A rural micro-plant integrated management method of electricity and carbon

Abstract

This invention discloses an integrated management and control method for electricity and carbon emissions in rural micro-energy stations, belonging to the field of intelligent regulation and control technology for integrated energy systems. This invention relies on a three-branch decoupled variational autoencoder and a cGAN hierarchical completion algorithm to achieve high-precision preprocessing of energy data, and combines ST-GCN, XGBoost, and causal inference algorithms to complete minute-level dynamic carbon accounting. It achieves on-chain trusted ownership confirmation by improving the IPBFT consensus mechanism and optimizing the Shapley algorithm, and adopts a two-layer reinforcement learning architecture to complete multi-energy distributed collaborative scheduling. Combined with digital twin simulation verification, national cryptographic encryption fault-tolerant control, and Bayesian parameter iterative optimization, a full-link closed-loop intelligent management and control system is constructed. This invention effectively improves the data processing accuracy, carbon accounting accuracy, and scheduling stability of rural micro-energy stations, while taking into account data privacy. It is suitable for the efficient and low-carbon operation of rural distributed energy systems and has extremely high engineering application value.

Description

Technical Field

[0001] This invention belongs to the field of intelligent regulation and control technology for integrated energy systems, specifically relating to an integrated management and control method for electricity and carbon in rural micro-energy stations. It primarily targets multi-source heterogeneous energy equipment clusters in rural areas, including distributed photovoltaic, energy storage, biomass, and biogas systems. By combining deep learning data processing, spatiotemporal graph carbon accounting, multi-agent reinforcement learning distributed scheduling, improved blockchain consensus storage, and digital twin closed-loop iterative technology, it achieves multi-dimensional collaborative intelligent management and control of electricity, carbon, heat, and gas in rural micro-energy stations. This method is applicable to scenarios involving high-precision carbon accounting, privacy-optimized scheduling, on-chain trusted ownership confirmation, and autonomous fault-tolerant operation in rural distributed integrated energy systems. Background Technology

[0002] Driven by dual-carbon goals and the construction of new power systems, rural micro-energy stations integrating distributed photovoltaics, energy storage, biomass cogeneration, biogas systems, and agricultural loads are core infrastructure for achieving low-carbon energy transformation in rural areas and revitalizing rural clean energy resources. With the continuous expansion of the national carbon market, the accurate quantification of carbon assets, multi-energy coordinated and optimized scheduling, reliable rights confirmation and traceability, and market-based compliance management in distributed rural energy scenarios have become core technical challenges that the industry urgently needs to overcome.

[0003] Currently, existing integrated energy-electricity-carbon coupling regulation, distributed carbon data storage, and optimized scheduling technologies are mostly designed for urban parks and large-scale interconnected microgrids. They are not adapted to the hardware operation characteristics of rural micro-energy stations, which are characterized by strong data heterogeneity, large seasonal load fluctuations, weak edge computing power, and unstable biomass feedstock properties. These technologies have many technical shortcomings and cannot meet the specific regulation needs of rural scenarios. Specific technical shortcomings are as follows: First, the energy, environmental, and load data collected by rural micro-energy stations are multimodal, heterogeneous, and have severe feature coupling and large amounts of missing data. Existing technologies only use basic data standardization processing methods and lack specific feature decoupling and hierarchical completion algorithms. Data errors are propagated at each level, leading to a significant decrease in the accuracy of backend accounting and scheduling systems. Second, traditional carbon accounting technologies use fixed emission factors and manually corrected parameters, lacking dynamic adaptive algorithm models. This makes it impossible to match the time-series fluctuation characteristics of rural biomass feedstock and grid marginal factors, resulting in large accounting errors and failing to meet the technical standards for high-precision carbon verification. Third, existing scheduling technologies have structural limitations. The existing technologies suffer from several structural defects. First, centralized scheduling architectures pose a risk of data leakage. Second, traditional distributed optimization algorithms are slow to converge and prone to getting trapped in local optima, making them unsuitable for the drastic seasonal load fluctuations in rural areas. Third, day-ahead scheduling lacks real-time performance and optimization accuracy. Fourth, traditional consortium blockchain consensus algorithms have high computational overhead and long response latency, failing to adapt to the low-computing-power hardware conditions of rural edge nodes. This results in asynchronous energy and carbon flow data, discrepancies between records and actual data, and an inability to achieve rapid and reliable notarization and confirmation of small-value, high-frequency carbon assets. Fifth, existing technologies lack a dedicated clustering adaptation mechanism for rural scenarios and have not optimized algorithm parameters for specific operating conditions such as seasonal agricultural production loads and on-site disposal of biomass waste. This leads to poor adaptability and significant performance degradation after hardware deployment. Sixth, existing technologies only optimize algorithms for single modules such as data processing, carbon accounting, scheduling, and confirmation of rights, lacking a comprehensive, integrated technical architecture covering data preprocessing, carbon accounting, collaborative scheduling, on-chain notarization, and closed-loop iteration. This prevents the automated, intelligent, closed-loop iterative upgrade of rural micro-energy station carbon management systems.

[0004] In summary, the current industry lacks an integrated intelligent management and control solution for electricity and carbon emissions that is suitable for rural micro-energy station scenarios and covers the entire process. This restricts the low-carbon, digital, and market-oriented development of rural distributed energy. Therefore, developing an integrated management and control method for electricity and carbon emissions in rural micro-energy stations that is closed-loop, highly accurate, highly reliable, and highly adaptable has significant practical significance and application value. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, this invention provides a method for integrated electricity and carbon management of rural micro-energy stations, comprising the following closed-loop steps: Step 1: Collect multimodal energy data, extract features through a three-branch decoupled variational autoencoder and constrain mutual information to achieve feature decoupling, use conditional generative adversarial network (cGAN) to hierarchically complete missing data, and screen and verify qualified data as a unified computing basis; Step 2: Dynamically calculate the carbon emission reduction of new energy power generation substitution based on Spatial-Temporal Graph Convolutional Networks (ST-GCN), calculate the carbon emission reduction corresponding to waste treatment based on Extreme Gradient Boosting Tree (XGBoost), dynamically generate deduplication coefficients using causal inference algorithm, and subtract the total life cycle carbon emissions to obtain the net carbon emission reduction. Step 3: Construct dual ledgers for energy flow and carbon flow, and simultaneously upload them to the blockchain. Use the improved IPBFT consensus to complete distributed notarization, and combine entropy weight-TOPSIS to correct the Shapley value to automatically allocate carbon asset rights of each entity. Step 4: Employ a two-layer reinforcement learning approach to perform co-optimization of electricity, carbon, heat, and gas. The upper layer uses the Deep Deterministic Policy Gradient (DDPG) algorithm framework to solve the monthly planning, while the lower layer uses the Attention Multi-Agent TwinDelayed Deep Deterministic Policy Gradient (AT-MA TD3) algorithm framework to implement distributed day-ahead scheduling. Step 5: Import the scheduling strategy into the digital twin simulation for verification, issue control commands via an encrypted protocol, and start autonomous fault-tolerant operation in the event of network outage or equipment failure; Step 6: Compare the simulation data with the field measurement data. If the deviation exceeds the limit, update all model parameters through Bayesian optimization iteration, and form a continuous optimization closed loop through iterative iteration.

