Carbon emission accounting method for biomass gas coupled with coal-fired power generation in carbon trading

Abstract

The application discloses a biomass gas coupling coal-fired power generation full life cycle carbon emission accounting method for carbon trading, and the method comprises the following steps: determining the accounting boundary covering raw material collection and transportation, storage and pretreatment, gas production and coupling power generation; collecting activity data, establishing the space-time mapping of upstream batch data and unit side continuous operation data, and performing data quality grading and uncertainty measurement; when the measured values of syngas components or low calorific value are missing, a soft measurement model is constructed based on operation data, batch characteristics are fed forwardly coupled, and estimated values and uncertainties are output; intermediate quantities are calculated based on a material-energy-carbon conservation mechanism model, a residual correction model with physical consistency constraints is used for dynamic correction, and the weight of the physical constraint is adjusted according to the uncertainty; combined with the dynamic emission parameter, an emission inventory and MRV data package are generated, which are used for monitoring, reporting, verification and transaction settlement, so that the accuracy and traceability are improved and the physical consistency is ensured.

Description

Technical Field

[0001] This invention relates to the fields of carbon emission accounting and power information technology, specifically to a method for carbon emission accounting of biomass gas coupled with coal-fired power generation throughout its entire life cycle for carbon trading, which falls under the category of computer-based emission measurement, traceability and verification, i.e., MRV data processing and management technology. Background Technology

[0002] Biomass gasification coupled with coal-fired power generation technology relies on existing facilities to achieve renewable energy substitution. However, converting its emission reduction benefits into tradable carbon assets requires a precise accounting system that meets the requirements of measurable, reportable, and verifiable (MRV). Existing technologies face multiple challenges in this regard.

[0003] Under the full life-cycle accounting framework, data fragmentation and spatiotemporal mismatch are prominent issues. Existing accounting methods are mostly limited to the combustion stage within the plant, neglecting emissions from upstream raw material collection, transportation, and storage. The biomass supply chain is complex, with differences between short-distance first-mile collection and long-distance trunk transportation in the transportation stage. This results in highly discrete batch characteristics in upstream data, making it difficult to effectively connect with continuous time-series operational data from the power plant. Due to the lack of an effective spatiotemporal mapping strategy, upstream emissions cannot be accurately allocated to the corresponding power generation periods in scenarios with inventory lag or frequent batch switching, causing systematic biases in the accounting results.

[0004] The strong heterogeneity of biomass fuels and the lag in measuring key parameters lead to insufficient completeness of accounting data. Parameters such as moisture and calorific value of biomass feedstocks are significantly affected by origin, season, and storage time, exhibiting far greater volatility than coal. Relying on online monitoring or offline testing after feedstock entry into the furnace is significantly lagging. When component analyzers malfunction or sampling frequency is insufficient, traditional methods lack effective soft-sensoring or other data supplementation methods, affecting the real-time performance and accuracy of accounting. Furthermore, the widely used static emission factor method cannot capture the characteristics of a single batch of feedstock and dynamic emission changes under specific unit loads, resulting in significant biases in carbon asset settlement.

[0005] The predictions of purely data-driven models often conflict with physical consistency requirements. When introducing artificial intelligence algorithms, traditional black-box models may violate the laws of conservation of mass or energy when samples are scarce or operating conditions fluctuate drastically. In carbon trading scenarios, such unexplained numerical drifts are difficult to audit by third-party verification agencies based on physical logic, failing to meet the stringent requirements of the carbon market for data compliance and the chain of evidence. Summary of the Invention

[0006] This invention aims to address the challenges in existing biomass gas-coupled coal-fired power generation carbon emission accounting, including spatiotemporal gaps and mismatches in full lifecycle data, missing and delayed measured key parameters, and insufficient physical consistency and compliance of models. Existing accounting systems often neglect emissions from upstream non-energy activities and lack effective strategies to accurately map discrete batch logistics data from upstream to the continuous operating timeline of downstream units. This results in the inability to accurately trace carbon emissions per unit of electricity generated under conditions of inventory lag or mixed combustion of multiple batches. Furthermore, the strong heterogeneity of biomass fuels causes significant fluctuations in moisture content and calorific value due to variations in origin and storage environment. Online monitoring equipment often experiences data gaps due to malfunctions or insufficient sampling frequency. Traditional static emission factor methods struggle to dynamically characterize actual operating conditions, severely limiting accounting accuracy. In addition, purely data-driven models are prone to outputting predictions that violate the laws of conservation of mass and energy under conditions of scarce samples or transient operating conditions. Moreover, the lack of uncertainty quantification and a complete chain of tamper-proof evidence makes it difficult for accounting results to meet the stringent requirements of the carbon trading market for the Monitoring, Reporting, and Verification (MRV) system in terms of interpretability and auditability.

[0007] To address the above problems, this invention proposes a biomass gas-coupled coal-fired power generation full life-cycle carbon emission accounting method for carbon trading, comprising the following steps:

[0008] S1: Determine the accounting boundary and accounting cycle. The accounting boundary covers the entire life cycle process, including at least the upstream biomass raw material collection, transportation, storage and pretreatment stages, and the downstream biomass gas preparation and coal-fired unit coupled combustion power generation stages. The accounting cycle is a preset cycle to meet the carbon trading declaration or settlement requirements.

[0009] S2: Collect activity data and associate batch identifiers and timestamps with upstream batch data and unit-side operation data to establish a spatiotemporal mapping relationship between upstream batch data and unit-side operation data. The spatiotemporal mapping relationship shall at least include an index key value composed of batch identifiers and timestamps.

[0010] S3: Perform consistency processing and missing data detection on the activity data, generate data quality classification labels and associate them with uncertainty measures;

[0011] S4: When the sampling frequency of at least one of the measured data of syngas components or lower heating value is lower than a preset threshold, a soft measurement model is constructed based on the unit-side operating data, and the estimated value and uncertainty measure are output.

[0012] S5: Construct a material-energy-carbon conservation mechanism model, and calculate the intermediate quantity of mechanism prediction based on the activity data and soft sensor model output. The intermediate quantity of mechanism prediction includes at least the carbon flow and energy flow on the biomass gas side, the carbon flow and energy flow on the coal combustion side, and the equivalent heat input of coupled combustion.

[0013] S6: Construct a data-driven residual correction model, and use the data-driven residual correction model to dynamically correct the intermediate quantity of the mechanism prediction to obtain the corrected intermediate quantity;

[0014] S7: Generate dynamic emission parameters based on upstream batch data, and align the dynamic emission parameters with the corrected intermediate amount by timestamp in combination with the spatiotemporal mapping relationship to generate a full life cycle emission inventory, and generate an MRV data package based on the full life cycle emission inventory.

[0015] Determining the accounting boundaries and accounting cycle includes the following sub-steps:

[0016] S11: Construct a physical boundary model that includes non-energy activity emission sources, and incorporate at least one by-product treatment unit into the accounting boundary, wherein the by-product treatment unit includes at least one of a biomass gasification ash treatment unit and a gas purification waste liquid treatment unit.

[0017] S12: Set a dynamic sliding window aligned with the carbon market settlement window, define the accounting period as a configurable sliding time window, and maintain a synchronized interface with the carbon trading platform's settlement or verification period.

[0018] Establishing the spatiotemporal mapping relationship between upstream batch data and unit-side operational data includes the following sub-steps:

[0019] S21: Assign a unique batch identifier to each batch of raw materials entering the factory, and establish a cross-system index between the batch identifier and the production timestamp;

[0020] S22: Record the time period during which the raw materials corresponding to the batch identifier enter the biomass gas preparation stage or the coal-fired unit coupled combustion power generation stage, forming a batch identifier-time stamp index key value.

[0021] The process of generating data quality grading identifiers and associating them with uncertainty measures includes the following sub-steps:

[0022] S31: Perform statistical analysis and logical verification on the activity data, and mark the activity data as measured data identifier, estimated data identifier, or default data identifier;

[0023] S32: Generate rules for associated uncertainty measures for different data identifiers, and pass the uncertainty measures to key estimates in the life cycle emission inventory and MRV data package according to the error propagation principle.

[0024] The output estimate and uncertainty measure include the following sub-steps:

[0025] S41: Select at least two of the following variables as auxiliary variables: gasifier temperature, pressure, gasifying agent flow rate, and bed pressure drop, and construct a machine learning model to back-infer at least one of the syngas components or lower heating value.

[0026] S42: Introduce an uncertainty estimation mechanism into the soft measurement model. The uncertainty estimation mechanism includes at least one of the following: an uncertainty estimation method based on random deactivation and an uncertainty estimation method based on Bayesian inference, to output a corresponding uncertainty metric.

[0027] The construction of the material-energy-carbon conservation mechanism model includes the following sub-steps:

[0028] S51: Establish a carbon balance equation set based on the law of conservation of mass, wherein the carbon balance equation set includes at least the carbon input of biomass feedstock, carbon from gasification conversion, carbon from bottom ash residue, carbon from fly ash residue, and carbon from flue gas emissions.

[0029] S52: Calculate the sensible heat and chemical heat brought in by biomass gas based on energy conservation, and convert them into equivalent heat input for coupled combustion with the same dimensions as coal.

[0030] The construction of the data-driven residual correction model includes the following sub-steps:

[0031] S61: When training or updating the data-driven residual correction model, a loss function with physical consistency constraints is introduced. The loss function includes a penalty term for violating mass conservation or energy conservation to constrain the intermediate quantity after correction to meet the preset physical consistency.

[0032] S62: Employs a multi-fidelity fusion method, pre-trains the model using simulation data, and updates the model using on-site measured data to achieve cross-domain deviation correction.

