Method and system for quantifying carbon emission factor of power grid based on environmental rights equity tracing

Abstract

The present application relates to the technical field of low-carbon power, and discloses a power grid carbon emission factor quantification method and system based on environmental rights and interests traceability, which collects actual on-grid green power data and corresponding output data of the power generation side, time-of-use electricity consumption data of the consumption side, and transaction contract data between the two sides; performs environmental rights and interests traceability and cancellation on the actual on-grid green power data, time-of-use electricity consumption data and transaction contract data to obtain environment attribute traceable power; inputs the output data and transaction contract data into a pre-trained TCN-LSTM hybrid prediction model for processing, and outputs unit output, cross-regional transmission power and traceable green power proportion in a future preset time period, and calculates the power grid carbon emission factor of the target region in combination with the environment attribute traceable power; the hour-level advanced prediction of the carbon emission factor is realized, the green power environmental rights and interests are accurately traced, and the timeliness and accuracy of carbon accounting are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of low-carbon power technology, and in particular to a method and system for quantifying power grid carbon emission factors based on environmental rights tracing. Background Technology

[0002] The grid carbon emission factor is a key parameter for measuring greenhouse gas emissions per unit of electricity consumption, and it is widely used in areas such as corporate carbon accounting, product carbon footprint certification, and carbon market trading. Currently, traditional carbon emission factor calculation methods, such as the measured method, material balance method, and emission factor method, mainly use static or low-frequency updated regional averages. These methods are outdated and cannot reflect the real-time impact of green electricity consumption. Furthermore, their data collection, processing, and dissemination processes are time-consuming, leading to delays in calculation results and a disconnect from the green electricity market, preventing users from managing carbon in a timely and effective manner.

[0003] Therefore, how to solve the problem of the delay in the existing carbon emission factor calculation method and its disconnect from the green electricity market has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] This invention provides a method and system for quantifying the carbon emission factor of the power grid based on environmental rights tracing, which solves the problems of delays and disconnection from the green electricity market in existing carbon emission factor calculation methods.

[0005] To address the aforementioned technical problems, the first aspect of this invention provides a method for quantifying power grid carbon emission factors based on environmental rights tracing, comprising: Collect actual grid-connected green electricity data and corresponding output data from the power generation side within the target area, time-of-use electricity data from the consumption side, and transaction contract data between the two sides; Environmental rights traceability and verification are performed on the actual green electricity data, the time-of-use electricity data, and the transaction contract data to obtain traceable electricity with environmental attributes. The power output data and the transaction contract data are input into a pre-trained TCN-LSTM hybrid prediction model for processing, and the output of unit power output, cross-regional power transmission and traceable green electricity ratio for a future preset period are output. The grid carbon emission factor of the target region is determined based on the traceable electricity volume of the environmental attributes, the unit output, the cross-regional electricity transmission volume, and the proportion of traceable green electricity.

[0006] A second aspect of the present invention provides a power grid carbon emission factor quantification system based on environmental rights tracing, comprising: The data collection module is used to collect actual grid-connected green electricity data and its corresponding output data on the power generation side within the target area, time-of-use electricity data on the consumption side, and transaction contract data between the two sides. The rights and interests traceability module is used to trace and verify the environmental rights and interests of the actual green electricity data, the time-of-use electricity data and the transaction contract data to obtain traceable electricity with environmental attributes. The model prediction module is used to input the power output data and the transaction contract data into the pre-trained TCN-LSTM hybrid prediction model for processing, and output the unit output, cross-regional power transmission and traceable green electricity ratio for a future preset period. The factor determination module is used to determine the grid carbon emission factor of the target area based on the traceable electricity of the environmental attributes, the unit output, the cross-regional transmission electricity, and the proportion of traceable green electricity.

[0007] Compared with the prior art, the beneficial effects of the embodiments of the present invention are as follows: By collecting time-of-use electricity consumption data from the consumer side and combining it with power output data from the generation side, the calculation of carbon emission factors can be refined to the hourly level. Compared with traditional annual or monthly average carbon emission factors, this method can dynamically reflect the changes in the actual carbon intensity of the power grid at different times, which is beneficial for refined applications such as demand-side response and time-of-use carbon labeling. By introducing an environmental rights traceability and verification mechanism, it can accurately identify and remove the environmental attributes of green electricity that have been explicitly transferred by trading contracts, preventing the same green electricity from being used for carbon emission reduction accounting by both the generation and consumption sides. This results in traceable electricity with environmental attributes, ensuring that the carbon emission factor truly reflects the physical emission intensity of the power grid and avoiding underestimation of the factor due to double calculation. The introduction of the TCN-LSTM hybrid prediction model enables advanced prediction of transmission volume, power generation, and the proportion of traceable green electricity, enhancing timeliness and foresight, and improving prediction accuracy and model robustness. It achieves hourly advanced prediction of carbon emission factors, accurately traces the environmental rights of green electricity, and significantly improves the timeliness and accuracy of carbon accounting. Attached Figure Description

[0008] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a flowchart of a method for quantifying power grid carbon emission factors based on environmental rights tracing, provided in a certain embodiment of the present invention; Figure 2 This is a structural diagram of a power grid carbon emission factor quantification system based on environmental rights tracing, provided in a certain embodiment of the present invention; Figure label: Among them, 10 is the data collection module; 20 is the rights and interests traceability module; 30 is the model prediction module; and 40 is the factor determination module. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings and examples. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0011] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art will be able to understand the specific meaning of the above terms in this application according to the specific circumstances.

