A method, system, device and medium for constructing a data-driven estimation model of carbon emissions of a coal-fired unit

Abstract

This invention discloses a method, system, equipment, and medium for constructing a carbon emission data-driven calculation model for coal-fired power units. The method includes: acquiring historical operating parameter data of the coal-fired power unit; preprocessing the historical operating parameter data to obtain first historical operating parameter data; performing correlation analysis on the first historical operating parameter data to extract key operating features; constructing input samples from the key operating features; using the input samples as input to construct a temporal convolutional network model using the first historical operating parameter data; performing global search and adaptive optimization on the temporal convolutional network model using an optimization algorithm to obtain an optimized temporal convolutional network model; predicting coal consumption for power supply using the optimized temporal convolutional network model; and calculating the carbon factor based on the predicted coal consumption for power supply. This method enables the construction of the calculation model to be driven by carbon emission data of the coal-fired power unit, enhancing the model's global optimization capability and generalization performance.

Description

Technical Field

[0001] This invention relates to the field of carbon emission accounting technology, and in particular to a method, system, equipment and medium for constructing a data-driven calculation model for carbon emissions from coal-fired power units. Background Technology

[0002] Currently, the calculation of the electricity carbon factor has become a core foundation for supporting green grid dispatch, formulating differentiated emission reduction strategies, and promoting carbon market trading. The electricity carbon factor characterizes the carbon emission intensity corresponding to a unit of electricity and is a key parameter connecting the generation and consumption ends in carbon footprint accounting. In the grid dispatch system, the generation side includes various types of power sources. Among them, renewable clean energy sources such as hydropower, wind power, and photovoltaics do not directly generate carbon emissions during power generation; their carbon footprint mainly originates from indirect emissions during the equipment manufacturing stage and has already been included in the upstream industry carbon accounting system. To avoid double counting between industries, clean energy is usually considered a zero-carbon emission power source in the calculation of the electricity carbon factor. In contrast, coal-fired power remains the main source of current grid carbon emissions. The carbon emission intensity of coal-fired power is not only much higher than other power source types but also exhibits significant dynamic and nonlinear characteristics due to factors such as coal quality, load changes, and operating conditions. Therefore, establishing a high-frequency, real-time, and accurate carbon emission estimation model for coal-fired power units is an important way to improve the accuracy of the electricity carbon factor and achieve carbon emission transparency.

[0003] Traditional carbon emission calculation methods include the positive balance method, the inverse balance method, and the empirical function method. The positive balance method, based on the principle of material balance, calculates carbon emissions by the ratio of coal consumption to power generation. It is characterized by its simplicity and wide applicability, but it relies on periodic statistical data and suffers from poor real-time performance. The inverse balance method, starting from the thermodynamic energy transfer path, indirectly estimates emissions based on the conversion efficiency of the boiler-turbine-generator unit. While theoretically precise, it is difficult to obtain parameters and is sensitive to actual operating conditions. The empirical function method uses long-term statistical fitting of the relationship between power generation load and coal consumption for online estimation. Its advantages include simple modeling and a certain degree of real-time performance, but it suffers from fixed model structure, limited consideration of factors, and poor generalization ability, failing to fully reflect the differences in carbon emissions caused by changes in equipment status. Furthermore, limited by the static modeling paradigm and manual parameter configuration mechanisms of traditional methods, these methods struggle to adapt to the highly dynamic, highly nonlinear, and multi-disturbance operating characteristics of modern power systems, failing to meet the practical needs of real-time carbon emission monitoring and refined management. With the increasing digitalization of power generation enterprises, online monitoring systems have been widely deployed in thermal power units, enabling them to collect and store a large amount of real-time operating parameters. This provides a practical foundation for building data-driven carbon emission estimation models. Compared to traditional methods, data-driven models can integrate multi-dimensional input variables and fully explore the temporal correlations and nonlinear characteristics in historical operating data, thereby achieving real-time prediction of coal consumption and even carbon emission intensity. By combining the predicted coal consumption results with carbon emission coefficient conversion, dynamic estimation of carbon emissions and the carbon factor of electricity can be achieved.