[0033] The process of generating a full life cycle emissions inventory and generating an MRV data package based on the full life cycle emissions inventory includes the following sub-steps:

[0034] S71: Generate dynamic emission parameters based on raw material batch parameters; the raw material batch parameters include at least transportation segment data and distinguish between the first kilometer segment and the trunk line segment; the dynamic emission parameters are calculated based on at least two of the following for each batch of biomass raw materials: moisture content, transportation distance, and storage time;

[0035] S72: Based on the spatiotemporal mapping relationship, the discrete emission data of the upstream batch data is mapped to the time axis of the unit-side operating data, and time-series alignment and accumulation are completed with the corrected intermediate quantity;

[0036] S73: Generate an MRV data packet containing batch traceability information, activity data source and processing records, data quality classification identifier, key estimators and uncertainty measures, model version and parameter summary, and accounting trajectory log, and generate verification information, wherein the verification information includes at least a hash digest or digital signature.

[0037] The strategy for establishing feedforward feature coupling between raw material batch parameters and machine learning models includes the following steps:

[0038] S81: Obtain raw material batch parameters and generate a priori feature vector by combining biomass degradation kinetics characteristics. The priori feature vector consists of at least two of the following: moisture content, transportation distance, and storage time.

[0039] S82: The prior feature vector and the unit-side operating features are used together as input variables for the machine learning model. The unit-side operating features include at least gasifier temperature, pressure, gasifying agent flow rate and bed pressure drop.

[0040] S83: Based on the machine learning model, output an estimate and uncertainty measure of at least one of the syngas components or lower heating value.

[0041] The establishment of an uncertainty-based physical constraint compensation strategy includes: when training or updating the data-driven residual correction model, adaptively adjusting the weights of the loss function containing physical consistency constraints based on the data quality classification identifier and uncertainty measure, so that the weights of the data fitting terms are reduced for samples with higher uncertainty measures and the penalty weights for violating mass conservation or energy conservation in the physical constraint terms are increased.

[0042] Compared with the prior art, the present invention has the following significant advantages and beneficial effects:

[0043] This invention achieves precise spatiotemporal fusion of full lifecycle data and adaptive compensation for the strong heterogeneity of biomass fuels. By constructing a spatiotemporal mapping index centered on batch identifiers and timestamps, it effectively breaks down the barriers between upstream discrete logistics data and downstream continuous production data. It is compatible with the differentiated characteristics of first-mile and trunk line transportation, accurately tracing the carbon emissions of each kilowatt-hour of power generation to the specific raw material batch and its entire storage and transportation process, completely eliminating accounting mismatches caused by batch switching and inventory lag. Simultaneously, a feedforward coupling mechanism based on raw material characteristics is established. It uses prior data such as upstream moisture content and storage time to predict the state of fuel entering the furnace, and inputs this as a feedforward feature into the downstream soft measurement model. This helps the system overcome the problems of online detection lag and missing key parameters, significantly improving the real-time accuracy and robustness of carbon accounting under drastic fluctuations in biomass quality and complex operating conditions.

[0044] This invention ensures the physical compliance of accounting results and provides a standardized chain of evidence directly adapted to the carbon market. By introducing a physical constraint compensation strategy based on uncertainty, the system can automatically strengthen the constraints of the laws of mass and energy conservation when data quality deteriorates, forcing the output of purely data-driven models to converge to a manifold that conforms to physical laws. This effectively solves the risk that traditional black-box AI models will output results that violate physical common sense and are unexplainable when samples are scarce. On this basis, the tamper-proof MRV data package generated by the system integrates the entire chain of evidence from batch traceability, model parameters, uncertainty quantification to accounting trajectory. This evidence is then used for ownership verification through hash digest or digital signature technology, forming an auditable digital certificate that can be directly used for carbon trading declaration, registration, and settlement. This significantly reduces the cost and risk of manual accounting and third-party compliance verification for enterprises. Attached Figure Description

[0045] Figure 1 A schematic diagram of the physical boundaries and scenarios for full lifecycle accounting;

[0046] Figure 2 This is a functional module architecture diagram of the accounting system;

[0047] Figure 3 The main flowchart for full life-cycle carbon emission accounting;

[0048] Figure 4 A flowchart illustrating the spatiotemporal mapping between discrete batch data and continuously running data;

[0049] Figure 5 This is a flowchart of soft measurement and correction based on physical constraints and uncertainties. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with the claims and embodiments. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0051] Implementation Method 1

[0052] like Figure 3 As shown, the biomass gas-coupled coal-fired power generation life-cycle carbon emission accounting method for carbon trading includes the following steps:

[0053] S1: Determine the accounting boundary and accounting cycle. The accounting boundary covers the entire life cycle process, including at least the upstream biomass raw material collection, transportation, storage and pretreatment stages, and the downstream biomass gas preparation and coal-fired unit coupled combustion power generation stages. The accounting cycle is a preset cycle to meet the carbon trading declaration or settlement requirements.

[0054] S2: Collect activity data and associate batch identifiers and timestamps with upstream batch data and unit-side operation data to establish a spatiotemporal mapping relationship between upstream batch data and unit-side operation data. The spatiotemporal mapping relationship shall at least include an index key value composed of batch identifiers and timestamps.

[0055] S3: Perform consistency processing and missing data detection on the activity data, generate data quality classification labels and associate them with uncertainty measures;

[0056] S4: When the sampling frequency of at least one of the measured data of syngas components or lower heating value is lower than a preset threshold, a soft measurement model is constructed based on the unit-side operating data, and the estimated value of the corresponding missing item and its uncertainty measure are output.

[0057] S5: Construct a material-energy-carbon conservation mechanism model, and calculate the intermediate quantity of mechanism prediction based on the activity data and soft sensor model output. The intermediate quantity of mechanism prediction includes at least the carbon flow and energy flow on the biomass gas side, the carbon flow and energy flow on the coal combustion side, and the equivalent heat input of coupled combustion.

[0058] S6: Construct a data-driven residual correction model, and use the data-driven residual correction model to dynamically correct the intermediate quantity of the mechanism prediction to obtain the corrected intermediate quantity;

[0059] S7: Generate dynamic emission parameters based on upstream batch data, and align the dynamic emission parameters with the corrected intermediate amount by timestamp in combination with the spatiotemporal mapping relationship to generate a full life cycle emission inventory, and generate an MRV data package based on the full life cycle emission inventory.

[0060] S11: Construct a physical boundary model that includes non-energy activity emission sources, and incorporate at least one by-product treatment unit into the accounting boundary, wherein the by-product treatment unit includes at least one of a biomass gasification ash treatment unit and a gas purification waste liquid treatment unit.

[0061] In practical application scenarios of physical boundary model instantiation and verification, taking a certain biomass gasification coupling as an example... Taking the operation accounting of a coal-fired power generating unit as an example, the system executes a physical constraint compensation strategy based on uncertainty. The accounting engine automatically retrieves the activity data of the biomass gasification ash treatment unit from the industrial time-series database through the monitoring point index, and calculates the amount of ash generated by the gasifier during that period. And the mass fraction of residual carbon in this batch of ash slag, analyzed synchronously through the laboratory analysis system. As an input variable.

[0062] The calculation engine further reads the disposal method configuration field of the node. If it detects that this field is configured for building material utilization, it determines, based on its built-in calculation logic, that the residual carbon within the calculation boundary has not undergone oxidation, and thus automatically calculates the unit's direct carbon dioxide emissions as... The system then displays the measured values. and As a carbon outflow term, it is substituted into the constructed carbon balance equations for consistency verification.

[0063] If the residual carbon loss in the main process chain after solving the equations is greater than a preset threshold, the system determines that there is a deviation between the topology of the physical boundary model and the actual working conditions, and identifies hanging branches, that is, there may be uncaptured carbon flow outside the current boundary. Then, it automatically triggers a topology integrity verification alarm and generates an accounting trajectory log. Through a data-driven closed-loop mechanism, the integrity and accuracy of the full life cycle accounting boundary are ensured.

[0064] Specifically, to achieve digital mapping and traceable accounting of the entire lifecycle accounting boundary, a physical boundary model based on a directed acyclic graph (DAG) topology is instantiated in memory, and the calculation traversal order of carbon emission streams and energy streams is determined based on a topology sorting algorithm. For the physical loops actually existing in the production process, boundary equivalence processing is performed according to a pre-defined accounting rule configuration table. Specifically, for the main return path, a time window slicing strategy is used to treat the current return working fluid as the input for the next calculation window. For secondary paths, a net flow reduction strategy is used. Within the calculation window, the net flow is obtained by subtracting the total return flow integral from the total input flow integral, logically transforming it into an acyclic directed path to ensure the model satisfies the topological constraints of the DAG.

[0065] like Figure 1 As shown, the physical boundary model after the above equivalent processing fully covers the main stages of the entire lifecycle in terms of accounting logic. In the upstream stage, such as... Figure 1 As shown on the left, the boundary extends from the starting point to the biomass raw material collection node, encompassing farmland and woodland, covering a multi-level logistics chain from first-mile transportation using small trucks to trunk transportation using large trucks, ultimately connecting to the storage and pre-processing area within the power plant. In the downstream stages, such as... Figure 1As shown in the right-hand section of the plant, the core conversion process includes the gasification island (biomass gas preparation) and the power generation island (coupled coal-fired power generation). Topologically, the gasification reactor node in the biomass gas preparation stage is defined as a critical convergence node in the main process chain. Branch paths originate from this node: one connects to the coupled combustion node of the coal-fired unit as the main energy flow path, and the other connects to... Figure 1 The by-product treatment unit node below serves as an auxiliary emission flow path. The by-product treatment unit node includes the ash and slag treatment unit and the waste liquid treatment unit, as well as the gasification tar treatment unit, to achieve complete capture of non-main process emissions.