[0012] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is merely for describing specific embodiments and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0013] In one embodiment, such as Figure 1 As shown, the first aspect of this invention provides a method for quantifying the carbon emission factor of a power grid based on environmental rights tracing, comprising: S1. Collect actual grid-connected green electricity data and corresponding output data from the power generation side within the target area, time-of-use electricity data from the consumption side, and transaction contract data between the two sides; First, determine the target area (such as a city or a pre-defined regional range), and use the power grid of the target area as the implementation target. Identify or sort the power generation side—power plants—within the target area by quantity, or divide them according to the dispatch nodes within the power grid. Then, identify the consumption side—large industrial users, commercial complexes, and typical residential transformer substations—within the target area by quantity and type. Next, install high-precision smart terminals or unified metering devices at all grid-connected power plants (including thermal power, hydropower, wind power, photovoltaic, biomass, etc.) within the target area. Collect actual grid-connected green electricity data (only for the actual grid-connected electricity of renewable energy power plants such as wind power and photovoltaic) and output data (actual active power of all units, load sequence, cross-regional power transmission power sequence, etc.) every 5 minutes. The data is encrypted and uploaded to the power dispatch center's data platform via the dedicated power network. The platform performs outlier removal, missing value imputation (using linear interpolation or nearby time average filling), and timestamp alignment on the raw data. Install user-side smart meters or unified metering on the consumption side. The device collects time-of-use electricity consumption data every 15 minutes, and the data is uploaded to the same data platform and undergoes the same preprocessing. It retrieves all green electricity trading contract data (including intra-provincial and inter-provincial / inter-regional green electricity transactions) between the power generation and consumption sides from the power trading center. Each contract contains at least the following fields: contract number, seller (power generation company), buyer (electricity user or electricity sales company), green electricity transaction volume (MWh), transaction start and end time, unique identifier of the corresponding power generation project (such as project code), and environmental rights attribution clauses. The contract data is parsed into a structured table and indexed according to timestamps and power generation company and user information. Simultaneously, it also collects relevant time data such as time, weekday / holiday, and season, as well as meteorological data such as wind speed, solar irradiance, and temperature, and performs similar preprocessing such as cleaning on this data.

[0014] Additionally, it should be noted that this invention achieves seamless data interoperability across four major platforms through standardized API interfaces: the Green Certificate Issuance and Trading System, the Power Grid Energy Management System (EMS), the Power Trading Institution System, and the Enterprise Energy Management System (EMS).

[0015] S2. Perform environmental rights tracing and verification on the actual green electricity data, the time-of-use electricity data, and the transaction contract data to obtain traceable electricity with environmental attributes. In one embodiment, step S2 includes: The actual green electricity data, the time-of-use electricity data, and the transaction contract data are matched spatiotemporally based on time period, region, and quantity. Based on the matching results, traceable green electricity data is determined, and traceable green electricity consumption certificates corresponding to the traceable green electricity data are generated. In response to the verification application from the consumer side, the traceable green electricity data and its corresponding traceable green electricity consumption certificate are verified to obtain the traceable electricity volume with environmental attributes.

[0016] Specifically, this invention defines a "matching unit" as a triple: (t, j, transaction pair), where t is time, preferably a 15-minute time period; j is the region; and the transaction pair refers to the generation side and user side corresponding to a contract. This matching unit compares the actual grid-connected green electricity data, time-of-use electricity data, and transaction contract data based on the time period (less than 15 minutes can be converted using a sliding average, and more than 15 minutes can be obtained by splitting), region (ensuring the grid connection region of the generation side, the electricity consumption region of the consumer side, and the settlement region stipulated in the contract are consistent, avoiding confusion regarding the cross-regional affiliation of green electricity), and quantity (the transaction pair matches the content of the transaction contract data; for example, extracting records of successful spatiotemporal matching, obtaining the electricity data for each time period of the three, and verifying whether the relative error of the three is within ±1%). If the electricity volume of the generation side is less than the contractually agreed... If the electricity consumption is 10% higher than the contractually agreed amount, exceeding the error tolerance limit, the amount will be calculated based on the contractually agreed amount, meaning the excess portion is considered over-generation and cannot be attributed to the user. If the user's electricity consumption is 10% lower than the contractually agreed amount, exceeding the error tolerance limit, the amount will be calculated based on the user's actual electricity consumption, meaning the unused portion cannot be reimbursed. Simultaneously, if the error exceeds the preset tolerance limit, it will be marked as an abnormal quantity. For example, if the power generation side's electricity consumption is 0, it indicates the unit is out of service; if the user's electricity consumption is 0, it indicates the user has not used electricity. Both of these situations are directly judged as untraceable. The data must be consistent across all three factors. Subsequently, the minimum value among the actual grid-connected green electricity data, time-of-use electricity data, and transaction contract data within the matching result will be taken as the traceable green electricity data. In other words, the traceable green electricity data is the minimum value among the matched user's electricity consumption, the contracted electricity consumption between the user and the green electricity generation company, and the electricity available for consumption by the green electricity generation company. For each traceable green electricity data value generated by a matching unit, an "Accountable Green Electricity Consumption Certificate" is automatically generated. The certificate includes at least the following: certificate number (unique hash value), power generation company name and project code, electricity user name and account number, time period (start and end time), region, traceable green electricity data, corresponding transaction contract number, generation timestamp and electronic signature. The certificate is stored in the blockchain or the centralized ledger of the power trading center, and consumers can query and download it at any time. The initial status of the certificate is "not verified".