[0006] Preferably, in step 1, data is supplemented according to the duration of the missing data: short-duration missing data is supplemented directly, medium-duration missing data is supplemented by merging neighboring data, and long-duration missing data triggers operation and maintenance alarms and is replaced by historical data from the same period. Only data with a verification pass rate of ≥99% can proceed to subsequent calculations.

[0007] Preferably, in step 2, the dynamic emission factor of particle size is output in real time for 15 minutes, the total error of carbon accounting is controlled within ±3%, and the biomass raw material quality and the time series fluctuation of the power grid marginal factor are automatically adapted.

[0008] Preferably, in step 3, only high-reputation nodes are selected to participate in consensus accounting, the latency of a single carbon asset confirmation is ≤200ms, and an tamper-proof electronic certificate with a unique hash is generated.

[0009] Preferably, in step 4, the system is divided into multiple independent intelligent entities, including photovoltaic, energy storage, biomass, biogas, and controllable load. A centralized training and distributed execution mode is adopted, with each entity performing local calculations and raw operating data not being uploaded, thus achieving privacy protection.

[0010] Preferably, in step 4, the typical rural scenarios of spring plowing, summer irrigation, autumn harvest, and winter warmth are divided based on K-means++ clustering, and the parameters of the reinforcement learning model are adaptively initialized in real time according to the scenario.

[0011] Preferably, step 5 uses MQTT (Message Queuing Telemetry Transport) lightweight communication combined with the national cryptographic standard SM2 to complete multi-subject data interaction. After network interruption, each intelligent agent operates independently to ensure power supply to the core agricultural load.

[0012] Preferably, after step 3 is completed, the system automatically connects to the carbon market interface, enabling one-stop online storage and traceability of the entire process of Chinese Certified Emission Reduction (CCER), green certificate application, trading, compliance verification, project changes, and decommissioning settlement.

[0013] Preferably, in step 6, a 5% indicator deviation threshold is set. If the deviation exceeds the threshold, the system automatically iterates back through the hyperparameters of three types of models: data decoupling, carbon accounting, and scheduling optimization, to continuously reduce system operating losses.

[0014] Compared to existing technologies, this invention addresses the hardware limitations and technical shortcomings of rural micro-energy stations. Through optimized proprietary algorithm models, innovative system architecture, and improved software and hardware adaptation, it constructs a full-chain intelligent carbon emission control technology system. This solves the technical problems of low accuracy, poor privacy, weak adaptability, and inability to achieve closed-loop iteration in traditional solutions. The specific beneficial effects of this technology are as follows: 1. Significantly improved accuracy and reliability of multimodal data processing. This invention employs a three-branch decoupled variational autoencoder to decouple the features of multi-source heterogeneous energy data, effectively breaking the problem of strong data coupling and reducing the coupling degree of data features by more than 35%. At the same time, based on the difference in missing duration, a cGAN hierarchical completion strategy is adopted, coupled with a strict data verification mechanism, increasing the data verification pass rate from the traditional 95% to 99.2%, and the missing data completion accuracy rate reaches 96%. This completely solves the industry pain points of missing data, high noise, and error propagation in rural scenarios, providing a high-precision unified data foundation for downstream carbon accounting and scheduling optimization.

[0015] 2. Achieve highly dynamic and accurate carbon emission reduction accounting. This invention integrates ST-GCN dynamic carbon factor calculation, XGBoost waste carbon accounting, and causal inference adaptive deduplication technology. It abandons the traditional fixed factor and manual parameter accounting mode, and can output 15-minute particle size dynamic emission factors in real time. It automatically adapts to the temporal fluctuations of biomass raw material quality and grid marginal factors, and controls the total carbon accounting error within ±2.8%, which is far better than the traditional ±5% accounting accuracy. It fully meets the business needs of high-precision CCER verification and real-time carbon asset statistics.

[0016] 3. Balancing real-time scheduling, optimization accuracy, and multi-agent privacy and security. This invention innovatively adopts a two-layer reinforcement learning collaborative optimization architecture. The upper-layer DDPG algorithm implements monthly long-cycle planning, while the lower-layer AT-MA TD3 algorithm implements day-ahead distributed scheduling, reducing the scheduling solution time from 30 seconds to less than 8 seconds, and improving optimization accuracy by 9.2% compared to traditional distributed optimization algorithms. Through multi-agent centralized training and distributed execution mode, each agent only performs local operations and does not upload raw data, achieving zero data leakage and perfectly adapting to the operation scenario of rural areas with multiple dispersed agents and high privacy protection requirements. At the same time, based on K-means++ clustering to classify typical rural seasonal scenarios, adaptive initialization of model parameters is achieved, significantly improving the scheduling adaptability in complex scenarios.

[0017] 4. Achieve rapid on-chain carbon asset storage and ownership confirmation with low computing power and high reliability. This invention innovatively optimizes and improves the Practical Byzantine Fault Tolerant (IPBFT). The consensus algorithm incorporates a high-reputation node screening mechanism, adapting to the low-computing-power hardware conditions of rural edge nodes and significantly reducing consensus computing power overhead. Combined with a dual-ledger architecture for energy flow and carbon flow synchronized on-chain and a modified Shapley algorithm, it achieves automatic and accurate allocation of carbon asset rights among multiple stakeholders, reducing single-transaction confirmation latency to within 200ms and generating unique, tamper-proof electronic evidence certificates. Simultaneously, it automatically connects to the carbon market system through standardized interfaces, enabling full-process on-chain evidence storage and automatic write-off of carbon assets, resolving the technical problems of mismatched computing power, delayed confirmation, and susceptibility to data tampering inherent in traditional technologies.

[0018] 5. Improve system hardware adaptability and operational stability under complex conditions. This invention adopts a lightweight MQTT communication protocol combined with the national cryptographic SM2 encryption algorithm, adapting to the weak network hardware environment in rural areas and ensuring secure and efficient data interaction among multiple devices and entities. It adds an autonomous fault-tolerant algorithm mechanism for equipment failures and network outages, allowing each intelligent agent to operate independently under abnormal conditions, ensuring uninterrupted power supply to core agricultural loads. Through the K-means++ scene clustering algorithm, it adapts to the differentiated operating conditions of rural areas throughout the four seasons, customizing and optimizing model parameters to effectively solve the system performance degradation problem caused by large fluctuations in rural resource loads and limited equipment computing power. Scene scheduling adaptability is improved by more than 20%, making it adaptable to rural micro-energy station hardware equipment clusters of different regions and scales.