[0066] The system employs a graph-based storage approach that separates configuration models from activity data to support the boundary model. Specifically, it persistently stores the metadata of physical entities through a node configuration table. Fields include a unique node identifier, a node type identifier, a monitoring point index, and an emission factor retrieval key. The monitoring point index is used to associate with the DCS monitoring point index or calculation parameter templates. The enumeration range of the node type identifier is... Figure 1 The physical entities shown correspond one-to-one, covering nodes for raw material collection, transportation, storage and pretreatment, biomass gas production, coupled combustion power generation, and various by-product treatment. Simultaneously, an edge configuration table defines the working fluid flow relationship, with fields including source node identifier, target node identifier, and flow type identifier to distinguish between material flow, energy flow, or waste flow. During system operation, the dynamic calculation logic retrieves timestamped activity data from the time-series data storage based on the monitoring point index in the node configuration table. This activity data includes flow rate, carbon content, and power consumption. Then, based on the emission factor retrieval key, it matches the corresponding dynamic or static factor from the emission factor database to perform emission calculations.

[0067] In particular, such as Figure 1 As shown in the upstream and midstream logistics section, for raw material transportation nodes, the transportation segment type identifier field is preset in the attribute configuration to strictly distinguish between first-mile transportation and trunk line transportation. During the calculation process, the transportation segment type identifier is used as one of the retrieval conditions for emission factor matching to automatically match differentiated transportation emission factors. To form a computable link between objects, data, and parameters, the system uses a data binding module to bind the activity data collection points of the by-product processing unit and the emission factor configuration items to the corresponding node records.

[0068] Specifically, the system retrieves the ash production volume and residual carbon mass fraction from a time-series database. The residual carbon fraction can be obtained from manually entered laboratory values ​​or converted using the loss on ignition rate. The system reads the treatment method configuration and executes logical judgments. If the configuration is for methods involving residual carbon oxidation, such as reburning, incineration, or co-processing, the system calculates the direct carbon dioxide emissions based on the ash production volume, residual carbon fraction, and residual carbon oxidation efficiency. If the configuration is for methods that do not involve significant carbon oxidation, such as building material utilization, landfill, or long-term storage, the direct emissions are set to zero, and the carbon content is included as a carbon outflow item in the balance check. Based on this, the system further collects the power consumption of the treatment equipment and the ash transportation distance, and calculates the indirect power emissions and transportation emissions by combining the grid emission factor and transportation emission factor. The total emissions of the unit are obtained by accumulating the direct emissions, indirect emissions, and transportation emissions.

[0069] Secondly, the system obtains the difference between the wastewater flow rate and the organic load removal amount, i.e., the chemical oxygen demand (COD) of the influent and effluent. To ensure dimensional consistency, the product of flow rate and concentration is first converted into the total COD removal mass using a comprehensive conversion factor, and differentiated calculations are performed based on the process type identifier. For anaerobic processes, a methane emission factor is matched, and the global warming potential of methane is introduced to calculate the carbon dioxide equivalent. For aerobic processes, a nitrous oxide emission factor is matched to calculate the corresponding greenhouse gas emissions. Simultaneously, the indirect electricity emissions from the wastewater treatment equipment are calculated, and when the biogas recovery configuration for replacing fossil fuels is detected and activated, the emission reduction corresponding to the recovered methane is deducted, and the total net emissions of the wastewater treatment unit are obtained.

[0070] To prevent omissions, the system performs topology integrity checks during physical boundary model instantiation. The system first traverses the graph structure. If it detects a biomass gas production node in an active state, but its output path lacks any byproduct processing nodes (i.e., there are dangling material flow branches), then based on mass conservation logic, it determines that the destination of ash or waste liquid is unknown, triggering a boundary missing anomaly. The system will automatically halt the calculation process and send an alarm until maintenance personnel complete the node configuration through the human-machine interface, thus preventing hidden emissions omissions in the auxiliary production process.

[0071] S12: Set a dynamic sliding window aligned with the carbon market settlement window, define the accounting period as a configurable sliding time window, and maintain a synchronized interface with the carbon trading platform's settlement or verification period.

[0072] In this embodiment, to ensure strict consistency between the accounting cycle and the settlement cycle of the carbon trading platform or third-party verification agency, the dynamic sliding time window is set as a configurable accounting cycle. For this window, the system uses an ordered structure for parameterized definition, and the current accounting boundary is thus uniquely locked through the configuration object W.

[0073] ;

[0074] In the formula, This is the start timestamp; End timestamp; This is the sliding step size; This is a configuration item for the alignment strategy. Based on the above definition, the effective accounting time interval generated by the system is a left-closed, right-open interval. :

[0075] ;

[0076] The timestamps all use a unified time base and are consistent with the settlement time zone of the carbon trading platform. The cutoff points of the intervals, such as the end of the month or the end of the year, strictly follow the platform rules.

[0077] The system obtains the set of settlement cut-off points from the carbon trading platform through a parameter synchronization interface or a pre-configured rule configuration file. ,in Representing the The standard timestamp for each compliant settlement date, based on configuration items. Perform boundary constraint solving, The dynamic constraint is a value in the platform's settlement cutoff point set, that is, it satisfies:

[0078] ;

[0079] Regarding parameter conflict handling, the system employs a built-in hierarchical priority strategy. When parameter sources conflict, the priority order is: parameters returned by the platform interface take precedence over manually configured parameters, which in turn take precedence over locally cached default parameters. To ensure continuous availability, when the synchronization interface fails, the system automatically degrades to using the local default configuration and generates a preset level of operational alarm. When the interface recovers, the system automatically updates the configuration and records version change logs, forming a traceable configuration evolution chain. This is implemented at each sliding step. Upon arrival, perform a window update operation:

[0080] ;

[0081] For upstream batch data that crosses the window boundary, let a certain batch... The effective start and end time interval is The system uses the interval intersection algorithm to perform time slicing processing, calculating whether the batch falls within the current accounting window. Valid time period : This indicates the valid start time of the batch of data, marking the beginning of the time period covered by this batch of data. This indicates the valid end time of this batch of data, marking the end of the time period covered by this batch of data.

[0082] ;in, This represents the function that takes the maximum value. This represents a function that takes the minimum value.

[0083] When the synchronization interface detects the arrival of a carbon market settlement or verification node, the system performs a snapshot freeze operation: locking all activity data, intermediate calculation variables, and cumulative emission results within the current sliding window, generating a snapshot of the accounting cycle. To ensure the immutability and auditability of the data, the system performs a digital fingerprint generation operation on the snapshot content:

[0084] ;

[0085] In the formula, This represents a snapshot of the accounting period that includes activity data and cumulative emissions results. It is a hash digest algorithm; This is a digital signature based on a private key. The system will generate a digital fingerprint. Write the snapshot data to WORM (Write Once, Read Once, Read Once) storage area or blockchain notarization node together with the snapshot data, as a trusted base input for the subsequent generation of MRV data packets.

[0086] Establishing a spatiotemporal mapping relationship between upstream batch data and unit-side operational data includes the following sub-steps:

[0087] S21: Assign a unique batch identifier to each batch of raw materials entering the factory, and establish a cross-system index between the batch identifier and the production timestamp;

[0088] like Figure 4 As shown, to achieve the correlation mapping between upstream discrete batch data and unit-side continuous operation data, the system adopts an identifier anchoring and time-series alignment strategy based on a three-lane interaction model. This model establishes three parallel logical lanes to manage heterogeneous data: the left lane manages batch-based static attribute data, generating a globally unique batch identifier for each train or batch of incoming raw materials. And it attaches discrete block data such as supplier information, quality inspection reports, and net weight to this... Under its name; the middle lane, as a bridge connecting the discrete and continuous domains, is responsible for capturing physical action events such as feeding and switching. When the left lane... When an entity enters the production process, record the precise timestamp of that moment. ) and physical location ( Anchor points are formed to lock logical identifiers with physical space and time.

[0089] The right lane is responsible for recording high-frequency streaming data generated by systems such as DCS and SIS. The system executes a timing alignment algorithm based on the anchor point established in the middle lane, monitors the data stream in the right lane, and reads the valid start time of the current batch. and valid end time Perform time-series slicing on a continuous data stream and divide the timestamps All data frames within the system are tagged to bind static batch attributes to dynamic process parameters. Furthermore, the system introduces a hysteresis compensation mechanism based on the physical distance from the feed inlet to the machining center, adjusting the compensation according to the conveyor speed. and distance Automatically add time lag parameters To correct the alignment time window, ensure the physical accuracy of data fusion.

[0090] Through the interactive collaboration of the three swimlanes described above, this implementation method eliminates the data silos between the ERP / MES management layer and the DCS / SCADA control layer, constructing a full lifecycle data chain from raw material entry to finished product output. This not only supports refined single-batch cost accounting but also significantly improves the ability to analyze the root causes of quality issues: when quality anomalies occur, the system can quickly locate and distinguish whether the problem stems from raw material properties or process fluctuations, providing a solid data foundation for production process optimization and end-to-end traceability in discrete manufacturing.

[0091] like Figure 4 As shown in the upstream logistics swimlane on the left, the system assigns a unique batch identifier (BID) to each smallest logistics unit of raw materials entering the plant in a single shipment or batch. At this stage, the data presents a discrete, block-like structure and records the static properties of the raw materials. For example... Figure 4 As shown in the right-hand swimlane for unit operation, the data collected by the DCS and SIS systems presents a continuous time-series format. To establish a correlation between the two, the system... Figure 4 The intermediate processing lanes perform data fusion. First, the system performs NTP or PTP clock synchronization between the upstream management system (covering weighing and metering) and the plant-wide control system (covering production execution). Then, based on a unified time zone, it performs linear correction for cross-system time deviations to form a unified plant-wide production timestamp, thereby constructing an index key-value pair centered on a batch identifier and timestamp. After indexing, the following steps are required: Figure 4The intermediate lane indicates the time period for entering the gasifier to determine the effective time window for each batch of raw materials. Specifically, for any batch, the system defines its effective consumption time period on the production line as a left-closed, right-open interval based on the principle of material conservation. This interval is defined as including the start point but excluding the end point to avoid duplicate assignments at batch switching boundaries. To ensure the accurate acquisition of the above time period, the system uses a combination of cumulative flow integration and event triggering to determine the boundaries. Specifically, the start time is determined by the feeder opening signal or the feed valve switching event. For the end time, the system integrates the instantaneous mass flow rate of the feeder in real time, and the boundary determination is achieved by summing the data over the sampling period in the discrete control system.