[0017] Consumers, based on their own carbon reduction needs (such as fulfilling annual carbon market obligations), can select one or more "unverified" traceable green electricity consumption certificates and initiate a verification application. The purpose of the verification must be specified in the application. In response to the consumer's verification application, this invention first verifies that the traceable green electricity data in the certificate has not yet been verified (to prevent duplicate verification). Then, it changes the traceable green electricity data from "unverified" to "verified," recording the verification time, applicant, purpose, and verification serial number. Simultaneously, the traceable green electricity data M is deducted from the available balance of the matching unit. Subsequent verifications of the remaining electricity in the same matching unit will only be performed on (M - verified electricity volume) until the balance is zero. This environmental rights verification process can promptly and permanently mark the corresponding environmental rights as "used" and attribute them to the user, preventing duplicate calculations and sales of environmental rights.

[0018] After the write-off process is completed, the "environmentally traceable electricity" is defined as the sum of the written-off amounts (M) among all generated traceable green electricity consumption certificates. In other words, only the portion of traceable green electricity that has been actively applied for and written off by the consumer is considered to have its environmental attributes removed from the grid and attributed to the specific user. The unwritten portion remains in the grid's public pool and participates in the calculation of the average carbon emission factor denominator. Furthermore, all matching results, consumption certificate generation records, write-off applications, and operation logs are written to the consortium blockchain in real time (with the grid company, trading center, and regulatory authorities serving as consensus nodes). Any third party can query the complete chain using the certificate number: generation time data → contract breakdown amount → user time-of-use electricity → calculation process of matching and taking the minimum value → write-off time and purpose. Since the data cannot be altered once written, the risk of subsequent tampering is eliminated.

[0019] This invention uses a time-region-quantity-based spatiotemporal matching system to link each green electricity consumption to a specific power generation time period and geographical location, providing a complete chain of evidence for the subsequent generation of a "traceable green electricity consumption certificate." By taking the minimum value of actual grid-connected green electricity data, time-of-use electricity data, and transaction contract data, it ensures that the final verified traceable green electricity volume will not exceed the constraints of any one source. This ensures that any missing or abnormal data from any party will automatically reduce the matching result, thus preventing violations such as false reporting of green electricity by the power generation side and excessive declarations of consumption by users, and safeguarding the credibility of carbon emission reduction accounting.

[0020] S3. Input the power output data and the transaction contract data into the pre-trained TCN-LSTM hybrid prediction model for processing, and output the unit output, cross-regional power transmission, and traceable green electricity ratio for the future preset time period; wherein, the TCN-LSTM hybrid prediction model includes a TCN layer, an LSTM layer, and a fully connected layer; due to the delay in the power system data acquisition, processing, and release process, the carbon emission factor based on measured data cannot meet the user's needs for real-time accounting and management of current and future carbon emissions. Therefore, a temporal convolutional network-long short-term memory network (TCN-LSTM hybrid prediction) model is introduced. Before the release of the time-of-use emission factor after deducting the time-of-use emission factor revised by the traceable green electricity, the power system operation status and key environmental rights variables at the unreleased time are predicted, thereby determining the emission factor to meet the needs of enterprises for real-time carbon emission accounting.

[0021] In one embodiment, step S3 includes: Multidimensional basic features are constructed based on the output data and the transaction contract data; The long-term temporal features in the multidimensional basic features are extracted through the TCN layer to obtain a high-order spatiotemporal feature sequence; The LSTM layer is used to capture the dynamic patterns in the high-order spatiotemporal feature sequence to obtain a temporal dynamic feature vector. Based on the fully connected layer, the time-series dynamic feature vector is mapped to the unit output, cross-regional power transmission, and traceable green electricity ratio for the future preset time period.

[0022] Specifically, the TCN-LSTM hybrid prediction model adopts a cascaded hybrid architecture of TCN and LSTM: the TCN layer acts as a front-end feature extractor, using its dilated convolutions and residual connections to capture long-term dependencies and complex feature patterns in the input multi-dimensional time series data; the LSTM layer inherits the high-order feature sequences extracted by the TCN layer, using its gating mechanism to model the time series dynamics, focusing on capturing complex time patterns such as the change of green electricity ratio with weather, load and trading plans; the fully connected layer maps the final output of the LSTM layer to the prediction target.

[0023] To ensure the quality and integrity of the model input data, this invention adopts a data preprocessing process based on Generative Adversarial Network (GAN). This process can not only perform conventional normalization processing, but also intelligently repair missing values ​​and outliers in historical time series data. Furthermore, it can synthesize high-quality data to balance the dataset when training samples are scarce, thereby fundamentally improving the robustness and accuracy of subsequent prediction models.

