energy data elements according to claim 1, characterized in that, In the hierarchical federated governance module, the Federated Learning Framework (FedEE) employs a dynamic weight allocation mechanism, specifically including: Calculate the data quality score for each local model: The aggregation weights are dynamically adjusted based on $Q_i$, allowing high-quality data nodes to dominate global model updates.
6. The AI-assisted intelligent mining system for new energy data elements according to claim 1, characterized in that, The physical constraint causal mining module integrates the DoWhy causal discovery framework to analyze the fault chain of new energy equipment and output the causal effect value β and significance p value; it constructs a cross-modal Transformer model, integrates acoustic print, vibration and infrared image data, and identifies hidden cracks in the blades.
7. The AI-assisted intelligent mining system for new energy data elements according to claim 1, characterized in that, The dynamic optimization decision-making module deploys a reinforcement learning agent based on the near-end policy optimization algorithm. By deeply sensing the real-time electricity price fluctuations, load demand and renewable energy output of the power grid, it dynamically generates multi-timescale charging and discharging strategies for the energy storage system. Synchronously connect to a high-fidelity digital twin simulation platform to verify the peak-valley arbitrage economic benefits and equipment lifespan loss of the strategy in a virtual environment, and achieve closed-loop optimization of the strategy. Simultaneously, it integrates a blockchain-based trusted incentive engine, which uses smart contracts to automatically parse the contribution of multimodal sensing data, including satellite remote sensing data quality and federated model parameter accuracy, to construct a data element and economic benefit mapping chain for accurate weighting and automated allocation of revenue, forming a sustainable energy economic closed loop of strategy optimization, simulation verification, and contribution confirmation.
8. An AI-assisted intelligent mining method for new energy data elements, applied to the AI-assisted intelligent mining system for new energy data elements as described in any one of claims 1-7, characterized in that, Includes the following steps: Step 1: Multi-source data perception and edge preprocessing A 50m×50m grid turbulence intensity map is generated by LiDAR scanning to mark high-risk areas of sudden wind speed changes; the infrared imaging unit of the UAV identifies hot spots on the photovoltaic module, and the IV curve diagnostic module locates the failed battery cell. Step 2: Cross-domain knowledge fusion under privacy protection Construct a four-dimensional spatiotemporal knowledge graph of resources, equipment, power grid, and carbon emissions, and use GNN to encode node relationships: Step 3: Causal-Driven Factor Value Mining The PC algorithm is applied to identify the causal structure of variables from SCADA data and output a DAG graph; the fault propagation time sequence is determined based on the Granger causality test. Step 4: Dynamic Decision-Making and Blockchain Evidence Storage Deep Q-Network (DQN) generates a green electricity trading strategy: $Q(s,a)\leftarrow Q(s,a)+\alpha[r+\gamma\max{a'}Q(s',a')-Q(s,a)]$; Write data usage records to the blockchain and distribute revenue according to smart contracts:
9. The AI-assisted intelligent mining method for new energy data elements according to claim 8, characterized in that, In step three, the fault propagation timeline is determined based on the Granger causality test, including abnormal bearing temperature and vibration.
10. The AI-assisted intelligent mining method for new energy data elements according to claim 8, characterized in that, In the causal-driven element value mining of step three, causal discovery and value quantification include: The DoWhy framework was used to calculate the average treatment effect (ATE) of photovoltaic penetration on voltage fluctuations: Where $X$ represents the photovoltaic penetration rate tier and $Y$ represents the voltage deviation rate; when ATE> the threshold $\delta$ and p<0.05, a power grid renovation early warning is triggered.