[0092] ;

[0093] In the formula The effective start time corresponding to this batch of material delivery is Summing the upper boundary Then the effective end time of the material delivery for this batch is... , for Instantaneous feed rate at any given moment This represents the net weight of the incoming batch. When the cumulative outflow is approximately equal to the net weight of the incoming batch, the system marks the end time and automatically switches to the next batch. This strategy effectively solves the inventory lag problem, and can calculate the accurate furnace entry time even when multiple batches are continuously received. During the mapping process, the system iterates through... Figure 4 The time axis of the unit's operating data on the right, at the sampling time When the data falls within the effective consumption period defined above, the system determines that the unit data at that moment is contributed by that batch.

[0094] S22: Record the time period when the raw material corresponding to the batch identifier enters the biomass gas preparation stage and the coal-fired unit coupled combustion power generation stage, forming a batch identifier-time stamp index key value.

[0095] To address overlapping operating conditions during raw material mixing or switching transitions within silos, the system further introduces a batch contribution coefficient to enhance mapping accuracy and generate data such as... Figure 4The extended index key value shown in the core operation box requires the contribution coefficient to meet a normalization constraint, meaning that the sum of the coefficients of all effective batches participating in the reaction at any given time is 1. This coefficient represents the proportion of a particular batch's mass contribution to the total input material at the current moment. The system performs emission allocation calculations based on this coefficient, multiplying the total emissions at a given time by the corresponding contribution coefficient to obtain the emissions belonging to each batch. In specific calculations, the value of the contribution coefficient can be determined through high-frequency measurement or model estimation. When separate metering is lacking, the system uses a FIFO (First-In, First-Out) model or a linear transition model to simulate the mixing process within the silo, specifically setting the contribution coefficient of the old batch to decrease linearly and the contribution coefficient of the new batch to increase linearly within the transition time window of batch switching. The final system is as follows: Figure 4 The bottom shows how to call the corresponding batch attributes and combine them with the contribution coefficient to generate a time-aligned full life cycle emission inventory, so that the emissions belonging to different batches can be accurately split according to weight and significantly reduce disputes over inter-period allocation.

[0096] The process of generating data quality grading identifiers and associating them with uncertainty measures includes the following sub-steps:

[0097] S31: Perform statistical analysis and logical verification on the activity data, and mark the activity data as measured data identifier, estimated data identifier, or default data identifier;

[0098] Specifically, to quantify the reliability of the accounting results, the system establishes a complete metrological link for quality grading, uncertainty assessment, and error propagation. The system groups the collected activity data by measurement point and timestamp, performs statistical analysis and logical verification, and writes data quality grading identifiers based on the verification results. The system divides the quality level of activity data into three levels and implements a dynamic marking and downgrading mechanism. Specifically, when activity data directly originates from the raw sampling values ​​of a calibrated metrological device or online analyzer, and simultaneously passes dimensional consistency verification, physical value range verification, and time series continuity verification, the system marks it as measured data; when activity data is estimated by soft measurement model output, multi-source data interpolation extrapolation, or data fusion algorithms, and passes data availability verification (i.e., the estimated value exists with a valid confidence interval and the value is within a physically reasonable range), the system marks it as estimated data; when activity data is missing, and the system fills it in based on the default values ​​of the emission factor library, historical statistical benchmarks, or conservative preset rules, it marks it as default data.

[0099] Based on this, the system runs mutation detection and drift detection algorithms in real time for the time series of the same measurement point. Specifically, the system uses the 3-Sigma criterion based on a sliding time window to perform mutation detection, with the current sampling time as the reference point. The length is denoted as The historical reference window, and extract the most recent data within that window. A reference sequence is formed by 10 historical valid sampling points. The system calculates the arithmetic mean of the reference sequence in real time. with standard deviation Construct by subtracting the average The lower limit is calculated as the standard deviation plus the mean. Dynamic confidence interval with 100 standard deviations as the upper limit , where the coefficient The value is typically set to 3 to correspond to extremely low probability events under a normal distribution. The system uses this interval to determine the measured data at the current moment. Whether a non-physical mutation has occurred, if measured data The value exceeding this range satisfies the condition that the absolute deviation between the measured value and the mean is greater than three standard deviations. The system determines that the current measurement point data is abnormal. Once an anomaly is determined, the system immediately triggers a graded degradation strategy, automatically downgrading the quality identifier of the data from the measured data identifier to the estimated data identifier, and calling the soft measurement model output value, multi-source fusion estimated value, or the effective hold value of the previous time point at the same timestamp to replace the abnormal data. At the same time, the system records the sudden change alarm event and the source of the replacement value in the calculation trajectory log for subsequent auditing.

[0100] S32: Generate rules for associated uncertainty measures for different data identifiers, and pass the uncertainty measures to key estimates in the life cycle emission inventory and MRV data package according to the error propagation principle.

[0101] After completing the classification and labeling, the system, based on the principles of metrological uncertainty assessment, assigns differentiated uncertainty assessment rules to data of different quality levels and generates standard uncertainties. The data is stored in the record. Specifically, for the identification of measured data, the system preferably adopts the Type B evaluation method, based on the accuracy class of the measuring instrument and the expanded uncertainty or maximum permissible error provided by the calibration certificate. The standard uncertainty can be calculated by combining it with a Type A assessment based on repeatability statistics; for the estimated data identifier, the system uses the root mean square error of the validation set of the soft sensor model. The uncertainty is determined by the width of the confidence interval of the cross-validation residual or the estimation algorithm output; for the default data identifier, the system sets the uncertainty to ±5% or ±10% based on the confidence interval given by the emission factor library, the data quality level mapping table, or the conservative error range specified by industry standards.

[0102] Based on the error propagation principle, the system propagates the uncertainty of each activity's data to the key estimator, namely, the total life-cycle greenhouse gas emissions E, in relation to the function E. ,in Representing the nth input parameter involved in the emission calculation, the system preferentially uses the sensitivity propagation law based on the first-order Taylor series expansion to calculate the combined standard uncertainty. :

[0103] ;

[0104] In the formula, Input variables Standard uncertainty; The sensitivity coefficient is obtained through analytical differentiation or numerical difference. The covariance is the variance between variables. When the correlation parameter is not specified in the system configuration table, the input quantities are assumed to be independent, i.e., the covariance term is zero.

[0105] To ensure computational accuracy under extreme conditions, the system incorporates an adaptive algorithm switching strategy. When the system detects that the input quantity exhibits a significant non-normal distribution or that the function has strong nonlinearity, it automatically switches to the Monte Carlo (MCM) method for error propagation. Specifically, the system adjusts the algorithm based on the input activity data. Construct the corresponding probability density function based on historical statistical distribution characteristics or metrological certificate parameters. The pseudo-random number generator is invoked to generate a sample size of [size missing] for each input variable based on its probability density function. A random sampling sequence, in which Usually set to This process is repeated at least once to ensure convergence. The system then substitutes each of the generated input variable sampling sequences point-by-point into the life-cycle emission calculation model. Perform iterative simulation calculations to obtain the results containing The output sample set of the calculation results. To ensure calculation accuracy under extreme conditions, when the system detects that the input quantity exhibits a significant non-normal distribution or the function has strong nonlinearity, it automatically switches to the Monte Carlo method to perform error propagation. This is based on the previously constructed simplified life-cycle carbon emission calculation model. For example, among which For emissions, Biomass consumption It has a low calorific value. The system first constructs probability density functions for each variable based on historical data, setting the consumption amount as the emission factor. Follows a normal distribution tons, lower calorific value Because raw material fluctuations follow a triangular distribution Emission factors Follows uniform distribution Subsequently, the system calls a pseudo-random number generator to generate the sample size. A random sampling sequence is used, and point-by-point substitution calculations are performed. In the... Substituting the sampled values ​​into the simulation Calculated In the Substituting into the next simulation Calculated This process is repeated tens of thousands of times. Finally, the system performs statistical analysis on the output sequence and calculates the mean. As the best estimate, and extract the sequence. quantiles )and quantiles As Confidence interval, in The data packet output has a defined uncertainty range The accounting results.

[0106] For this output sample set, the system performs statistical analysis to calculate the arithmetic mean of the output quantities as the best estimate, and then sorts the output sample set in ascending order and extracts the values ​​from the sorted samples. and The value at the quantile directly defines the probability of inclusion. The confidence interval is determined. Finally, the expanded uncertainty of the system is calculated. This value is equal to the inclusion factor. Multiply by the combined standard uncertainty ,in Typically, a value of 2 is used to correspond to approximately a 95% confidence level; the system will include key estimators and expanded uncertainties. The quality classification identifier of the original data is written into the MRV data packet, thereby forming a consistent evidence chain of data quality classification, uncertainty quantification, and confidence of accounting results.

[0107] The output includes the following sub-steps: the estimated value of the missing term and its uncertainty measure.

[0108] S41: Select at least two of the following variables as auxiliary variables: gasifier temperature, pressure, gasifying agent flow rate, and bed pressure drop, and construct a machine learning model to back-infer at least one of the syngas components or lower heating value.