[0019] 6. Construct a fully automated closed-loop iterative optimization technology architecture. This invention sets up a precise indicator deviation threshold judgment mechanism. Through real-time comparison between the digital twin simulation system and the actual measured data of the on-site hardware, when the operating indicators exceed the tolerance, the Bayesian optimization algorithm is automatically invoked to iteratively update the hyperparameters of the entire link model, including data decoupling, carbon accounting, and scheduling optimization. The system performance can be iteratively upgraded without manual intervention, continuously reducing the operating losses of micro-energy station equipment and system control errors, and ensuring the long-term efficient, low-carbon, and stable operation of the energy system. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the five-layer architecture of the integrated electricity and carbon control system for rural micro-energy stations in an embodiment of the present invention; Figure 2 This is a schematic diagram of the physical source-grid-load-storage topology of a rural multi-energy complementary micro-energy station in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the intelligent accounting and adaptive deduplication logic for two-dimensional carbon emission reduction in this embodiment of the invention. Figure 4 This is a flowchart illustrating the automatic carbon asset ownership confirmation process using a dual-ledger blockchain in this embodiment of the invention. Figure 5 This is a block diagram of the two-layer deep reinforcement learning electro-carbon co-optimization model in an embodiment of the present invention; Figure 6 This is a closed-loop flowchart of the entire process of electricity-carbon collaborative management in rural micro-energy stations, as described in this embodiment of the invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be described below with reference to the accompanying drawings. The described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without innovative effort are within the scope of protection of the present invention.

[0022] First, an integrated system for implementing integrated management and control of electricity and carbon emissions from rural micro-energy stations will be established, including: 1. Multimodal data perception decoupling layer This layer provides a high-confidence data foundation for the system, collecting full-dimensional data with a 15-minute time granularity. All data undergoes standardized mapping according to national / industry standards such as GB / T 33602 and GB / T 40607. The core innovation lies in introducing a multi-branch decoupled variational autoencoder and a conditional generative adversarial network to achieve feature decoupling and missing data completion for multi-source heterogeneous data, achieving a data verification pass rate of ≥99%. It comprises 6 functional modules: (1) Energy equipment parameter acquisition module It covers all types of energy equipment in rural micro-energy stations, collects rated parameters and real-time operating parameters of the equipment, and the parameter values ​​are matched with the technical level of mainstream domestic equipment. The core parameter ranges are as follows: Distributed photovoltaic: Module conversion efficiency 20%-23%, annual degradation rate 0.45%-0.7%, unit investment cost 3.2~4.0 yuan / W; Electrochemical energy storage: rated capacity 0.5-2.0 MWh, AC side charge-discharge efficiency 90%-94%, SOC operating range 0.2-0.9 pu, maximum charge-discharge power 0.5C-1.0C, cycle life 6000~10000 times; Biomass cogeneration units: rated power generation 0.5-2.0 MW, power generation efficiency 25%-30%, heating efficiency 50%-60%, minimum technical output 30%-40% of rated, ramp rate 5%~10% of rated / minute; Biogas engineering system: Anaerobic digester volume 300-1000 m³ 3 Gas production rate: 0.3-0.6 m³ 3 / kgVS, biogas methane content 55%-65%, generator set efficiency 28%-32%; Electric heating equipment: COP 2.8-3.5, heat load per unit area 40-60 W / m² 2 .

[0023] (2) Resource endowment collection module Data on three types of resources—solar energy, biomass, and meteorology—were collected at the county level. Solar energy resources: annual equivalent utilization hours (North China 1200-1400 h, Central China 1000-1200 h, Northwest China 1400-1800 h), hourly irradiance data with 15-minute resolution; Biomass resources: Calorific value of crop straw (wheat 14.5-16.5 MJ / kg, corn 13.5-15.5 MJ / kg), moisture content 15%-25%, collectability coefficient 0.6-0.8; VS content of livestock and poultry manure (pig manure 10%-15%, cow manure 15%-20%), methane conversion potential 0.25-0.35 m 3 / kgVS; Calorific value of forestry waste: 16-18 MJ / kg; Meteorological parameters: ambient temperature, wind speed, and relative humidity are collected hourly and used for equipment output correction.

[0024] (3) Load characteristic acquisition module The model uses "typical days in four seasons + monthly coefficients" to collect four types of core load data: Residential electricity consumption: Average daily electricity consumption per person is 1.5~2.5 kWh, peak hours are 18:00-22:00, off-peak hours are 23:00-6:00 the next day, and the peak-to-valley difference rate is 30%~50%; Electricity consumption for agricultural production: During the summer irrigation season (June-August), the daily electricity consumption for irrigation is 8-12 hours; during the autumn harvest season (September-November), the load factor for agricultural product processing is 0.6-0.8. Heating load: During the winter heating season from November to March of the following year, the indoor design temperature is 18±2℃, and the daily load fluctuation coefficient is 0.7~1.2; Gas load: Cooking gas consumption has two peaks in the morning and evening, biogas supply pressure ≥2kPa, and hourly unevenness coefficient 1.5~2.0.

[0025] (4) Market boundary parameter acquisition module Collect price and quota parameters from the electricity, carbon, and green certificate markets as boundary conditions for optimized dispatching: Electricity prices: Benchmark on-grid price for desulfurized coal: 0.35-0.45 yuan / kWh; Peak-valley price difference: 0.2-0.4 yuan / kWh; Residential electricity price: 0.5-0.6 yuan / kWh; Agricultural production electricity price: 0.4-0.5 yuan / kWh. Carbon pricing: The national carbon market quota transaction price is 50-90 yuan / tCO2, the CCER transaction price is 40-80 yuan / tCO2, and the local carbon market price is 30-70 yuan / tCO2; Green certificates: Onshore wind power green certificates are priced at 30-60 yuan each, and photovoltaic green certificates are priced at 40-80 yuan each. One certificate corresponds to 1000 kWh of renewable energy. Fuel costs: Straw purchase price 200-350 yuan / ton, livestock and poultry manure transportation and treatment cost 30-50 yuan / ton.

[0026] (5) Main technical attribute acquisition module Collect the technical attribute parameters of each participating node for system configuration matching and rights allocation calculation: Equipment capacity ratio, data upload capability, and operation and maintenance technology level; Node computing power level, network stability, and historical data accuracy.

[0027] (6) Multi-branch decoupling feature extraction module As the core innovative module of this layer, a three-branch variational autoencoder (VAE) is adopted. The architecture implements feature decoupling and noise reduction for multimodal data. The specific structure and algorithm are as follows: Three-branch parallel encoder: Time series data encoder: For time series data such as equipment output and load with 15-minute granularity, a 3-layer one-dimensional convolution + time attention mechanism structure is adopted. The convolution kernel size is 3 and the stride is 1. The attention layer captures intraday cycle and interday mutation features and outputs a 64-dimensional time series feature vector. Spatial Data Encoder: For county-level resource and load spatial distribution data, a 2-layer graph convolutional network (GCN) is used, with administrative villages as nodes and geographical distance and resource correlation as edge weights, to extract spatial correlation features and output a 32-dimensional spatial feature vector; Discrete Attribute Encoder: For discrete data such as device parameters and subject attributes, an embedding layer + 2 fully connected network is used to map discrete variables into continuous embedding vectors and output a 32-dimensional attribute feature vector.