[0024] In one embodiment, before constructing the multidimensional basic features based on the output data and the transaction contract data, the following steps are included: Anomaly detection is performed on the output data and the transaction contract data to obtain anomaly detection results; The anomaly detection results are input into a pre-trained GAN model for repair and verification, and the verification results are output. The verification results are normalized, and the normalized results are used as the time-series input dataset.

[0025] For input output data, transaction contract data, time-related data, and meteorological data, this invention performs the following steps: First, obvious outliers and missing segments in the input data are initially identified using statistical methods or rule bases (or univariate anomaly detection methods, distance-based clustering, or isolated forest algorithms), and these are marked to obtain anomaly detection results containing marked data and non-anomaly input data. Second, the marked data (including missing / anomalies) from the anomaly detection results and relevant features of the corresponding time period (such as meteorological conditions and load levels) are input into the generator in a pre-trained GAN network. The generator generates reasonable values ​​to replace outliers or fill in missing values ​​based on the contextual association and historical patterns of the data. The discriminator in the GAN network evaluates these generated data to ensure consistency with the real data distribution and prevent the introduction of unreasonable biases. The evaluated generated data and non-anomaly input data are output as verification results. The GAN network constructed in this invention, containing a generator G and a discriminator D, is used for data repair and enhancement. The generator G learns the real data distribution and can generate synthetic data that conforms to physical and statistical laws; the discriminator D is responsible for distinguishing between real data and generated data. Both are optimized together through adversarial training. The objective function adopts the standard objective function of Generative Adversarial Network (GAN) in existing technology, which is often referred to as the objective function of Minimax Game.

[0026] The validation results were then Z-score standardized to eliminate the influence of dimensions and accelerate model convergence. The normalized results, which included normalized meteorological data, normalized periodic data, normalized power output data, and normalized transaction contract data, were then input into the model as a time-series input dataset for processing.

[0027] This invention locates problematic data through an anomaly detection module and then uses a generative adversarial network (GAN) to repair the abnormal parts, so that the time series input dataset entering the TCN-LSTM prediction model has high integrity and high reliability, which greatly improves the quality of the original data, ensures the reliability of subsequent predictions, and thus improves the accuracy of carbon emission factors.

[0028] In one embodiment, the time-series input dataset further includes normalized meteorological data and normalized periodic data; wherein, The construction of multi-dimensional basic features based on the output data and the transaction contract data includes: Electricity features and green electricity rights features are constructed based on the normalized output data and normalized transaction contract data in the time-series input dataset, respectively; and meteorological features and time features are constructed based on the normalized meteorological data and the normalized periodic data, respectively. The electricity characteristics, green electricity rights characteristics, meteorological characteristics, and time characteristics are combined to form the multidimensional basic characteristics.

[0029] Specifically, the model extracts the cross-regional power transmission power sequence, the output sequence of various types of generating units, and the regional total load sequence from the normalized output data as power features. These features reflect the inertia and patterns of power interaction between regions, the overall operating status and dispatch strategy of the power grid, and the cross-regional traceable green electricity ratio. The model uses the signed green electricity transaction contract volume and traceable green electricity volume from the normalized transaction contract data as green electricity rights features, enabling the model to infer the planned green electricity base volume during the forecast period. The model uses wind speed, light intensity, and temperature from the normalized meteorological data as meteorological features, which directly affect the output of wind and solar power in the region. The model uses hours, weekdays / holidays, and seasons from the normalized periodic data as time features, which are used to capture the periodic operating patterns of power supply and demand.

[0030] Finally, power characteristics, green electricity rights characteristics, meteorological characteristics, and temporal characteristics are combined to form multi-dimensional basic features. This invention introduces normalized meteorological data and explicitly constructs meteorological features, enabling the TCN-LSTM model to directly learn the nonlinear mapping relationship between meteorological elements and wind and solar power output, thus more accurately predicting unit output in future periods. By introducing normalized periodic data, the model can automatically learn the intraday bi-peak characteristics of power load and output, the differences between working and rest days, and seasonal changes in wind and solar resources. Features are explicitly divided into four categories: power characteristics (reflecting physical power generation capacity), green electricity rights characteristics (reflecting market transactions and environmental rights attribution), meteorological characteristics (reflecting natural resource endowment), and temporal characteristics (reflecting human activities and natural cycles). This structured feature combination method not only facilitates the adjustment of weights from different data sources in engineering but also enhances the interpretability of the model's prediction results.

[0031] After constructing the multidimensional basic features, they are input into the TCN layer of the TCN-LSTM hybrid prediction model. The TCN layer is the front-end feature extractor, consisting of multiple dilated convolutional layers and residual connections (preferably 4 residual blocks, each containing two dilated convolutional layers. The dilation coefficients are 1, 2, 4, and 8, the kernel size is set to 3, the stride is 1, and causal convolution is used to ensure that the output at time t depends only on information before t. Each dilated convolution is followed by batch normalization and ReLU activation function, and dropout is added with a ratio of 0.2). The convolution operation of the dilated convolution can be represented as: In the formula, x The input sequence is the multidimensional basic feature. y The output sequence represents the output of the dilated convolution at time step t; These are the weights of the convolution kernel; K The kernel size; d Expansion rate; k This is the summation variable, used to iterate through all weight positions of the convolution kernel to complete the weighted summation.