[0109] Specifically, such as Figure 5 As shown, the system constructs a fusion computing layer comprising a soft sensor model and a mechanistic model to handle multi-source heterogeneous input data. When high-frequency measured data on syngas components or lower heating values ​​are missing, the system constructs a soft sensor model based on unit-side operating data to output estimated values ​​of the missing terms and their corresponding uncertainty metrics. For scenarios with missing measured data on syngas components or lower heating values, the system employs a soft sensor strategy based on deep neural networks for data imputation. In the model construction phase, the system first performs heterogeneous data alignment, starting from... The system obtains the true values ​​of laboratory tests and from... Database captures data up to this point in time. Stable operating conditions for minutes The training process uses the running features as input samples and includes two hidden layers, each containing... and one neuron Model. During real-time operation, when a fault is detected in the online analyzer, the system automatically collects current instantaneous operating data and inputs it into the model. The model then outputs a low-mounted calorific value estimate through forward inference. and utilize The standard uncertainty is calculated by performing multiple random inferences. The system ultimately combines this estimate with the uncertainty to form... By incorporating data into the emissions inventory and labeling it as estimated data, the continuity and traceability of carbon emissions accounting are ensured in the absence of hard measurements.

[0110] The system monitors the status and data stream of the online analyzer in real time. It will detect when a device fault flag is set or the number of consecutive missing data points exceeds a threshold. Or the data missing rate within a unit time window exceeds the threshold. At this time, the soft measurement completion process is triggered. To construct training samples, the system adopts a heterogeneous data spatiotemporal alignment strategy, selecting auxiliary variables including gasifier temperature, pressure, gasifying agent flow rate, and bed pressure drop from the unit-side operating data to form a high-frequency feature flow, and defining the intermittently acquired measured values ​​of syngas components, i.e., laboratory chromatographic analysis results or low-frequency outputs of online analyzers, as sparse label vector z. .

[0111] Label It is a collection A real-valued vector representing the content of each component. The system uses sparsely labeled sampling timestamps. Use this as the anchor point to capture a preset time window preceding that moment. Dense feature data within, The time window length refers to the time span over which feature data is traced backward. For the first The time points from which sparse labels are generated are considered. Steady-state segments are then selected based on preset variance or slope thresholds. Statistical convergence operations are performed on the features within these steady-state segments to extract the mean and variance, which are then used as aligned input features. This allows for the construction of a training sample set with input and output time aligned.

[0112] During the model building phase, the system preferentially adopts a deep neural network (DNN) as the basic architecture to achieve the goal of learning from operational features. To target variable The model employs a nonlinear mapping. It constructs a multilayer perceptron structure comprising an input layer matching the dimension of the aggregated feature vector, several hidden layers for deep feature extraction, and an output layer corresponding to the dimension of the target variable. Each hidden layer uses nonlinear activation functions such as ReLU (Rectified Linear Unit) to fit the strong nonlinear coupling relationship between the unit operating parameters and syngas components. Dropout layers are embedded between the hidden layers to randomly freeze neuron connections at a predetermined ratio during training and inference, providing a structural basis for subsequent uncertainty estimation based on Bayesian approximation.

[0113] To fully utilize historical data of different quality levels, a weighted loss function is constructed. Furthermore, a sample weighting strategy based on label data quality grading is introduced:

[0114] ;

[0115] In the formula: For model parameters; It is a deep neural network; For the first The confidence weight of each sample is determined by the label. The quality identifier determines the weight of the data. Measured data identifiers have high weights, while estimated and default data identifiers have low weights, thus increasing the contribution of high-quality samples to the loss function. The system records the model version number, training data time range, and key hyperparameter summaries into the accounting trajectory log. It also periodically calculates the root mean square error (RMSE) or population stability index (PSI) on the validation set. When the monitored index exceeds the drift threshold, it automatically triggers model retraining or incremental updates.

[0116] S42: Introduce an uncertainty estimation mechanism into the soft measurement model. The uncertainty estimation mechanism includes at least one of the following: an uncertainty estimation method based on random deactivation and an uncertainty estimation method based on Bayesian inference, to output a corresponding uncertainty metric.

[0117] During the online inference phase, the system utilizes the constructed deep neural network architecture to perform uncertainty estimation. The system preferably employs Monte Carlo Dropout (MC Dropout) to keep the randomly deactivated units in the network active during inference, for the same input. Configurable number of executions For example, after at least 20 random forward computations, a set of predicted sample sequences is obtained. The system calculates the mean of this sequence as the best estimate. The variance of the sequence is calculated to characterize cognitive uncertainty, and the inherent observation noise variance obtained from validation set residual analysis is superimposed. Thus, the total combined uncertainty is obtained. :

[0118] ;

[0119] in The final combined uncertainty is given by the left-hand side of the formula. That is, the total variance. The number of inference samples for Monte Carlo random inactivation. For the first A single prediction value obtained from random forward inference. For all The arithmetic mean of the prediction results. This represents the inherent observation noise variance.

[0120] Furthermore, when the input features of the soft measurement model When the input variable contains a variable that has been marked as an estimated data identifier or a default data identifier, the system executes the input uncertainty propagation logic. The system reads the standard uncertainty associated with the input variable, calculates the influence component of the input error on the output based on the error propagation principle or first-order sensitivity analysis, and combines it with the total combined uncertainty of the model output. The uncertainty measure of the final output is amplified and corrected by performing a quadratic summation and synthesis. The corrected estimate and its uncertainty measure are then written into the input data structure of the subsequent mechanism model and the full life cycle accounting link along with the timestamp, ensuring the continuity of uncertainty information in the gap filling stage of measured data.

[0121] The construction of the material-energy-carbon conservation mechanism model includes the following sub-steps:

[0122] S51: Establish a carbon balance equation set based on the law of conservation of mass, wherein the carbon balance equation set includes at least the carbon input of biomass feedstock, carbon from gasification conversion, carbon from bottom ash residue, carbon from fly ash residue, and carbon from flue gas emissions.

[0123] Specifically, the system establishes a three-dimensional conservation mechanism model of materials, energy, and carbon elements for the biomass gas production and coal-fired power generation coupling link, and uses discrete timestamps. The index performs standardized calculations for each physical quantity, unifying its dimensions and reference state. The discrete timestamp... Sampling period configured by the system For example, generation occurs every 1 minute or 5 minutes, and conservation calculations are performed in each... The system operates independently. It uses activity data that has been cleaned and graded according to quality standards and the output of the soft sensor model as input variables. When key parameters such as syngas composition or lower heating value are missing, the system directly calls the estimated values ​​and their uncertainty measures output by the soft sensor model to participate in the conservation calculation, and updates the composite uncertainty of the output quantity synchronously based on the error propagation principle.

[0124] The system at each timestamp A carbon conservation relationship is established, and the carbon input, carbon output, and residual terms are combined into a carbon balance equation system to generate the carbon flow on the biomass gas side and the carbon flow on the coal combustion side. Physical consistency verification is performed on key measurement points. Specifically, the system is based on the mass flow rate of the fuel entering the furnace. Carbon content with elemental analysis Calculate biomass-side carbon input Carbon input from coal-fired side :

[0125]

[0126] In the formula: and For real-time measurement of mass flow rate; and To correspond to the carbon mass fraction of the fuel, for coal quality data with low testing frequency, the system updates the data according to the latest batch test values ​​and takes a fixed value within its corresponding time period.

[0127] To address the solid residue from the gasification process, the system collects the mass flow rate of the bottom ash. Mass flow rate of fly ash And its corresponding residual carbon mass fraction. Calculate the gasification residual carbon output. :

[0128] ;

[0129] In the formula and For real-time measurement of bottom ash and fly ash mass flow rate; and The corresponding residual carbon mass fraction is obtained through laboratory testing or online ash and carbon meters; if it is an offline test value, it remains unchanged according to the latest batch.

[0130] For end-of-pipe emissions, the system is based on the standard flue gas volumetric flow rate monitored by CEMS. With carbon dioxide volume fraction Utilizing carbon molar mass Compared with standard molar volume The carbon emissions from the flue gas side were calculated. :

[0131] ;

[0132] In the formula: The unit is standard cubic meters per hour; For example, 22.414 standard cubic meters per kilomolar; The molar mass of carbon is 12 kilograms per kilomolar.

[0133] Based on this, the system employs a dual-path approach to calculate and verify the converted carbon content of the gasification subsystem. The first path, based on the solid-phase mass balance principle, defines the theoretical gasification conversion carbon content from the solid phase to the gas phase and ultimately into the coupled combustion link. The first approach is to subtract the residual carbon output from the gasification carbon input from the biomass carbon input. The second approach, based on gas phase component monitoring, directly calculates the gas phase carbon flow according to the syngas flow rate and the concentration of carbon-containing components. :

[0134] ;

[0135] In the formula: This represents the standard volumetric flow rate of syngas. It is a collection of carbon-containing components, including carbon monoxide, carbon dioxide, and methane, etc. Components Volume fraction; Components The number of carbon atoms in a molecule.

[0136] System construction of normalized total carbon conservation residuals Used to characterize physical consistency:

[0137] ;

[0138] In the formula: To prevent small amounts with a denominator of zero, when the residual exceeds a preset physical constraint threshold, the system marks the timestamp as a physical consistency anomaly and outputs the anomaly information along with relevant data quality classification identifiers, providing a strong constraint basis for subsequent residual correction models. The various carbon flow data calculated from the above equations constitute the carbon flow component in the intermediate quantities of mechanism prediction.

[0139] S52: Calculate the sensible heat and chemical heat brought in by biomass gas based on energy conservation, and convert them into equivalent heat input for coupled combustion with the same dimensions as coal.