[0028] Unified implicit space and decoupling constraints: The three types of feature vectors are concatenated into a 128-dimensional joint latent variable, and the original data is reconstructed through a decoder. The mutual information minimization adversarial constraint is introduced to ensure that the mutual information between the four core feature dimensions of energy output, load demand, resource endowment, and market price is ≤0.05, thereby achieving complete decoupling of features.

[0029] Parameter basis: Through testing on the actual data set of 12 typical rural micro-energy stations in China, when the mutual information between features is ≤0.05, the error of the downstream carbon accounting model is reduced by more than 35% and the optimization accuracy of the scheduling model is improved by 8%. Therefore, this constraint threshold is set.

[0030] cGAN missing data completion unit: To address data loss caused by unstable data collection equipment in rural areas, a conditional generative adversarial network (GAN) consisting of a bidirectional LSTM generator and a fully connected discriminator is constructed. Using historical data from the same period over the past 7 days, neighboring data from three adjacent administrative villages, and real-time meteorological data as conditional variables, data is completed in stages based on the duration of the missing data: Missing data ≤ 4 hours: direct completion with an accuracy ≥ 96%; Missing data 4–12 hours: combined completion using weighted mapping of neighboring data and the generator, with an accuracy ≥ 88%, triggering a level 1 alarm; Missing data > 12 hours: triggering a level 2 alarm, notifying maintenance personnel to investigate the data collection equipment, and simultaneously using weighted replacement data from the same period of the same month.

[0031] Data quality assessment subunit: The completed data undergoes consistency verification, outlier detection, and confidence assessment. Data with a verification pass rate of ≥99% is input into the upper-level model; unqualified data is returned to the completion unit for secondary processing, forming a closed-loop data quality control system.

[0032] 2. Intelligent Verification and On-Chain Rights Confirmation Layer for Carbon Assets This layer is the core accounting module of the system. Breaking through the limitations of traditional static accounting, it establishes a full-process technical system of "dynamic dual-dimensional accounting + adaptive deduplication + full lifecycle deduction + consortium blockchain rights confirmation," controlling the accounting error within ±3%, meeting the accuracy requirements of the CCER project verification. It comprises four functional modules: (1) Dual-dimensional carbon emission reduction intelligent accounting module A dual-dimensional accounting system of "emission reduction through power generation substitution + emission reduction through waste treatment" is constructed, and all of them adopt dynamic prediction models to replace traditional fixed factors, which greatly improves the accuracy and dynamism of the accounting.

[0033] 1) The calculation of carbon emission reduction by new energy power generation adopts the time-by-time marginal carbon emission factor dynamic prediction method based on spatiotemporal graph convolutional network (ST-GCN), which replaces the traditional annual average factor method, improving the calculation accuracy by more than 18%.

[0034] The ST-GCN emission factor prediction model takes as input real-time power output of each unit in the regional power grid, power flow of lines, and total output of renewable energy. Spatially, it extracts the topological relationships of power grid nodes through graph convolution, and temporally, it captures temporal dependencies through causal convolution. The output is the regional power grid marginal carbon emission factor EFgrid(d,t) with a 15-minute granularity. The typical value for the North China power grid is 0.55-0.65 tCO2 / MWh, and for the Central China power grid it is 0.45-0.55 tCO2 / MWh.

[0035] Emission reduction calculation formula: In the formula: , , The active power (kW) of photovoltaic, biomass, and biogas power generation during the t-th time period on day d. The time step is set to 0.25h; This is the power grid line loss correction factor, with a value ranging from 0.94 to 0.96.

[0036] 2) Calculation of carbon emission reduction from agricultural waste treatment The emission factor is dynamically calculated using a physically constrained XGBoost regression model, adapted to the characteristics of large fluctuations in the quality of raw materials in rural areas, and is divided into two categories of calculation: Emission reduction by replacing open burning of straw: In the formula: The annual dry weight of straw consumed (t); The dynamic straw burning emission factor output by the XGBoost model is 0.60-0.70 tCO2 / t for wheat straw and 0.55-0.65 tCO2 / t for corn straw. The ash content of the straw is 5% to 10%; The regional correction coefficients are 1.0 for Northeast / North China, 0.95 for Central China, and 0.9 for South China.

[0037] Methane emission reduction from livestock and poultry manure: In the formula: The annual processing volume of livestock and poultry manure (t) is given. Content of volatile solids; The maximum methane production potential is 0.25-0.35 m3CH4 / kgVS; MCFbaseline is the baseline methane conversion factor of 0.1-0.2; MCFproject is the project methane conversion factor of 0.8-0.9; calculated based on 1tCH4 = 28tCO2 equivalent.

[0038] (2) Adaptive deduplication verification and net emission reduction calculation module The core innovation is a dynamic deduplication algorithm based on the causal inference front-door adjustment criterion, which replaces the manually set fixed deduplication coefficient and solves the problem of double accounting for biomass emission reduction from a technical perspective.

[0039] 1) Adaptive deduplication algorithm Define causal variables: biomass fuel consumption M is the treatment variable, and waste treatment emission reduction is the causal variable. As the outcome variable, the emission reduction from biomass power generation substitution The mediating variable is used. The causal path of "waste treatment → fuel combustion → power generation and emission reduction" is identified using the front-door adjustment criterion. The proportion of the mediating effect to the total effect is calculated, and a deduplication coefficient is dynamically generated. The value ranges from 0.3 to 0.7 and is updated every 24 hours.

[0040] Waste emission reduction after deweighting: If the assumption of complete carbon neutrality is adopted, then Biomass power generation is not included in the emission reduction from power generation substitution, but only in the emission reduction from waste treatment.

[0041] 2) Calculation of total net carbon emission reduction Deductions for dynamic allocation of carbon emissions over the entire life cycle and conservative deductions based on certification: In the formula: The verification deduction rate is set at 5%-10%; ELCA is the annual carbon emission reduction for the entire life cycle, including all stages of equipment manufacturing, transportation, construction, operation and maintenance and scrapping, and is dynamically allocated according to the actual operating loss rate of the equipment, with a typical value of 5%-8% of the total emission reduction.

[0042] (3) Consortium blockchain dual ledger rights confirmation module The core innovation is "parallel accounting of energy flow and carbon flow dual ledgers + improved PBFT consensus", which realizes real-time synchronous confirmation of carbon assets and tamper-proof evidence storage, with a confirmation latency of ≤200ms.

[0043] 1) Dual-ledger parallel accounting mechanism Set up an energy transaction ledger and a carbon asset ledger, and bind the two ledgers together using the same timestamp hash value, mapping them one-to-one: Energy transaction log: Records data on the output, load, electricity purchase and sale, heating and gas supply of each device at a 15-minute interval, serving as the original basis for carbon accounting; Carbon asset ledger: records the net carbon emission reductions, equity shares of each entity, and certification status for the corresponding period.