[0004] Therefore, a data-driven carbon emission prediction model is needed. Unlike traditional methods that rely on theoretical assumptions or single load parameters, a data-driven carbon emission prediction model should be able to fully utilize high-frequency dynamic data collected from thermal power units under different operating conditions to deeply explore the nonlinear relationships and time-series dependencies among various operating parameters. Furthermore, by constructing an intelligent model with generalization capabilities, it can achieve accurate prediction of coal consumption for power generation and, combined with standard conversion factors, deduce carbon emissions and the carbon factor of electricity in real time, achieving the goals of high resolution and high real-time performance. Summary of the Invention

[0005] In view of the above-mentioned existing problems, the present invention provides a method, system, equipment and medium for constructing a data-driven calculation model for carbon emissions from coal-fired power units.

[0006] This invention provides a method, system, equipment, and medium for constructing a data-driven calculation model for carbon emissions from coal-fired power units, addressing the problem that traditional models lack the ability to model time-series characteristics and cannot depict the dynamic changes in carbon emissions over time.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for constructing a data-driven calculation model for carbon emissions from coal-fired power units, comprising: acquiring historical operating parameter data of the coal-fired power unit, preprocessing the historical operating parameter data, and obtaining first historical operating parameter data; Correlation analysis is performed on the first historical operating parameter data to extract key operating features. The key operating features are then constructed to obtain input samples. Using the input sample as input, a temporal convolutional network model is constructed using the first historical running parameter data; The time convolutional network model is subjected to global search and adaptive optimization using an optimization algorithm to obtain the optimized time convolutional network model. An optimized temporal convolutional network model is used to predict coal consumption for power generation. Based on the predicted coal consumption for power generation, the carbon factor is calculated, thereby enabling the construction of the calculation model driven by carbon emission data from coal-fired units.

[0008] As a preferred embodiment of the data-driven calculation model construction method for carbon emissions from coal-fired power units described in this invention, the key operating features extracted include: Set up a candidate operating feature set, calculate the correlation coefficient between each candidate operating feature and power supply coal consumption, and use it as the linear correlation index of the corresponding feature; Calculate the mutual information value between each candidate operating feature and power supply coal consumption, as a nonlinear dependence index of the corresponding feature; Normalize the linear correlation index and the nonlinear dependence index; The normalized linear correlation index and nonlinear dependence index are weighted and summed according to preset weight parameters to obtain a comprehensive score for each candidate operational feature. The candidate operational features are ranked according to the comprehensive score to obtain the key operational features.

[0009] The beneficial effect of this preferred technical solution is that by comprehensively evaluating linear and nonlinear relationships, key operating characteristics that significantly affect coal consumption for power supply are screened out, thereby improving the accuracy of model prediction and the reliability of feature selection.

[0010] As a preferred embodiment of the data-driven calculation model construction method for carbon emissions from coal-fired power units according to the present invention, the construction of the key operating characteristics includes: Based on key operational features, input samples are constructed using the first feature engineering method. Each input sample consists of key operational feature values ​​for the current time and each time within the previous preset time period, and the power supply coal consumption value corresponding to the current time is used as the output label; The input samples are divided into training set, test set and validation set.

[0011] As a preferred embodiment of the data-driven calculation model construction method for carbon emissions from coal-fired power units according to the present invention, the construction of the temporal convolutional network model includes: A temporal convolutional network model includes at least an input embedding layer, a causal convolutional layer, a dilated convolutional layer, residual connections and layer normalization structures, a regularization layer, and an output mapping layer. An optimizer is used to train the temporal convolutional network model, and the network parameters are updated by minimizing the error loss function between the predicted coal consumption for power supply and the actual value.

[0012] As a preferred embodiment of the data-driven calculation model construction method for carbon emissions from coal-fired power units according to the present invention, the global search and adaptive optimization of the temporal convolutional network model includes: Set the sand cat population size and maximum number of iterations, and randomly initialize hyperparameter combinations for each sand cat to obtain the initial population location; The model is trained using each set of hyperparameters, and the prediction error on the validation set is used as the fitness function to evaluate the quality of each individual and obtain the current global best individual. In each iteration, the control parameters and rotation angle parameters of the population are updated to simulate the hearing acuity and spatial positioning ability of the sand cat. For each individual, a search direction angle is generated through a random walk mechanism.