[0140] The system calculates the chemical and sensible heat brought into the boiler by biomass gas based on the law of conservation of energy, and converts them into equivalent heat input for coupled combustion with the same dimensions as coal. The system first calculates the chemical heat input of the syngas. It is equal to the syngas mass flow rate. With low heating value The product of the components; when LHV is missing, the output value of the soft-sensor model is used directly; when only volumetric flow rate is provided on-site, the system converts it to mass flow rate based on component density. Simultaneously, the system is based on syngas temperature. Compared with reference temperature For example, a 25-degree Celsius difference and the specific heat capacity of syngas at constant pressure. Calculate sensible heat input Thus, the total heat input of the syngas is obtained. :

[0141] ;

[0142] In the formula: for Total heat input of synthesis gas at time It is a chemical heat input, which is equal to the syngas mass flow rate. With low heating value The product; Sensible heat input; It can be calculated by weighting the specific heat capacity of each component proportionally or by taking an empirical constant from engineering. This refers to the real-time temperature of the synthesis gas. The reference temperature is used. To achieve a normalized comparison between biomass gas and coal in terms of energy, the system uses the received lower heating value of the coal fed into the furnace. The total heat input of syngas is converted into equivalent coal heat input. :

[0143] ;

[0144] The system simultaneously calculates the chemical heat input on the coal-fired side. and input the total heat of the synthesis gas. Coal-fired heat input With equivalent coal heat input Together, they serve as the energy flow component output in the intermediate quantities of mechanism prediction, and are used for subsequent residual correction and time-series alignment and emission allocation calculations in the full life cycle accounting link.

[0145] The construction of the data-driven residual correction model includes the following sub-steps:

[0146] S61: When training or updating the data-driven residual correction model, a loss function with physical consistency constraints is introduced. The loss function includes a penalty term for violating mass conservation or energy conservation to constrain the intermediate quantity after correction to meet the preset physical consistency.

[0147] Specifically, based on the material-energy-carbon conservation mechanism model, the system introduces a data-driven residual correction model to dynamically correct intermediate quantities in mechanism predictions, thereby compensating for systematic biases caused by mechanism simplification, parameter drift, or unmodeled dynamic characteristics. The system denotes intermediate quantities in mechanism predictions as multidimensional vectors. Its components include at least the carbon and energy flows from the biomass gas side, the carbon and energy flows from the coal combustion side, and the equivalent heat input from coupled combustion. The system synchronously acquires high-confidence field measurement data and aligns it to discrete timestamps using a timestamp anchoring and window aggregation strategy. To form a supervisory signal vector Considering that the field measurement data may only cover some intermediate dimensions, the system introduces an observation selection matrix. To construct residual label vectors under sparse supervision :

[0148] ;

[0149] Based on this, a training sample set containing input features and residual labels is constructed. System training data drives the residual correction model. This establishes a nonlinear mapping from the system state to the correction increment. The input feature vector of this model... It includes not only the current unit-side operating data, soft measurement model output and its uncertainty measurement, and data quality classification indicators, but also the aforementioned variables in... The model uses sliding statistical features, such as mean and variance, within a pre-defined time window to capture the dynamic temporal characteristics of the system. The model output is an estimate of the correction increment. The system calculates the corrected intermediate quantity based on this. :

[0150] ;

[0151] In the formula: These are model parameters; This is a correction intermediate vector after both physical and data constraints, used for timing alignment accumulation and MRV data packet generation in subsequent steps.

[0152] To avoid the output deviating from physical laws due to purely data-driven correction, the system performs parameter training or updates. When introducing a loss function with physical consistency constraints, the intermediate quantity after forced correction is... The system satisfies the preset mass and energy conservation constraints. The system will use the loss function... Defined as a weighted combination of the data fitting term and the physics penalty term:

[0153] ;

[0154] Among them, the data fitting term The weighted mean square error form is used to constrain the learning accuracy of the measured residuals. Physical penalty term. and Based on the established conservation equations, the system will construct the corrected intermediate quantities. Substitute the values ​​into the conservation equations to calculate the physical residuals, and then perform normalization to eliminate dimensional differences:

[0155] ;

[0156] In the formula: and These are the absolute conservation residual functions obtained from the carbon balance equation and the energy balance equation, respectively. and For example, a benchmark value or historical statistical variance for the corresponding physical quantity; and As a penalty weight, the system can adaptively adjust it based on the level of the conserved residual at the current moment or the uncertainty measure of the input data, thereby achieving a dynamic balance between data fitting accuracy and physical consistency.

[0157] Select A set of historical operating data at a given time was used as a sample, in which the amount of coal fed into the furnace was... Syngas flow rate is Based on the above inputs, the theoretical carbon emission rate calculated by the mechanistic model is: The actual carbon emission rate measured by the on-site CEMS system was: At this point, the target residual that the model needs to learn is .

[0158] The data-driven model uses the current coal quantity, gas quantity, and furnace temperature as input vectors. Output predicted residuals Assume that in the early stages of training iterations, the predicted residual output by the model is... At this point, the corrected total emissions become The system immediately performs a physical consistency check: calculations revealed... carbon outflow Exceeding the total carbon input of fuel This violates the law of conservation of mass. At this point, the loss function... The physical constraint penalty in the code is activated:

[0159] ;

[0160] in, For mean square error loss, For the quantity that violates conservation, These are preset large weighting coefficients. This will generate a huge gradient signal, forcing the model parameters to correct the predicted values ​​downwards in the next iteration until the predicted residuals are close to the true values. And satisfy The physical boundary.

[0161] S62: Employs a multi-fidelity fusion method, pre-trains the model using simulation data, and updates the model using on-site measured data to achieve cross-domain deviation correction.

[0162] To address the cross-domain bias caused by distribution differences between simulation data and field-measured data, the system employs a multi-fidelity fusion strategy to train the model. First, the system generates a simulation dataset covering a wide range of operating conditions using thermodynamic simulation software. This dataset is then used to pre-train the model to obtain initial parameters with basic representational capabilities. Subsequently, the model is incrementally updated using a small, high-confidence field-collected dataset. During the incremental update phase, the system employs distribution alignment constraints or sample reweighting mechanisms to correct statistical distribution biases between simulation and field data, ensuring that the model parameters are adapted to actual field conditions.

[0163] During system operation, an online update trigger mechanism is established. When the cumulative amount of newly added field-measured data meets preset conditions, or when the physical conservation residual continuously exceeds a preset alarm threshold, the system triggers the online update process. To ensure system stability, the update process operates in shadow mode, meaning the new model is tested in parallel in the background and only switched to the main model after verification. After each update, the system records the model version number, training data summary, training data hash value, key hyperparameters, and update timestamp, using this information as part of the model evidence chain for subsequent MRV data packets to ensure the traceability and tamper-proof capability of the calculation results. In the model architecture and simulation pre-training phase, a deep neural network (DNN) with three hidden layers (64 neurons per layer) is constructed, with the input vector... This includes gasifier temperature, pressure, and syngas flow rate. Using Aspen Plus thermodynamic simulation software, in... to 50,000 sets of simulation samples under ideal operating conditions were generated within a wide temperature range. Data from a single sample (such as temperature) was then used to generate these samples. ,flow Substituting into the model, the theoretical carbon emission rate obtained from simulation calculations is... The model was initially trained on a large amount of such data, learning the thermodynamic mapping relationship under ideal conditions, and establishing the model's basic weight parameters. .

[0164] During the fine-tuning phase of the measured data, in actual operation, due to equipment aging or incomplete combustion, similar issues also arise. Under operating conditions, the actual emission rate measured by the on-site CEMS system is ,exist Systematic bias. Only 500 sets of high-confidence field measurement data were collected within the most recent week. A small learning rate was used, such as... For the pre-trained model parameters Fine-tuning is performed. The model quickly captures the deviation characteristics in the measured data and updates the parameters accordingly. This allows the system to output correction values ​​that match actual field conditions. When the system detects new operating conditions, a purely mechanistic / simulation model might predict emissions of... The model trained using multi-fidelity fusion will automatically incorporate the learned bias characteristics to output corrected predictions. This allows for the achievement of high-precision accounting goals using a small amount of high-cost data on the basis of low-cost data.

[0165] The process of generating a full life cycle emissions inventory and generating an MRV data package based on the full life cycle emissions inventory includes the following sub-steps:

[0166] S71: Generate dynamic emission parameters based on raw material batch parameters; the raw material batch parameters include at least transportation segment data and distinguish between the first kilometer segment and the trunk line segment; the dynamic emission parameters are calculated based on at least two of the following for each batch of biomass raw materials: moisture content, transportation distance, and storage time;

[0167] Specifically, the system generates dynamic emission parameters based on upstream batch data and, combined with spatiotemporal mapping relationships, synchronously maps and accumulates the dynamic emission parameters with corrected intermediate quantities according to timestamps, thereby generating a full life cycle emission inventory and outputting an MRV data package that meets carbon trading verification requirements. The dynamic emission parameters are stored and used in calculations in the form of batch-level structured records. Their content includes at least a set of batch-level life cycle inventory (LCI) entries bound to the batch identifier, and the batch-level dynamic emission factor and its composite uncertainty calculated from the LCI entries. Each batch record carries raw material batch parameters, a calculation caliber version number, and a data quality classification identifier to support subsequent verification and traceability.

[0168] The system collects a set of raw material batch parameters for each batch, including at least transportation segment data, moisture content, transportation distance, and storage time. To improve accounting accuracy, the system divides the transportation process into the raw material collection segment (the first kilometer) and the main transportation segment, calculating the carbon emission components for each segment separately. The system represents the upstream emissions of a batch as the sum of the transportation component, the storage component, and the optional upstream pretreatment component.

[0169] Specifically, the transport volume Calculate separately for each section:

[0170] ;

[0171] In the formula: Net weight of raw materials for the batch; and These refer to the transportation distances for the collection section and the main line section, respectively. and This refers to the transportation emission factor that is matched to the vehicle type and road conditions for the corresponding section.