[0032] By utilizing its dilated causal convolutional structure, the features at each time step are mapped to a higher-dimensional abstract space, effectively capturing long-term dependencies and multi-dimensional features in the input time-series data. Subsequently, the output of the dilated convolutional layer is residually connected to the input sequence to obtain a high-order spatiotemporal feature sequence. This sequence retains the entire time step of the original input, but the features at each time step have incorporated historical information spanning several days. This invention, through residual connections, prevents gradient vanishing, ensuring the model's training stability and its ability to retain historical information.

[0033] The LSTM layer consists of an input gate, a forget gate, an output gate, and a cell state (preferably a double-layer stacked LSTM is used, with 128 hidden units in the first layer and 64 hidden units in the second layer, and Dropout set to 0.3 for both layers). The forget gate determines which old information in the cell state needs to be retained and which needs to be forgotten; the input gate determines the proportion of new information stored and generates candidate new information; the cell state is used to store long-term dependency information of the sequence; and the output gate is used to filter key information in the cell state, generating the hidden state at the current moment, which will also serve as one of the inputs for the next moment. The LSTM layer captures the dynamic patterns in high-order spatiotemporal feature sequences, thus obtaining a temporal dynamic feature vector. This vector condenses all the dynamic patterns extracted from the input data, such as: the trend of thermal power output changes over the next 24 hours (based on past multi-day peak-valley patterns); the probability of blockage in transmission channels under similar weather and contract conditions and its impact on cross-regional power generation; the daily variation pattern of green electricity ratio over the past week; and the offset affected by contract execution progress. The specific processing procedure is similar to the data processing procedure of existing LSTM models and will not be elaborated here.

[0034] The fully connected layer is a 3-layer fully connected network (multilayer perceptron). The input is the temporal dynamic feature vector output by the LSTM. The intermediate hidden layers have dimensions of 128 and 64 respectively, with ReLU activation. The output dimension of the last layer equals the number of predicted targets. The predicted targets are multiple variables for a future preset time period (e.g., 24 hours, with 96 15-minute intervals): predicted output of various types of generators (hourly power generation of thermal power units, wind power units, photovoltaic units, hydropower units, etc.); cross-regional power transmission (power transmitted from other regions to the target region); and the proportion of traceable green electricity (a value between 0 and 1) of the power transmitted between the target region and other regions (i.e., regions that transmit power across regions to the target region). The output values ​​are linearly transformed (without activation function) to obtain the final prediction result.

[0035] In addition, for model training, real-time power output data, transaction contract data, and corresponding meteorological data from the past three years were collected. Historical data was used as input in 72-hour windows, and the actual power output of each unit, cross-regional power transmission, and the proportion of traceable green electricity in each region for the next 24 hours were used as labels. The training, validation, and test sets were divided in an 8:1:1 ratio. The Adam optimizer was used with an initial learning rate of 0.001, decaying to 0.9 every 10 epochs. The batch size was set to 64, the maximum number of training epochs was 100, and an early stopping mechanism was implemented (the model stopped if the validation set loss did not decrease for 10 consecutive epochs). The mean squared error (MSE) was used as the loss function, and supervised learning was performed through backpropagation until the model converged. The trained model was deployed to a real-time prediction server. Every 15 minutes, the latest power output data, transaction contract data, and meteorological forecast data from the past 72 hours were automatically retrieved, standardized, and then input into the model. The model output the prediction results for the next 24 hours within seconds, which were used for subsequent dynamic calculation of the grid carbon emission factor.

[0036] This invention utilizes the dilated convolutional structure of the TCN layer to significantly expand the receptive field without increasing the number of layers, effectively extracting long-term temporal features spanning multiple days or even weeks. Simultaneously, the LSTM layer, through its gating mechanism, excels at capturing short-term dynamic changes within sequences. The combination of these two elements enables the model to accurately model patterns across different time scales in the power grid, resulting in more stable and realistic predictions. Furthermore, by incorporating transaction contract data into the multidimensional basic features, the model can anticipate the impact of market behavior on unit output and the proportion of green electricity, significantly improving the business interpretability and numerical accuracy of the prediction results.

[0037] S4. Determine the grid carbon emission factor of the target region based on the traceable electricity of the environmental attributes, the unit output, the cross-regional transmission electricity, and the traceable green electricity ratio. In one embodiment, step S4 includes: The original total carbon emissions corresponding to all the electricity consumed in the target area are determined based on the unit output and the cross-regional power transmission. Based on the traceable electricity of the environmental attributes, the cross-regional transmission electricity of the unit output, and the proportion of traceable green electricity, the remaining electricity obtained after deducting all the green electricity that has been cancelled from the total electricity consumed in the target area is determined; The grid carbon emission factor is determined by the original total carbon emissions and the remaining electricity.

[0038] Specifically, the traditional power grid carbon emission factor = total carbon emissions / total electricity consumption. The zero-carbon benefits of green electricity are diluted across all users, resulting in identical carbon emission factors for users consuming and not consuming green electricity, failing to reflect the emission reduction value of green electricity. This invention does not distinguish between green and non-green electricity. It calculates the original carbon emissions (numerator) corresponding to all electricity consumed in the target area (local generation + imported electricity). From the total electricity consumed in the target area, it deducts all traceable green electricity that has been verified and clearly attributed (the emission reduction benefits of this green electricity have been locked by users and are no longer included in subsequent factor averaging). The total carbon emissions are divided by the non-green electricity after deducting green electricity. The final factor only reflects the carbon emission intensity of electricity that has not been verified as green electricity, achieving precise attribution of the emission reduction benefits of green electricity.