[0013] As a preferred embodiment of the data-driven calculation model construction method for carbon emissions from coal-fired power units described in this invention, it further includes: If the absolute value of the rotation angle parameter is less than or equal to 1, a local search strategy is used to guide the sand cat to accurately adjust its search direction in the high-frequency region. If the absolute value of the rotation angle parameter is greater than 1, then the global search strategy is used to expand the search range and escape the local optimum. If a sand cat goes out of bounds, it is corrected back into the search space, the current best individual is updated, and the sand cat position matrix is ​​maintained. When the maximum number of iterations is reached, or the optimal solution does not improve for several consecutive generations, the iteration is terminated early, and the elite individual is output as the final optimal hyperparameter combination.

[0014] The beneficial effects of this preferred technical solution are that it efficiently searches for hyperparameters through intelligent optimization algorithms, significantly improving the model's global optimization capability and convergence speed.

[0015] As a preferred embodiment of the data-driven calculation model construction method for carbon emissions from coal-fired power units described in this invention, the calculated electrical carbon factor includes: The optimized temporal convolutional network model is trained using the training set; The validation set is used to monitor the loss changes during training, and an early stopping mechanism is introduced to prevent model overfitting. The power supply coal consumption prediction performance of the trained temporal convolutional network model was evaluated using a test set; The coal consumption for power supply is predicted using a trained temporal convolutional network model to obtain the unit coal consumption value for power supply. The unit coal consumption for power generation is combined with the carbon content of the coal and the incomplete combustion coefficient. Through the preset carbon and carbon dioxide conversion relationship, the carbon dioxide emissions corresponding to the unit power generation are calculated as the carbon factor.

[0016] The beneficial effects of this preferred technical solution are that it enables high-precision coal consumption prediction and dynamic calculation of the carbon factor, thereby improving the accuracy of carbon emission management.

[0017] Secondly, the present invention provides a data-driven calculation model construction system for carbon emissions from coal-fired power units, comprising: The preprocessing module is used to acquire historical operating parameter data of the coal-fired power unit, preprocess the historical operating parameter data, and obtain the first historical operating parameter data. The feature extraction module is used to perform correlation analysis on the first historical operating parameter data, extract key operating features, construct the key operating features, and obtain input samples. The model building module is used to take the input sample as input and build a temporal convolutional network model using the first historical running parameter data; The model optimization module is used to perform global search and adaptive optimization on the temporal convolutional network model using optimization algorithms to obtain the optimized temporal convolutional network model. The calculation module is used to predict the coal consumption for power generation using an optimized temporal convolutional network model. Based on the predicted coal consumption for power generation, the carbon factor is calculated, thereby driving the construction of the calculation model using carbon emission data from coal-fired units.

[0018] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the method for constructing a data-driven calculation model for carbon emissions from coal-fired power units.

[0019] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the method for constructing a data-driven calculation model for carbon emissions from coal-fired power units.

[0020] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention introduces a Temporal Convolutional Network (TCN) as the core prediction structure, which can fully leverage the advantages of TCN in long-term time-series dependency modeling and sequence feature extraction. It can capture the deep feature patterns of the evolution of unit operating parameters over time, achieving high-precision fitting and prediction of power supply coal consumption and carbon emission factors. This invention addresses key hyperparameters involved in model training, such as network depth, kernel width, learning rate, and regularization parameters, by incorporating the Sand Cat Population Optimization Algorithm (SCSO) for global search and adaptive tuning. Compared with traditional heuristic algorithms, SCSO has stronger capabilities in convergence stability and escaping local optima, effectively improving the generalization ability and prediction accuracy of the model. This invention's method does not rely on static parameters such as coal quality and equipment efficiency that cannot be obtained in real time. Instead, it can perform multi-source feature modeling based on the online monitoring system operating data of thermal power plants, integrating key monitoring variables such as power load, steam temperature and pressure, wind speed, and water level to construct a dynamically responding carbon factor calculation system. This invention not only overcomes the problems of low modeling accuracy, poor real-time performance, and insufficient feature coverage of traditional methods, but also possesses good scalability and applicability, making it suitable for multiple key application scenarios such as power grid carbon factor accounting, power dispatch optimization, and regional carbon footprint management. This invention provides theoretical support and technical means for building efficient and accurate low-carbon power systems. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0022] Figure 1 This is a schematic diagram of the overall process logic of a method for constructing a data-driven calculation model for carbon emissions from coal-fired power units, provided as an embodiment of the present invention. Detailed Implementation