[0044] The system automatically synchronizes all data for the day to the blockchain at midnight every day, enabling the synchronous generation and confirmation of energy output and carbon assets, thus resolving issues such as discrepancies between accounts and actual assets and delays in asset confirmation.

[0045] 2) Improved Practical Byzantine Fault-Tolerant Consensus Algorithm (IPBFT) To address the limited computing power of rural nodes, a node reputation value mechanism is introduced to optimize the consensus process. Node reputation score calculation: ,in To improve data upload accuracy, For the speed of response to historical consensus, To ensure data integrity, updates are performed every 24 hours. Consensus node selection: Only the top 30% of core nodes with the highest reputation scores are selected to participate in the three-stage consensus process of "pre-preparation-preparation-commitment". The remaining nodes serve as verification nodes to synchronize the ledger. Fault tolerance: When the number of core nodes is N, it can tolerate ⌊(N-1) / 3⌋ malicious nodes, and the consensus latency is reduced by more than 40% compared with the traditional PBFT, making it suitable for rural low-computing-power node scenarios.

[0046] 3) Carbon asset equity allocation unit Based on the entropy weight-TOPSIS method to quantify the contribution of entities, the initial allocation of Shapley values ​​is corrected to achieve an objective and fair division of carbon asset rights. Step 1: Objectively calculate the weights of three types of indicators—resource contribution, equipment contribution, and operation and maintenance contribution—using the entropy weight method to avoid subjective weighting bias; Step 2: Calculate the overall contribution score of each subject using the TOPSIS method; Step 3: Generate correction coefficients to adjust the initial allocation of Shapley values: In the formula: Initially allocate funds to Shapley values; This is a contribution correction factor, ranging from -0.2 to +0.3; This is a technical risk correction factor, ranging from -0.2 to +0.2. Constraints: Total equity equals total net carbon emission reduction; each entity's equity is not less than the carbon asset corresponding to its independent contribution; equity value is non-negative.

[0047] Once the ownership is confirmed, an electronic certificate with a unique hash value is generated for each carbon asset, which includes the time, quantity, owner, and verification status, and cannot be tampered with.

[0048] (4) Carbon Asset Lifecycle Management Module Covering the entire process of carbon assets from generation to cancellation, it achieves trusted management and control of the entire lifecycle on the blockchain. It has 4 sub-units: 1) The transaction declaration management unit supports the batch declaration and listing of carbon assets such as CCER and green certificates, automatically generates accounting drafts, project documents and declaration materials that comply with carbon market norms, and connects to the national carbon market and local carbon market trading interfaces; it realizes one-click listing, transaction transfer and fund settlement of carbon assets on the blockchain, and automatically updates the carbon asset ledger after the transaction is completed, and generates transaction vouchers for on-chain storage at the same time, reducing the declaration time by more than 90% compared with the manual mode.

[0049] 2) The compliance and write-off management unit connects to the carbon market compliance system, automatically matches the quota gap of the emission control entities, and automatically deducts the corresponding carbon asset quota to complete the compliance and write-off after confirmation by the certification body; the write-off record is on the blockchain and cannot be tampered with, and compliance completion certificates are automatically generated to achieve full traceability of the write-off process.

[0050] 3) The project change registration unit automatically recalculates the carbon asset baseline and emission reduction for scenarios such as expansion, reduction, and change of operating entity of micro-energy station equipment, and generates change plans. The change plans take effect after being confirmed on the chain by the verification agency node, and the carbon asset ledger and full life cycle archive are updated simultaneously to ensure that asset changes are compliant and traceable.

[0051] 4) When a project reaches its operating period or is decommissioned early, the decommissioning and liquidation management unit automatically completes the liquidation and accounting of carbon assets throughout their entire life cycle, calculates the cumulative emission reductions, cumulative trading revenues, remaining asset quotas, and carbon footprint throughout their entire life cycle, and generates a decommissioning and liquidation report; it also completes the cancellation of on-chain assets and permanent archiving of files, thus achieving a closed loop for the entire life cycle of carbon assets.

[0052] 3. Electro-carbon synergistic dual-layer intelligent optimization layer This layer is the core computing module of the system, adopting a two-layer distributed reinforcement learning architecture of "upper-layer monthly planning - lower-layer day-day scheduling". The solution speed is ≤8 seconds / day-day scheduling and ≤8 minutes / monthly plan. After optimization, the overall system performance is improved by more than 20% compared with traditional solutions. It consists of 4 functional modules: (1) Quadruple Coupled Optimization Objective Function With the multiple objectives of minimizing overall system operating cost and maximizing carbon emission reduction, a quaternary coupled optimization model of electricity, carbon, heat, and gas is constructed, with the objective function as follows: In the formula: To calculate the annualized investment cost, the equal annuity method is adopted, with a discount rate of 6%-8%, and the equipment lifespan is calculated as 25 years for photovoltaic, 10 years for energy storage, and 15 years for biomass units. For operation and maintenance costs, photovoltaic power accounts for 1%-1.5% of the investment per year, energy storage accounts for 2%-3% per year, and biomass power units account for 4%-6% per year. Cost of biomass fuel; Net cost of electricity purchase and sale to the power grid; Net cost of carbon allowances; This is the carbon emission reduction weighting coefficient, used for multi-objective trade-offs.

[0053] The core constraints include: power balance constraints, energy storage operation constraints, biomass unit ramp-up constraints, grid-connected power constraints, carbon quota constraints, green certificate consumption constraints, and heating / gas storage balance constraints.

[0054] (2) Upper-level monthly planning unit: DDPG deep reinforcement learning algorithm The Deep Deterministic Strategy Gradient Algorithm (DDPG) is used to solve the medium- and long-term operation plan on a monthly scale, which is suitable for scenarios with strong uncertainty.

[0055] State space: monthly electricity price forecast series, carbon price forecast series, biomass resource forecast output, load forecast, equipment health status parameters; Action plan: monthly biomass fuel procurement volume, monthly carbon asset reserve ratio, risk hedging ratio, and monthly energy storage operation strategy; Reward function: With minimizing monthly comprehensive operating costs as the core, a conditional value at risk (CVaR) constraint with a 95% confidence level is introduced, and a penalty term is applied when the CVaR exceeds 10% of the annual expected cost; Output: Monthly operating boundary plan, including monthly total electricity purchase and sale, total carbon asset declaration, green certificate trading plan, and fuel procurement plan, which is distributed to the lower-level day-ahead dispatching units.

[0056] (3) Lower-level day-ahead scheduling unit: Attention-guided multi-agent TD3 algorithm Attention-guided multi-agent dual-delay deep deterministic strategy The gradient algorithm (Attention-MA-TD3) implements distributed day-ahead scheduling, balancing data privacy and solution efficiency.