[0039] Calculate the product of the power generation of each unit in the target area and the carbon emission factor of each unit, and then sum them to obtain the total carbon emissions from local power generation. ,in, The carbon emission factor of generator unit i within the target area's power grid coverage is taken as the corresponding carbon emission factor of the generator unit. To predict the power generation of generator unit i within the target area's power grid coverage area, calculate the product of the electricity transmitted from each external region to the target region and the power grid carbon emission factor of that region (which can be the average of the power grid carbon emission factors of other regions over a preset period; or calculated using the material balance method), and sum them to obtain the total carbon emissions from external electricity. ,in, For the power grid carbon emission factor corresponding to other regions j, the average value of the carbon emission factor of that region over a preset period in the past can be taken. For the predicted area The total amount of electricity delivered to the target area is calculated; local refers to the target area; the total carbon emissions of local power generation are added to the total carbon emissions of imported electricity to obtain the original total carbon emissions corresponding to all the electricity consumed in the target area (local power generation + imported electricity), without distinguishing between green electricity and non-green electricity.

[0040] The total local power generation of the target area And the predicted total amount of electricity transmitted across regions, which is the total amount of electricity imported. Add them together to get the total electricity consumption base of the target area; multiply the total local power generation of the target area by the local traceable green electricity ratio to get the local green electricity volume that has been written off. ,in, To predict the traceable green electricity ratio corresponding to the target area, multiply the electricity transmitted from each external area to the target area by the traceable green electricity ratio of that area to obtain the traceable green electricity amount after deducting the amount of external electricity that has already been verified. ; Let J be the traceable green electricity ratio for region j. Add the locally verified green electricity to the traceable green electricity that has been verified after deducting external electricity to obtain the total verified green electricity. Calculate the deviation between the total verified green electricity and the traceable green electricity based on environmental attributes. If it is less than a preset threshold (e.g., 1%), use the total verified green electricity as the final verified green electricity; otherwise, take the minimum of the two as the final verified green electricity. Finally, subtract the final verified green electricity from the total electricity consumed by the target region to obtain the remaining electricity after deducting all verified green electricity from the total electricity consumed by the target region. This remaining electricity can no longer enjoy the green electricity emission reduction benefits and is the main bearer of carbon emissions.

[0041] Finally, dividing the original total carbon emissions by the remaining electricity yields the grid carbon emission factor for the target area. This invention explicitly deducts all physical electricity corresponding to previously written-off green electricity (i.e., the written-off portion of environmentally traceable electricity) from the denominator, ensuring that the carbon emission factor reflects only the average carbon intensity of the remaining electricity. This prevents separately written-off green electricity from further reducing the grid average factor, avoiding double counting of environmental rights on both the supply and consumption sides, and aligns with the international principle of preventing double counting in green electricity trading and carbon markets.

[0042] In one embodiment, after step S4, the following is included: Obtain the actual carbon emission factor for the corresponding time period of the power grid carbon emission factor, and calculate the deviation between the power grid carbon emission factor and the actual carbon emission factor; When the deviation exceeds a preset threshold, supplementary training of the TCN-LSTM hybrid prediction model is triggered.

[0043] Specifically, due to the delayed release of the grid carbon emission factor, after the official release of the actual carbon emission factor for the corresponding period, the average absolute percentage error between the two is calculated as the deviation. If the calculated deviation is greater than 5% and lasts for more than 2 hours in a single instance, a log is triggered and maintenance personnel are notified. If the calculated deviation is greater than 8% and exceeds 8% for three consecutive evaluation windows (i.e., three consecutive hours), or if a single deviation is greater than 12%, supplementary training is immediately triggered: all data since the last complete training (or at least the most recent two weeks) is extracted from the historical database, including real-time output, transaction contracts, meteorological data, cyclical data, and actual write-off results, etc.; the past 72 hours are used as the input window, and the actual factor and its related intermediate variables (machine) corresponding to the next 24 hours are used as the input. The model uses labels (group output, cross-regional electricity consumption, and traceable green electricity ratio) to form several training samples. More recent samples are assigned higher weights (e.g., by a time decay function) to make the model more focused on current data patterns. Instead of re-initializing the model randomly, the weights of the current TCN-LSTM model are loaded as initial values, and a low initial learning rate (e.g., 1 / 10 of the original learning rate) is used for a small number of iterations (e.g., 5-10 rounds). The first few residual blocks of the TCN layer can be frozen initially, training only the LSTM layer and the last fully connected layer to accelerate adaptation. If the bias persists, all layers are unfrozen for complete fine-tuning. Errors are monitored on the validation set (the latest 20% of the training data). Supplementary training is stopped when the validation set error no longer decreases for three consecutive rounds to prevent overfitting. The required data is automatically retrieved from the data lake, and the training module (in the same environment as the initial training) is invoked. The training process runs in the background without interrupting ongoing real-time predictions (predicting continues using the old model until the new model is validated). After training, the performance of the new and old models is compared on the historical test set: the MAPE of the new model should be less than that of the old model and below the trigger threshold; otherwise, the old model is retained and a failure log is recorded. If the new model passes the validation, the TCN-LSTM model in the current production environment is automatically replaced, and the new model is used for subsequent predictions.