[0023] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0024] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for constructing a data-driven calculation model for carbon emissions from coal-fired power units is provided, comprising: S100: Obtain historical operating parameter data of the coal-fired unit, preprocess the historical operating parameter data, and obtain the first historical operating parameter data; In this embodiment of the invention, the historical operating parameter data of the coal-fired power unit includes the operation monitoring information system and other auxiliary monitoring systems deployed in the coal-fired power unit, covering historical operating data of the unit under different load levels and operating conditions. Data acquisition includes, but is not limited to, key parameters such as power generation, active load, boiler main steam temperature and pressure, reheat steam temperature, fuel flow rate, induced draft fan speed, inlet and outlet water temperature difference, and flue gas oxygen content. These operating variables can comprehensively reflect the combustion state and energy conversion efficiency of the power unit and are important inputs for modeling coal consumption and carbon emission behavior in power supply.

[0025] Considering the potential for abnormal fluctuations and data acquisition errors during the operation of thermal power units, to ensure data consistency and accuracy, the original dataset is first preliminarily reviewed to identify and remove records with a large number of missing data and outliers that clearly exceed the physical or engineering reasonable range. For example, for parameters such as boiler temperature, fuel consumption, and power generation, outliers are identified and removed or replaced by setting engineering boundary value ranges and sliding window statistical rules.

[0026] Specifically, preprocessing includes a preliminary review of the raw dataset to identify and remove records with a large number of missing data and outliers that are clearly outside the physical or engineering reasonable range. For example, for parameters such as boiler temperature, fuel consumption, and power generation, outliers are identified and removed or replaced by setting engineering boundary value ranges and sliding window statistical rules.

[0027] Because the physical dimensions and value ranges of various operating parameters differ significantly, directly using them for modeling may lead to an imbalance in the weights between features. Therefore, it is necessary to standardize their numerical scale. To this end, this invention uses the range standardization method to normalize each variable, expressed as: in, This is the original data. These are the minimum and maximum values ​​of the data, respectively. This is the normalized data.

[0028] It should be noted that the preprocessing method of the present invention effectively removes outliers and normalizes data, ensuring the consistency and accuracy of input data, significantly improving the stability and prediction accuracy of the model, and laying a solid foundation for high-precision carbon emission modeling.

[0029] S200: Perform correlation analysis on the first historical operating parameter data to extract key operating features, construct the key operating features, and obtain the input sample; In one optional embodiment, the correlation analysis can be Spearman rank correlation coefficient analysis, in which the observed values ​​of each feature variable and the target variable are sorted to obtain the rank, the difference between the rank of the feature variable and the rank of the target variable is calculated for each sample, the Spearman rank correlation coefficient is calculated, the calculated Spearman rank correlation coefficient is normalized, and the normalized Spearman rank correlation coefficient is combined with other indicators such as mutual information to construct a comprehensive feature score for feature selection. In another optional embodiment, the correlation analysis can also be maximum information coefficient analysis, which divides the values ​​of the feature variable and the target variable into different intervals to form a two-dimensional grid. At the resolution of each bin, the mutual information between the feature variable and the target variable is calculated. Among all bin resolutions, the maximum mutual information value is selected as the correlation index of the variable pair. The maximum information coefficient is normalized and combined with other correlation coefficients to construct a comprehensive feature score for feature selection. In this embodiment of the invention, the correlation analysis includes a method that combines Pearson correlation coefficient and mutual information analysis; Specifically, the core prediction target of the intelligent model is the coal consumption per unit of electricity supply. Carbon emissions and the carbon factor for electricity were both calculated. Therefore, to improve the generalization ability and stability of the coal consumption prediction model, it is necessary to select the key variables most significantly correlated with changes in coal consumption from the original operating parameters. To this end, this invention uses a method combining Pearson correlation coefficient and mutual information analysis to comprehensively characterize the linear and nonlinear dependencies between the characteristic variables and the target variable.