[0057] Intelligent agents are classified into five categories: photovoltaic units, energy storage units, biomass units, biogas systems, and controllable loads. Each intelligent agent only accesses local operating data, naturally protecting the data privacy of multiple entities. Global attention mechanism: A global attention network is designed to dynamically adjust the information interaction weights between agents based on the real-time global state of the system, such as power deficit and carbon emission reduction gap. It focuses on global constraints such as power balance and total carbon emission reduction, while weakening non-critical information interaction. The convergence speed is improved by more than 30% compared with the traditional MA-TD3. Training framework: It adopts a "centralized training, distributed execution" architecture. During the training phase, the global state is aggregated to calculate the reward, and during the execution phase, each agent outputs actions based only on local observations. Constraint handling: Hard constraints such as energy storage SOC, biomass ramp-up rate, and grid-connected power are embedded into the algorithm through action space pruning; soft constraints such as carbon quotas and green certificate quotas are transformed into penalty terms and added to the reward function.

[0058] (4) Seasonal scene adaptive scheduling unit Using the K-means++ unsupervised clustering algorithm, four typical scenario libraries—spring plowing, summer irrigation, autumn harvest, and winter warmth—are generated based on historical operational data, corresponding to different resource-load characteristics. Spring plowing season (March-May): Photovoltaic power output rebounds, irrigation load starts, priority is given to meeting local electricity needs, and surplus electricity is fed into the grid; Summer irrigation season (June-August): Photovoltaic output is at its highest for the year, irrigation electricity consumption is at its peak, and priority is given to selling electricity and green certificates; Autumn harvest season (September-November): Large quantities of straw are produced, agricultural product processing reaches its peak, biomass units increase their load, and carbon assets are stored. During the warm winter season (December-February), heating load peaks, with combined heat and power (CHP) prioritizing heating and carbon quota trading as the main activity. The system automatically matches scenarios based on real-time meteorological and load data, initializes reinforcement learning model parameters, and improves scheduling adaptability and training efficiency.

[0059] 4. Multi-entity collaborative regulation and execution layer This layer is responsible for the implementation of optimization strategies and multi-entity technical collaboration. It consists entirely of technical interaction rules and execution logic, excluding business operation rules. It comprises four functional modules: (1) Multi-subject data interaction protocol unit Define data interaction standards for four types of nodes: village-level nodes, operator nodes, power grid nodes, and verification agency nodes. Communication protocol: Employs the MQTT lightweight communication protocol, adapted to rural network environments; Data security: Data is encrypted and transmitted and authenticated using the national cryptographic algorithm SM2. Interaction Specifications: Define the data upload frequency, command response latency (≤5s), and data format standards for each node to achieve standardized data interaction among multiple nodes.

[0060] (2) System configuration adaptive matching unit Based on the input resource endowment, equipment scale, and main technical attribute parameters, the system automatically matches the optimal system technical configuration scheme: Output the optimal device capacity ratio, algorithm parameters, and consensus node configuration; It outputs the technical and economic indicators of the system operation, including equipment utilization rate, carbon emission reduction intensity, investment payback period, etc., and supports horizontal comparison of multiple schemes.

[0061] (3) Control command issuance and feedback unit The scheduling strategy output by the optimization layer is broken down into precise control commands for each device and sent to the corresponding execution terminal; real-time device operation feedback data is collected, scheduling deviations are calculated and corrected in a closed loop, and the command execution deviation rate is ≤3%.

[0062] (4) Autonomous operation unit under extreme operating conditions 1) Off-grid autonomous subunit: Automatically switches to off-grid autonomous mode when the network is interrupted. Each local agent operates independently based on preset strategies and local data to ensure the supply of core loads. After the network is restored, it automatically synchronizes operating data and returns to the grid-connected scheduling mode. 2) Fault-tolerant subunit: Monitors the operating status of equipment in real time. When equipment fails, it automatically triggers fault-tolerant scheduling, calls on backup resources to fill the gap, and issues fault alarms. After the fault is cleared, it automatically restores normal scheduling.

[0063] 5. Digital Twin Visualization and Interaction Layer This layer serves as the system's human-computer interaction and simulation verification platform, upgrading traditional visualization functions to a digital twin to achieve full-element mapping and strategy pre-playback. It comprises five functional modules: (1) Full-element digital twin modeling unit A three-dimensional digital twin of rural micro-energy stations—source, grid, load, and storage—is constructed, mapping physical equipment and network topology in a 1:1 ratio. The dynamic changes of energy flow, carbon flow, and information flow are rendered in real time, with rendering accuracy reaching the equipment level.

[0064] (2) Strategy simulation and deduction unit After optimizing the scheduling strategy output, a 24-hour simulation is conducted in a digital twin environment to verify technical indicators such as power balance, equipment operating status, carbon emission reduction, and constraint satisfaction rate, and a feasibility assessment report is output to avoid actual operational risks.

[0065] (3) Technical indicator analysis unit It provides statistical analysis of technical indicators such as carbon accounting accuracy, solution speed, equipment utilization rate, carbon emission reduction intensity, and data qualification rate; it supports single-factor / multi-factor sensitivity analysis of eight types of parameters, including electricity price, carbon price, photovoltaic power output, and biomass production, and outputs sensitivity coefficients and tornado maps.

[0066] (4) Model parameter inverse iteration unit Regularly compare the results of digital twin simulations with actual operating data. When the deviation of key indicators exceeds 5%, use the Bayesian optimization algorithm to iteratively update the hyperparameters of the data decoupling model, carbon accounting model, and reinforcement learning scheduling model. After the iteration is completed, the model configuration is automatically updated to achieve continuous improvement in system performance.

[0067] (5) Scheme Derivation Unit Generate Word version technical solution report, PDF version operation analysis report, Excel version operation data report and carbon accounting draft with one click.

[0068] Based on the above system, the integrated electricity and carbon control method for rural micro-energy stations provided by this invention includes the following steps: S1. Multimodal heterogeneous data acquisition and decoupling feature extraction Collect five categories of basic data on rural micro-energy stations in the target area: energy equipment parameters, resource endowment, load characteristics, market boundaries, and main entity attributes, and perform standardized mapping according to national / industry standards. A three-branch decoupled VAE model is used for multimodal feature extraction. Feature decoupling is achieved by minimizing mutual information and adversarial constraints, thereby reducing feature coupling. Conditional GANs are used to perform hierarchical intelligent completion of missing data, and corresponding completion strategies and alarm mechanisms are triggered according to the duration of missing data. The processed data undergoes a quality assessment. Once the pass rate is ≥99%, a standardized feature database is constructed. Unqualified data is returned and supplemented, forming a closed loop for data quality.