[0044] This invention periodically compares predicted values ​​(grid carbon emission factors) with actual values ​​(post-calculated true factors), automatically triggering supplementary training when the deviation exceeds a threshold. This ensures the model always closely reflects the latest data distribution, avoiding the lag and subjectivity of manual periodic retraining. It guarantees the long-term stability of carbon emission factor prediction accuracy, achieving adaptive continuous optimization of the model and improving long-term prediction accuracy. Furthermore, it does not retrain after every prediction, but only triggers supplementary training when performance degradation exceeds the tolerance range. This ensures model quality while avoiding the waste of computing power caused by frequent training.

[0045] This application addresses the issues of delays and disconnect between existing carbon emission factor calculation methods and the green electricity market by designing a grid carbon emission factor quantification method based on environmental rights traceability. This method collects time-of-use electricity consumption data from the consumer side and combines it with power generation data from the generator side, enabling the calculation of carbon emission factors to be refined to the hourly level. Compared to traditional annual or monthly average carbon emission factors, this method can dynamically reflect the actual carbon intensity changes of the grid at different times, which is beneficial for refined applications such as demand-side response and time-of-use carbon labeling. By introducing an environmental rights traceability and verification mechanism, it can accurately identify and remove transactions that have already been traded. By clearly defining the environmental attributes of transferred green electricity, the same green electricity is prevented from being used simultaneously by the generation and consumption sides for carbon emission reduction accounting. This results in traceable electricity with environmental attributes, ensuring that carbon emission factors truly reflect the physical emission intensity of the power grid and avoiding underestimation of factors due to dual calculations. The introduction of the TCN-LSTM hybrid prediction model enables advanced prediction of transmission volume, generation volume, and the proportion of traceable green electricity, enhancing timeliness and foresight, as well as improving prediction accuracy and model robustness. Hourly-level advanced prediction of carbon emission factors is achieved, accurately tracing the environmental rights and interests of green electricity and significantly improving the timeliness and accuracy of carbon accounting.

[0046] It should be noted that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order requirement for the execution of these steps, and they can be executed in other orders.

[0047] In another embodiment, such as Figure 2 As shown, a second aspect of the present invention provides a power grid carbon emission factor quantification system based on environmental rights tracing, comprising: The data collection module 10 is used to collect actual grid-connected green electricity data and its corresponding output data on the power generation side within the target area, time-of-use electricity data on the consumption side, and transaction contract data between the two sides. The rights and interests traceability module 20 is used to trace and verify the environmental rights and interests of the actual green electricity data, the time-of-use electricity data and the transaction contract data to obtain traceable electricity with environmental attributes. Model prediction module 30 is used to input the power output data and the transaction contract data into a pre-trained TCN-LSTM hybrid prediction model for processing, and output the unit output, cross-regional power transmission and traceable green electricity ratio for a future preset period. The factor determination module 40 is used to determine the grid carbon emission factor of the target area based on the traceable electricity of the environmental attributes, the unit output, the cross-regional transmission electricity and the traceable green electricity ratio.

[0048] It should be noted that each module in the aforementioned grid carbon emission factor quantification system based on environmental rights traceability can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module. For specific limitations regarding the grid carbon emission factor quantification system based on environmental rights traceability, please refer to the limitations of the grid carbon emission factor quantification method based on environmental rights traceability mentioned above; both have the same function and role, and will not be repeated here.

[0049] In summary, this invention relates to the field of low-carbon power technology and discloses a method and system for quantifying grid carbon emission factors based on environmental rights traceability. It collects actual grid-connected green electricity data and its corresponding output data from the power generation side within a target area, time-of-use electricity data from the consumption side, and transaction contract data between the two sides. Environmental rights traceability and verification are performed on the actual grid-connected green electricity data, time-of-use electricity data, and transaction contract data to obtain traceable electricity volume with environmental attributes. The output data and transaction contract data are input into a pre-trained TCN-LSTM hybrid prediction model for processing, outputting the unit output, inter-regional transmission electricity volume, and traceable green electricity ratio for a future preset time period. Combined with the traceable electricity volume with environmental attributes, the grid carbon emission factor for the target area is calculated. This achieves hourly-level advance prediction of carbon emission factors, accurately traces the environmental rights of green electricity, and significantly improves the timeliness and accuracy of carbon accounting.

[0050] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0051] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this invention, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. A method for quantifying power grid carbon emission factors based on environmental rights tracing, characterized in that, include: Collect actual grid-connected green electricity data and corresponding output data from the power generation side within the target area, time-of-use electricity data from the consumption side, and transaction contract data between the two sides; Environmental rights traceability and verification are performed on the actual green electricity data, the time-of-use electricity data, and the transaction contract data to obtain traceable electricity with environmental attributes. The power output data and the transaction contract data are input into a pre-trained TCN-LSTM hybrid prediction model for processing, and the output of unit power output, cross-regional power transmission and traceable green electricity ratio for a future preset period are output. The grid carbon emission factor of the target region is determined based on the traceable electricity volume of the environmental attributes, the unit output, the cross-regional electricity transmission volume, and the proportion of traceable green electricity.