[0172] For the stored portion, the system determines the dry matter loss rate function based on storage time and raw material moisture content, calculates the dry matter loss or greenhouse gas emissions of the batch of raw materials during storage, and multiplies it by the equivalent emission factor of the biomass degradation process to obtain the storage emissions. For the pre-treated portion, when the system detects that the batch has undergone pre-treatment processes such as crushing, drying, or packaging and collects the power or fuel consumption, it calculates the pre-treatment emissions based on the emission factor of the corresponding energy source; otherwise, this item is set to zero.

[0173] Based on this, the system generates a batch-level dynamic lifecycle inventory set of entries, and then sums the transportation, storage, and pretreatment components and divides the sum by the net mass of the batch raw material to obtain the batch-level dynamic emission factor per ton of raw material in tons of carbon dioxide equivalent. The system writes the above calculation results into the dynamic emission parameter database and records its calculation rule identifier, factor source identifier, and combined uncertainty calculated by the uncertainty propagation formula.

[0174] S72: Based on the spatiotemporal mapping relationship, the discrete emission data of the upstream batch data is mapped to the time axis of the unit-side operating data, and time-series alignment and accumulation are completed with the corrected intermediate quantity;

[0175] The system utilizes a spatiotemporal mapping index key composed of batch identifiers and timestamps to map discrete emission data of upstream batches to the unit-side operating data timeline, and aligns and accumulates it with the corrected intermediate quantities at the same timestamp. To ensure the auditability and conservation consistency of the allocation, the system constructs a mass sharing weight for each valid batch under each discrete timestamp. This weight must satisfy the normalization constraint, that is, at any given time, the sum of the weights of all valid batches participating in coupled combustion equals 1; simultaneously, this weight must satisfy the batch mass conservation constraint across the time dimension, that is, the integral of the consumption of this batch over its entire life cycle (i.e., the time-cumulative sum of the product of the total mixed fuel flow, the sharing weight, and the sampling period) is approximately equal to its net mass upon entry into storage.

[0176] Accordingly, batch-level upstream emission factors are mapped to time-series interval upstream emissions, and the corrected intermediate values ​​are converted into in-plant process emissions within the same sampling period. The system synthesizes full life-cycle emission inventory entries under the same timestamp. :

[0177] ;

[0178] In the formula: For a moment Total mass flow rate of the mixed fuel entering the furnace; The sampling period; Assign weights to batches; For batch dynamic emission factors, for The collection of all effective biomass and coal batches that participate in coupled combustion at all times;

[0179] The system writes the allocated weight, corresponding batch set, and mapping basis for each timestamp into the accounting trajectory log. The system accumulates the inventory items within the accounting period by timestamp to obtain the cumulative emissions for the accounting period, and further aligns them with the time series of power generation during the same period to calculate the life-cycle greenhouse gas emission intensity per unit of power generation, forming the statistical output required for carbon trading declaration.

[0180] S73: Generate an MRV data packet containing batch traceability information, activity data source and processing records, data quality classification identifier, key estimators and uncertainty measures, model version and parameter summary, and accounting trajectory log, and generate verification information, wherein the verification information includes at least a hash digest or digital signature.

[0181] After generating the full lifecycle emissions inventory, the system constructs an MRV data package to meet the verification requirements for the integrity and traceability of the evidence chain. The MRV data package includes batch traceability information, activity data source and cleaning records, raw data with data quality grading identifiers, uncertainty measures, key estimators, model version and parameter summaries, and an accounting trajectory log containing allocated weights and intermediate result summaries.

[0182] To achieve tamper-proof verification, the system performs normalized serialization on core objects within the MRV data packet, converting the data objects into unique byte sequences. A hash digest is calculated from the normalized sequence, and the system digitally signs the digest using the private key of the accounting entity stored in the secure hardware or key management module. The system then writes the hash digest and digital signature into the MRV data packet's metadata area. This allows third-party verification agencies to verify the signature and recalculate the hash using a public key without accessing the original system environment, thus completing data consistency verification and audit review.

[0183] The strategy for establishing feedforward feature coupling between raw material batch parameters and machine learning models includes the following steps:

[0184] S81: Obtain raw material batch parameters and generate a priori feature vector by combining biomass degradation kinetics characteristics. The priori feature vector consists of at least two of the following: moisture content, transportation distance, and storage time.

[0185] Specifically, to enable batch-to-batch differences in upstream feedstocks to feedforward and constrain the soft-sensor estimation of syngas composition or lower heating value, the system establishes a feedforward feature coupling mechanism between feedstock batch parameters and the soft-sensor model. The system extracts batch parameters for each batch of feedstock and constructs a batch-level prior feature vector. Then, based on the spatiotemporal mapping relationship and mass-allocated weights, this vector is mapped to the unit-side continuous time axis to obtain a comprehensive prior feature vector. This comprehensive prior feature vector is then fused with the unit-side operating feature vector as input to the soft-sensor model to output estimated values ​​for missing terms.

[0186] The system acquires a set of raw material batch parameters for each batch of raw materials, including at least moisture content, transportation distance, and storage time. To characterize the impact of storage decay on convertible components and calorific value, the system introduces a kinetic correction term and forms a batch-level prior feature vector. Preferably, the system constructs a correction formula based on an exponential decay model or the Arrhenius equation, introducing a baseline coefficient of degradation rate and a moisture content influence coefficient obtained from laboratory measurements or fitting of historical operating data. When the above-mentioned refined coefficients are lacking, the system can degrade to using the product of storage time and moisture content as the storage decay state variable in the modeling. The system combines the normalized moisture content, transportation distance, storage time, and the kinetic correction term to form a batch-level prior feature vector, thereby transforming discrete batch attributes into computable physical prior features.

[0187] S82: The prior feature vector and the unit-side operating features are used together as input variables for the machine learning model. The unit-side operating features include at least gasifier temperature, pressure, gasifying agent flow rate and bed pressure drop.

[0188] To align discrete batch characteristics with continuous operational data from the unit side, the system at any timestamp Determine the effective batch set for combustion or gasification, and read the calculated mass allocation weights. The system's prior feature vectors for valid batches Perform weighted synthesis to obtain timestamps. The comprehensive prior feature vector :

[0189] ;

[0190] In the formula: for The set of all effective biomass and coal batches participating in coupled combustion at any given time, where the sum of the weights of all effective batches at that time is 1.

[0191] S83: Based on the machine learning model, output an estimate and uncertainty measure of at least one of the syngas components or lower heating value.

[0192] The system extracts the unit-side operating data into an operating feature vector, whose components include at least gasifier temperature, pressure, gasifying agent flow rate and bed pressure drop. It then uses a splicing method to fuse this feature vector with the comprehensive prior feature vector to construct a high-dimensional input vector containing "unit status + feedstock status". This input vector is then fed into the soft measurement model to output estimated values ​​of syngas composition or lower heating value.

[0193] To suppress physical violations caused by data noise, the system introduces a monotonicity constraint regularization term during the soft measurement model training phase, ensuring that the low-order heat value output by the model is consistent with the actual physical properties. With other operating parameters remaining constant, the system does not increase with increasing moisture content or storage time. The system is constructed with a monotonic penalty term. And add it to the total loss function:

[0194] ;

[0195] In the formula: For moisture; Storage time; The function is used to filter out positive gradient directions that violate physical laws. Through the above-mentioned feedforward coupling and physical monotonicity constraints, the system can still output stable and interpretable estimation results under the constraints of raw material state, even when measured data is missing or sparse, and maintain logical consistency from batch parameters to calculation results.

[0196] The establishment of a physical constraint compensation strategy based on uncertainty includes:

[0197] When training or updating the data-driven residual correction model, the weights of the loss function containing physical consistency constraints are adaptively adjusted based on the data quality classification and uncertainty measure. This reduces the weight of the data fitting term for samples with higher uncertainty measures and increases the penalty weight for violating mass conservation or energy conservation in the physical constraint term.

[0198] Specifically, to ensure that the data-driven residual correction model still meets the preset mass and energy conservation consistency in operating conditions where measured data is scarce or of low quality, the system establishes a physical constraint compensation mechanism based on uncertainty. The system first assigns each timestamp... The following sample association data quality classification label With normalized uncertainty measure .in, Divide the absolute uncertainty index by the corresponding characteristic reference value This eliminates the dimensional differences. Based on this, the system constructs an adaptive weight scheduling strategy with high uncertainty, weak fitting, and strong physical constraints, dynamically adjusting the weights of the data fitting term and the physical penalty term in the loss function.

[0199] The system defines the training objective of the data-driven residual correction model as a weighted physical consistency loss function, which consists of a data fitting term and a physical constraint term, with the weights of each term dynamically changing with sample uncertainty. The weights of the data fitting term are... The system constructs its normalized uncertainty measure. The negative correlation mapping relationship is established, and a numerical upper limit is set to prevent gradient explosion:

[0200] ;

[0201] In the formula: The preset weight truncation threshold; To prevent smooth terms with a denominator of zero; For discrete confidence coefficients determined based on data quality grading identifiers, when The formula uses the larger value for actual data labels, the median value for estimated data labels, and the smaller value for default data labels. This formula automatically reduces the contribution of high-uncertainty samples to gradient descent.

[0202] For the weight of physical constraint terms The system constructs its normalized uncertainty measure. Positive correlation mapping relationship:

[0203] ;

[0204] In the formula: As the physical constraint benchmark weight; This is an enhancement coefficient. When sample uncertainty is high or measured data is missing, Automatically increasing the loss function shifts the dominant term from data fitting error to physical conservation residuals, forcing the model to converge to a feasible solution space that satisfies mass and energy conservation constraints in regions lacking data support.