[0069] S2, Dual-dimensional intelligent carbon emission reduction accounting and adaptive deduplication verification The hourly marginal carbon emission factor of the power grid is predicted based on the spatiotemporal graph convolutional network, and the carbon emission reduction by power generation substitution is calculated by combining the hourly output data of new energy sources. Based on the XGBoost model with physical constraints, the carbon emission reduction of straw burning and livestock manure treatment is dynamically calculated to obtain the total carbon emission reduction of waste treatment. Based on the causal inference front-door adjustment criterion, the deduplication coefficient is dynamically calculated to eliminate the error in the repeated emission reduction accounting of biomass fuel; The annual and hourly net carbon emission reductions are output by deducting the dynamic allocation of carbon emissions over the entire life cycle and the conservative deduction of certification.

[0070] S3, Synchronous Consortium Blockchain Carbon Asset Ownership Confirmation and Rights Allocation Energy flow data and net carbon emission reductions are simultaneously uploaded to the consortium blockchain via a dual ledger mechanism; An improved IPBFT consensus algorithm is used to complete distributed notarization, ensuring that the data is tamper-proof; Based on the entropy weight-TOPSIS method, the overall contribution of each entity is quantified, the initial allocation of Shapley values ​​is corrected, and the carbon asset equity share is automatically allocated. Generate electronic carbon asset certificates with unique hash values ​​and distribute them to various principal nodes and certification authority nodes.

[0071] S4, Two-layer deep reinforcement learning-based co-scheduling optimization of electric carbon The upper layer uses the DDPG algorithm, which takes monthly forecast data and risk constraints as input, solves the monthly operation plan, and outputs the daily operation boundary. The lower layer uses the Attention-MA-TD3 multi-agent algorithm to solve the optimal strategies for hourly equipment output, energy storage charging and discharging, and load regulation based on the data from 96 points before the current day. Based on real-time weather and load data, the system automatically matches seasonal scenarios, optimizes model parameters, and improves scheduling adaptability.

[0072] S5, Digital Twin Simulation and Collaborative Execution The optimized scheduling strategy is input into the digital twin for simulation and deduction to verify the technical feasibility and constraint satisfaction rate. Control commands are sent to each execution terminal through a multi-entity interaction protocol; Real-time collection of operational feedback data and closed-loop correction of scheduling deviations; Under extreme operating conditions, it automatically switches to autonomous operation mode to prioritize the supply of core loads.

[0073] Step S6: Iterative Model Optimization Based on actual operational data and feedback, the parameters of the data decoupling model, carbon accounting model, and reinforcement learning scheduling model are updated iteratively in reverse order to continuously improve system performance and form a closed-loop technology chain.

[0074] All embodiments and comparative examples use the same hardware boundaries and computing benchmarks, control resource conditions and equipment configurations, and only change the technical solution system.

[0075] Calculation benchmark: discount rate 6.5%; calculation period: full life cycle 25 years, equipment is depreciated according to its own lifespan; core comparative technical indicators include: net carbon emission reduction calculation error, day-ahead scheduling solution time, data verification pass rate, consensus confirmation delay, annual return volatility, and optimization accuracy deviation.

[0076] Example 1: Rural Micro-Energy Stations in Agricultural Areas of the Central China Plain 1. Basic conditions of the project This is an administrative village in a plain area of ​​Central China, with 350 households and a permanent population of 1,120. It has 3,200 mu of arable land, producing 4,800 tons of corn stalks annually, and a large-scale pig farm producing 2,100 tons of pig manure annually. The annual equivalent photovoltaic utilization hours are 1,150 hours, and the regional power grid marginal carbon emission factor baseline is 0.52 tCO2 / MWh. There is no traditional centralized heating system.

[0077] 2. System Hardware Configuration Distributed photovoltaic: 1.8MW, monocrystalline silicon TOPCon modules, conversion efficiency 22%; Electrochemical energy storage: 0.6MW / 1.2MWh lithium iron phosphate battery, AC side efficiency 92%, SOC operating range 0.2~0.9; Biomass cogeneration: 0.8MW straw direct combustion unit, with a power generation efficiency of 28% and a heating efficiency of 55%; Biogas project: 600m3 mesophilic anaerobic digester, gas production rate 0.4m3 / kgVS, equipped with a 150kW biogas generator; Terminal renovation: The entire village will undergo air source heat pump electric heating renovation, covering a heating area of ​​42,000 square meters.

[0078] 3. System Technical Parameter Configuration Data layer: A multi-branch decoupled VAE model is adopted, consisting of a 3-layer convolutional temporal encoder and a 2-layer GCN spatial encoder, with a mutual information constraint threshold of 0.05; conditional GAN ​​missing completion is used, with a completion window of 4 hours. Carbon accounting layer: ST-GCN emission factor prediction model, 3-layer spatiotemporal graph convolution; causal front-door adjustment for deduplication, dynamic deduplication coefficient γ baseline value of 0.4; life cycle carbon footprint deduction of 7%; Optimization layer: Upper layer DDPG monthly planning, 95% confidence CVaR constraint; Lower layer Attention-MA-TD 3-day day scheduling, 5 types of agents, 4 attention heads; Rights confirmation layer: 4 core nodes of the consortium blockchain, IPBFT consensus, reputation value updated daily; entropy rights - TOPSIS rights allocation.

[0079] 4. Operational testing technical indicators Example 2: Rural Micro-Energy Stations in Northern Agricultural and Livestock Integration Models 1. Basic conditions of the project This is an agro-pastoral administrative village in North China, with 280 households, 4,500 mu of cultivated land, an annual output of 6,200 tons of wheat and corn stalks, and an annual output of 3,600 tons of cattle manure from a beef cattle breeding base. The winter heating season is 5 months, the annual equivalent photovoltaic utilization hours are 1,320 hours, and the grid marginal carbon emission factor baseline value is 0.58 tCO2 / MWh. Previously, heating was mainly achieved through scattered coal burning.

[0080] 2. System Hardware Configuration Distributed photovoltaic: 2.2MW, installed on village collective rooftops and idle wasteland; Electrochemical energy storage: 1.0MW / 2.0MWh lithium iron phosphate energy storage; Biomass cogeneration: 1.2MW straw direct combustion unit, with supporting centralized heating network; Biogas project: 1000m3 mesophilic anaerobic digester, equipped with a 200kW biogas generator; Terminal load: centralized heating for all residents of the village + insulation and heating for livestock pens.

[0081] 3. System Technical Parameter Configuration Carbon accounting: The dynamic deweighting factor γ has a baseline value of 0.3, and the emission reduction from biomass heating replacing scattered coal is additionally included; Optimization: Prioritize heating dispatch for winter heating scenarios and prioritize electricity sales dispatch for summer irrigation scenarios; risk hedging ratio is 50%. Confirmation of rights: In the joint venture model, the allocation of equity and the weighting of resource contributions have increased.

[0082] 4. Operational testing technical indicators Example 3: Small-scale rural micro-energy stations in hilly and mountainous areas of southern China 1. Basic conditions of the project This is an administrative village in a mountainous area of ​​southern China, with 180 households and 1200 mu of arable land. The annual collection capacity of forestry waste and crop straw is 1100 tons, and there is no large-scale livestock farming. The annual equivalent photovoltaic utilization hours are 1080 hours, and the baseline value of the grid marginal carbon emission factor is 0.48 tCO2 / MWh. The load is mainly for residential electricity consumption and tea processing.