2. The method for quantifying power grid carbon emission factors based on environmental rights tracing according to claim 1, characterized in that, The process of tracing and verifying environmental rights based on the actual green electricity data, the time-of-use electricity data, and the transaction contract data to obtain traceable electricity with environmental attributes includes: The actual green electricity data, the time-of-use electricity data, and the transaction contract data are matched spatiotemporally based on time period, region, and quantity. Based on the matching results, traceable green electricity data is determined, and traceable green electricity consumption certificates corresponding to the traceable green electricity data are generated. In response to the verification application from the consumer side, the traceable green electricity data and its corresponding traceable green electricity consumption certificate are verified to obtain the traceable electricity volume with environmental attributes.

3. The method for quantifying power grid carbon emission factors based on environmental rights tracing according to claim 2, characterized in that, The determination of traceable green electricity data based on matching results includes: The minimum value among the actual on-grid green electricity data, time-of-use electricity data, and transaction contract data in the matching results is taken as the traceable green electricity data.

4. The method for quantifying power grid carbon emission factors based on environmental rights tracing according to claim 1, characterized in that, The TCN-LSTM hybrid prediction model includes a TCN layer, an LSTM layer, and a fully connected layer.

5. The method for quantifying power grid carbon emission factors based on environmental rights tracing according to claim 4, characterized in that, The process of inputting the power output data and the transaction contract data into a pre-trained TCN-LSTM hybrid prediction model for processing, and outputting the unit output, cross-regional power transmission volume, and traceable green electricity ratio for a future preset time period, includes: Multidimensional basic features are constructed based on the output data and the transaction contract data; The long-term temporal features in the multidimensional basic features are extracted through the TCN layer to obtain a high-order spatiotemporal feature sequence; The LSTM layer is used to capture the dynamic patterns in the high-order spatiotemporal feature sequence to obtain a temporal dynamic feature vector. Based on the fully connected layer, the time-series dynamic feature vector is mapped to the unit output, cross-regional power transmission, and traceable green electricity ratio for the future preset time period.

6. The method for quantifying power grid carbon emission factors based on environmental rights tracing according to claim 5, characterized in that, Before constructing the multidimensional basic features based on the output data and the transaction contract data, the following steps are included: Anomaly detection is performed on the output data and the transaction contract data to obtain anomaly detection results; The anomaly detection results are input into a pre-trained GAN model for repair and verification, and the verification results are output. The verification results are normalized, and the normalized results are used as the time-series input dataset.

7. The method for quantifying power grid carbon emission factors based on environmental rights tracing according to claim 6, characterized in that, The time-series input dataset also includes normalized meteorological data and normalized periodic data; wherein... The construction of multi-dimensional basic features based on the output data and the transaction contract data includes: Electricity features and green electricity rights features are constructed based on the normalized output data and normalized transaction contract data in the time-series input dataset, respectively; and meteorological features and time features are constructed based on the normalized meteorological data and the normalized periodic data, respectively. The electricity characteristics, green electricity rights characteristics, meteorological characteristics, and time characteristics are combined to form the multidimensional basic characteristics.

8. The method for quantifying power grid carbon emission factors based on environmental rights tracing according to claim 1, characterized in that, The determination of the grid carbon emission factor for the target region based on the traceable electricity generation of the environmental attributes, the unit output, the inter-regional power transmission, and the proportion of traceable green electricity includes: The original total carbon emissions corresponding to all the electricity consumed in the target area are determined based on the unit output and the cross-regional power transmission. Based on the traceable electricity of the environmental attributes, the cross-regional transmission electricity of the unit output, and the proportion of traceable green electricity, the remaining electricity obtained after deducting all the green electricity that has been cancelled from the total electricity consumed in the target area is determined; The grid carbon emission factor is determined by the original total carbon emissions and the remaining electricity.

9. The method for quantifying power grid carbon emission factors based on environmental rights tracing according to claim 1, characterized in that, After determining the grid carbon emission factor of the target region based on the traceable electricity volume of the environmental attributes, the unit output, the inter-regional transmission electricity volume, and the proportion of traceable green electricity, the process includes: Obtain the actual carbon emission factor for the corresponding time period of the power grid carbon emission factor, and calculate the deviation between the power grid carbon emission factor and the actual carbon emission factor; When the deviation exceeds a preset threshold, supplementary training of the TCN-LSTM hybrid prediction model is triggered.

10. A power grid carbon emission factor quantification system based on environmental rights tracing, characterized in that, include: The data collection module is used to collect actual grid-connected green electricity data and its corresponding output data on the power generation side within the target area, time-of-use electricity data on the consumption side, and transaction contract data between the two sides. The rights and interests traceability module is used to trace and verify the environmental rights and interests of the actual green electricity data, the time-of-use electricity data and the transaction contract data to obtain traceable electricity with environmental attributes. The model prediction module is used to input the power output data and the transaction contract data into the pre-trained TCN-LSTM hybrid prediction model for processing, and output the unit output, cross-regional power transmission and traceable green electricity ratio for a future preset period. The factor determination module is used to determine the grid carbon emission factor of the target area based on the traceable electricity of the environmental attributes, the unit output, the cross-regional transmission electricity, and the proportion of traceable green electricity.

Images