[0205] To further ensure the compliance of the output results, the system introduces an optional physical feasible region projection step during the model inference phase. When a residual exceeding a preset adaptive tolerance is detected that a corrected intermediate quantity violates physical constraints, the system will proceed accordingly. At that time, the system outputs the model value. Using initial values, solve the constrained optimization problem to obtain the final output. :

[0206] The constrained optimization problem is represented as:

[0207] ;

[0208] In the formula: The mass and energy conservation residual function vector; The allowable physical tolerance can be set as a constant or vary with uncertainty. Dynamic relaxation; These are non-negativity engineering physical boundary constraints. The Euclidean norm of a vector is the square root of the sum of the squares of the magnitudes of its components.

[0209] Example 2

[0210] like Figure 2 As shown, the biomass gas-coupled coal-fired power generation life-cycle carbon emission accounting system for carbon trading includes:

[0211] The accounting boundary and accounting cycle management module is used to determine the accounting boundary and accounting cycle. The accounting boundary covers the entire life cycle process, including at least the upstream biomass raw material collection, transportation, storage and pretreatment stages, and the downstream biomass gas preparation and coal-fired unit coupled combustion power generation stages. The accounting cycle is a preset cycle to meet the carbon trading declaration or settlement requirements.

[0212] The data acquisition and spatiotemporal mapping module is used to collect activity data and associate batch identifiers and timestamps with upstream batch data and unit-side operation data, and establish a spatiotemporal mapping relationship between upstream batch data and unit-side operation data. The spatiotemporal mapping relationship includes at least an index key value composed of batch identifiers and timestamps.

[0213] The data quality management module is used to perform consistency processing and missing data detection on activity data, generate data quality classification labels and associate them with uncertainty measures;

[0214] The soft measurement model module is used to construct a soft measurement model based on the unit-side operating data when the sampling frequency of at least one of the measured data of syngas components or lower heating value is lower than a preset threshold, and output the estimated value of the corresponding missing item and its uncertainty measure.

[0215] The Material-Energy-Carbon Conservation Mechanism Model Module is used to construct a material-energy-carbon conservation mechanism model. Based on activity data and soft sensor model output, the intermediate quantity of mechanism prediction is calculated. The intermediate quantity of mechanism prediction includes at least the carbon flow and energy flow on the biomass gas side, the carbon flow and energy flow on the coal combustion side, and the equivalent heat input of coupled combustion.

[0216] The data-driven residual correction model module is used to construct a data-driven residual correction model, and to dynamically correct the intermediate quantity of the mechanism prediction through the data-driven residual correction model to obtain the corrected intermediate quantity.

[0217] The full life cycle calculation module is used to generate dynamic emission parameters based on upstream batch data, and align the dynamic emission parameters with the corrected intermediate amount by timestamp in combination with the spatiotemporal mapping relationship to generate a full life cycle emission inventory and generate MRV data packets based on the full life cycle emission inventory.

[0218] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0219] The preset parameters or preset thresholds mentioned above are all set by those skilled in the art based on actual conditions or obtained through large-scale data simulation.

[0220] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for carbon emission accounting throughout the entire life cycle of biomass gas coupled with coal-fired power generation for carbon trading, characterized in that, Includes the following steps: The accounting boundaries and accounting cycle are determined. The accounting boundaries cover the upstream collection, transportation and pretreatment of biomass raw materials and the downstream gasification and coal-fired power generation coupled with the coal-fired power units. The accounting cycle is the preset cycle for carbon trading declaration or settlement. Collect activity data, label upstream batch data and unit-side operation data with batch identifiers and timestamps, and establish a spatiotemporal mapping relationship; Perform consistency processing and missing data detection on activity data, generate data quality grading labels and associate them with uncertainty measures; When the measured sampling frequency of syngas components or lower heating value is lower than the threshold, a soft measurement model is constructed based on the unit-side operating data and combined with the feedforward feature coupling strategy to output the estimated value and uncertainty measure. Construct a material-energy-carbon conservation mechanism model, and use computer science to predict intermediate quantities based on activity data and estimated values; A data-driven residual correction model is constructed to dynamically correct the intermediate quantity of the mechanism prediction, and the corrected intermediate quantity is obtained. Dynamic emission parameters are generated based on upstream batch data, and timestamps are aligned with the corrected intermediate values ​​according to the spatiotemporal mapping relationship to generate a full life cycle emission inventory and MRV data package. The construction of the material-energy-carbon conservation mechanism model includes the following sub-steps: S51: Establish a carbon balance equation set based on the law of conservation of mass, wherein the carbon balance equation set includes at least the carbon input of biomass feedstock, carbon from gasification conversion, carbon from bottom ash residue, carbon from fly ash residue, and carbon from flue gas emissions. S52: Calculate the sensible heat and chemical heat brought in by biomass gas based on energy conservation, and convert them into equivalent heat input of coupled combustion with the same dimensions as coal combustion; The construction of the data-driven residual correction model includes the following sub-steps: S61: When training or updating the data-driven residual correction model, a loss function with physical consistency constraints is introduced. The loss function includes a penalty term for violating mass conservation or energy conservation to constrain the intermediate quantity after correction to meet the preset physical consistency. S62: Employs a multi-fidelity fusion method, pre-trains the model using simulation data, and updates the model using on-site measured data to achieve cross-domain deviation correction; The construction of the soft measurement model based on unit-side operational data and combined with a feedforward feature coupling strategy includes the following steps: S81: Obtain raw material batch parameters and generate a priori feature vector by combining biomass degradation kinetics characteristics. The priori feature vector consists of at least two of the following: moisture content, transportation distance, and storage time. S82: The prior feature vector and the unit-side operating features are used together as input variables for the machine learning model. The unit-side operating features include at least gasifier temperature, pressure, gasifying agent flow rate and bed pressure drop. S83: Based on the machine learning model, output an estimate and uncertainty measure of at least one of the syngas components or lower heating value.

2. The method for calculating the carbon emissions of biomass gas coupled with coal-fired power generation for carbon trading as described in claim 1, characterized in that, Determining the accounting boundaries and accounting cycle includes the following sub-steps: Construct a physical boundary model that includes emission sources from non-energy activities, and incorporate at least one by-product treatment unit into the accounting boundary. The by-product treatment unit includes at least one of a biomass gasification ash treatment unit and a gas purification waste liquid treatment unit. Set a dynamic sliding window aligned with the carbon market settlement window, define the accounting period as a configurable sliding time window, and maintain a synchronized interface with the carbon trading platform's settlement or verification period.

3. The method for full life-cycle carbon emission accounting of biomass gas coupled with coal-fired power generation for carbon trading as described in claim 2, characterized in that, Establishing a spatiotemporal mapping relationship between upstream batch data and unit-side operational data includes the following sub-steps: S21: Assign a unique batch identifier to each batch of raw materials entering the factory and establish a cross-system index of batch identifier and production timestamp; S22: Record the time period during which the raw materials corresponding to the batch identifier enter the biomass gas preparation stage or the coal-fired unit coupled combustion power generation stage, forming a batch identifier-time stamp index key value.

4. The method for calculating the carbon emissions of biomass gas coupled with coal-fired power generation for carbon trading as described in claim 3, characterized in that, The process of generating data quality grading identifiers and associating them with uncertainty measures includes the following sub-steps: S31: Perform statistical analysis and logical verification on the activity data, and mark the activity data as measured data identifier, estimated data identifier, or default data identifier; S32: Generate rules for associated uncertainty measures for different data identifiers, and pass the uncertainty measures to key estimates in the life cycle emission inventory and MRV data package according to the error propagation principle.

5. The method for calculating the carbon emissions of biomass gas coupled with coal-fired power generation for carbon trading as described in claim 4, characterized in that, The output estimate and uncertainty measure include the following sub-steps: S41: Select at least two of the following variables as auxiliary variables: gasifier temperature, pressure, gasifying agent flow rate, and bed pressure drop, and construct a machine learning model to back-infer at least one of the syngas components or lower heating value. S42: Introduce an uncertainty estimation mechanism into the soft measurement model. The uncertainty estimation mechanism includes at least one of the following: an uncertainty estimation method based on random deactivation and an uncertainty estimation method based on Bayesian inference, to output a corresponding uncertainty metric.

6. The method for calculating the carbon emissions of biomass gas coupled with coal-fired power generation for carbon trading as described in claim 5, characterized in that, The process of generating a full life cycle emissions inventory and generating an MRV data package based on the full life cycle emissions inventory includes the following sub-steps: S71: Generate dynamic emission parameters based on raw material batch parameters; the raw material batch parameters include at least transportation segment data and distinguish between the first kilometer segment and the trunk line segment; the dynamic emission parameters are calculated based on at least two of the following for each batch of biomass raw materials: moisture content, transportation distance, and storage time; S72: Based on the spatiotemporal mapping relationship, the discrete emission data of the upstream batch data is mapped to the time axis of the unit-side operating data, and time-series alignment and accumulation are completed with the corrected intermediate quantity; S73: Generate an MRV data packet containing batch traceability information, activity data source and processing records, data quality classification identifier, key estimators and uncertainty measures, model version and parameter summary, and accounting trajectory log, and generate verification information, wherein the verification information includes at least a hash digest or digital signature.

7. The method for full life-cycle carbon emission accounting of biomass gas coupled with coal-fired power generation for carbon trading as described in claim 6, characterized in that, The loss function with physical consistency constraints includes: when training or updating the data-driven residual correction model, the weights of the loss function with physical consistency constraints are adaptively adjusted according to the data quality classification and uncertainty measure. When the uncertainty measure of a sample exceeds a preset uncertainty threshold, the weight of the data fitting term of the sample is reduced, and the penalty weight of violating mass conservation or energy conservation in its physical constraint term is increased accordingly.

Images