[0083] 2. System Hardware Configuration Distributed photovoltaic power: 0.8MW, from villagers' rooftops to the rooftops of village collective public buildings; Electrochemical energy storage: 0.2MW / 0.5MWh; Biomass heating unit: 0.3MW gasification unit, supplying hot water for tea processing and public buildings; Small-scale biogas project: 200m3 ambient temperature anaerobic tank for treating domestic sewage and free-range manure.

[0084] 3. System Technical Parameter Configuration Carbon accounting: The dynamic deweighting factor γ has a baseline value of 0.5, which includes emission reductions from replacing coal-fired power with forestry waste; Optimization: Primarily self-generated and self-used, with surplus power fed into the grid; lightweight model adapted for small-scale systems; Confirmation of rights: Village-level autonomous model, with a weighting of farmers' resource contributions.

[0085] 4. Operational testing technical indicators Comparative Example 1: Traditional Data Processing + Static Carbon Accounting Scheme The basic hardware is exactly the same as in Example 1, using traditional standardized data acquisition, fixed emission factor carbon accounting, and manually set deduplication coefficients. Comparative Example 2: Traditional ADMM Distributed Optimization Scheme The basic hardware and carbon accounting scheme are the same as in Example 1, using traditional ADMM two-layer distributed optimization without reinforcement learning and attention mechanisms. Comparative Example 3: Traditional PBFT Consortium Blockchain Rights Confirmation Scheme The basic hardware and accounting optimization scheme are the same as in Example 1, using the traditional PBFT consensus algorithm without a dual ledger synchronization mechanism. From the perspective of data processing and carbon accounting performance, the multi-branch decoupled VAE and dynamic dual-dimensional accounting system of this invention improves the data verification pass rate to over 99% and reduces carbon accounting error by nearly half, effectively solving the technical pain points of poor data quality and low accounting accuracy in rural scenarios. From the perspective of optimization and scheduling performance, the Attention-MA-TD3 algorithm improves the solution speed by more than 3 times and significantly improves the optimization accuracy, while achieving distributed data privacy protection. From the perspective of rights confirmation performance, the IPBFT dual ledger mechanism reduces the rights confirmation latency by more than 75% and achieves synchronous storage of energy flow and carbon flow. The synergistic effect of the entire chain of technical solutions has qualitatively improved the overall system performance compared with traditional solutions and is highly adapted to the scenario characteristics of rural micro-energy stations.

[0086] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computers or available storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are 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. A method for integrated management and control of electricity and carbon emissions in rural micro-energy stations, characterized in that, It includes the following closed-loop steps across the entire supply chain: Step 1: Collect multimodal energy data, extract features using a three-branch decoupled variational autoencoder and constrain mutual information to achieve feature decoupling, use cGAN to hierarchically complete missing data, and screen and verify qualified data as a unified computing basis; Step 2: Calculate the carbon emission reduction of new energy power generation substitution based on ST-GCN, calculate the carbon emission reduction corresponding to waste treatment based on XGBoost, dynamically generate the deduplication coefficient using the causal inference algorithm, and subtract the carbon emission of the whole life cycle to obtain the net carbon emission reduction. Step 3: Construct dual ledgers for energy flow and carbon flow, and simultaneously upload them to the blockchain. Use the improved IPBFT consensus to complete distributed notarization, and combine entropy weight-TOPSIS to correct the Shapley value to automatically allocate carbon asset rights of each entity. Step 4: Employ a two-layer reinforcement learning approach to perform electricity-carbon-heat-gas collaborative optimization. The upper layer uses the DDPG algorithm framework to solve the monthly planning, while the lower layer uses the AT-MA TD3 algorithm framework to implement distributed day-ahead scheduling. Step 5: Import the scheduling strategy into the digital twin simulation for verification, issue control commands via an encrypted protocol, and start autonomous fault-tolerant operation in the event of network outage or equipment failure; Step 6: Compare the simulation data with the field measurement data. If the deviation exceeds the limit, update all model parameters through Bayesian optimization iteration, and form a continuous optimization closed loop through iterative iteration.

2. The integrated electricity and carbon control method for rural micro-energy stations according to claim 1, characterized in that, In step 1, data is supplemented according to the duration of missing data: short-duration missing data is supplemented directly, medium-duration missing data is supplemented by merging neighboring data, and long-duration missing data triggers operation and maintenance alarms and is replaced by historical data from the same period. Only data with a verification pass rate of ≥99% can proceed to subsequent calculations.

3. The integrated electricity and carbon control method for rural micro-energy stations according to claim 1, characterized in that, In step 2, the dynamic emission factor of particle size is output in real time for 15 minutes, and the total error of carbon accounting is controlled within ±3%. It automatically adapts to the time-series fluctuations of biomass raw material quality and power grid marginal factor.

4. The integrated electricity and carbon control method for rural micro-energy stations according to claim 1, characterized in that, In step 3, only high-reputation nodes are selected to participate in consensus accounting, the latency of a single carbon asset confirmation is ≤200ms, and an tamper-proof electronic certificate with a unique hash is generated.

5. The integrated electricity and carbon control method for rural micro-energy stations according to claim 1, characterized in that, Step 4: Divide the system into multiple independent intelligent entities, including photovoltaic, energy storage, biomass, biogas, and controllable load. Adopt a centralized training and distributed execution mode, with each entity performing local calculations and raw operating data not being uploaded to protect privacy.

6. The integrated electricity and carbon control method for rural micro-energy stations according to claim 1, characterized in that, Step 4: Based on K-means++ clustering, divide the rural typical scenarios into four categories: spring plowing, summer irrigation, autumn harvest, and winter warmth, and match the scenarios in real time to adaptively initialize the reinforcement learning model parameters.

7. The integrated electricity and carbon control method for rural micro-energy stations according to claim 1, characterized in that, Step 5: Use MQTT lightweight communication combined with the national cryptographic standard SM2 to complete multi-agent data interaction. After network interruption, each intelligent agent operates independently to ensure power supply to the core agricultural load.

8. The integrated electricity and carbon control method for rural micro-energy stations according to claim 1, characterized in that, Step 3: After the confirmation of rights is completed, the system will automatically connect to the carbon market interface to complete the entire process of CCER, green certificate application, trading, performance verification, project change, and retirement liquidation on the blockchain with one-stop on-chain evidence storage and traceability.

9. The integrated electricity and carbon control method for rural micro-energy stations according to claim 1, characterized in that, Step 6: Set a 5% threshold for indicator deviation. If the deviation exceeds the threshold, automatically reverse the iteration of hyperparameters for three types of models: data decoupling, carbon accounting, and scheduling optimization, to continuously reduce system operating losses.

Images