[0030] S300: Take the input sample as input and use the first historical running parameter data to build a temporal convolutional network model; S400: The optimization algorithm is used to perform global search and adaptive optimization on the temporal convolutional network model to obtain the optimized temporal convolutional network model. In one optional embodiment, the optimization algorithm can be a genetic algorithm. A set of initial solutions is randomly generated, each solution representing a set of model hyperparameters. The prediction error on the validation set is used as the fitness function to evaluate the performance of each individual. Based on fitness, high-fitness individuals are selected for the next generation. Two parent individuals are randomly selected, and new offspring individuals are generated by exchanging some of their genetic information. The newly generated offspring individuals undergo random mutation, altering certain genes with a certain probability. The newly generated offspring individuals replace individuals with lower fitness in the population, forming a new generation. When the preset maximum number of iterations is reached or the fitness no longer significantly improves, iteration stops, and the individual with the highest fitness is selected from the final population as the optimal hyperparameter combination. In another alternative embodiment, the optimization algorithm can also be a particle swarm optimization algorithm, which randomly initializes a group of particles, with the position of each particle representing a set of model hyperparameters, and initializes the velocity of each particle. The fitness of each particle at its current position is calculated, and the optimal position found by each particle during the search process is recorded. Among the individual optimal solutions of all particles, the one with the highest fitness is selected as the global optimal solution. Based on the individual optimal position and the global optimal position, the velocity and position of each particle are updated. The steps are repeated until the maximum number of iterations is reached, and the global optimal position is used as the optimal combination of hyperparameters. In this embodiment of the invention, the optimization algorithm includes the Sand Cat Population Optimization Algorithm (SCSO). SCSO simulates the behavior of sand cats, which rely on their keen hearing to perceive the location of prey, and introduces a perceptual domain and target guidance mechanism to achieve collaborative search among individuals. During the optimization process, each sand cat perceives direction based on its current optimal position and searches for a better solution within its perceptual radius to improve the efficiency of hyperparameter optimization. When the fitness changes little or gets stuck in a local optimum, some individuals will randomly jump out of the current search domain, reinitialize their positions, enhance global search capabilities, and avoid premature convergence. The algorithm iterates until the fitness function converges or the set maximum number of iterations is reached, and then terminates, outputting the optimal hyperparameter configuration for training the TCN model.

[0031] S500: It uses an optimized temporal convolutional network model to predict coal consumption for power generation. Based on the predicted coal consumption for power generation, it calculates the carbon factor for electricity, thereby driving the construction of the calculation model with carbon emission data of coal-fired units.

[0032] It should be noted that this invention proposes a carbon emission prediction method for coal-fired power units based on temporal convolutional neural networks (TCN), which can effectively mine the temporal features in operational data and improve the real-time performance and accuracy of carbon factor calculation. This invention introduces the Sand Cat Population Optimization (SCSO) algorithm to intelligently search and adaptively tune model hyperparameters, enhancing the model's global optimization capability and generalization performance. This invention establishes a data-driven modeling framework from multi-source real-time operational parameters to the output of the electric carbon factor, supporting minute-level dynamic estimation, and is suitable for power grid dispatching and carbon emission management scenarios.

[0033] In this embodiment of the invention, step S200 includes the following sub-steps A1-A5; In A1: Set up a candidate operating feature set, calculate the correlation coefficient between each candidate operating feature and power supply coal consumption, and use it as the linear correlation index of the corresponding feature; In A2: Calculate the mutual information value between each candidate operating feature and power supply coal consumption, as a nonlinear dependence index of the corresponding feature; In A3: Normalize the linear correlation index and the nonlinear dependency index; In A4: The normalized linear correlation index and nonlinear dependence index are weighted and summed according to preset weight parameters to obtain the comprehensive score of each candidate operating feature; In A5: Candidate operational features are ranked according to the comprehensive score to obtain key operational features.

[0034] In this embodiment of the invention, the sample size is set to be... The candidate feature set is , among which, the The first feature Each sample value is denoted as The coal consumption for power generation (predicted target) is denoted as , its first The value of each sample is Their sample means are respectively , .

[0035] Specifically, the Pearson correlation coefficient, used to measure the linear correlation between a feature and a target, is expressed as follows: in, For the first One candidate running feature, is the Pearson correlation coefficient, with a value range of [−1, 1]. The stronger the linear correlation, the larger the absolute value.

[0036] Taking the absolute value yields the linear correlation index. .

[0037] Considering the numerous nonlinear coupling relationships in the operation of coal-fired power units, mutual information is further introduced to characterize the nonlinear dependence between the features and the coal consumption for power supply. After binning, the variables are treated as discrete random variables, and the mutual information is expressed as: in, Features Values ,Target Values The joint probability, Features Marginal probability distribution, For the goal Marginal probability distribution, This is the mutual information value; the larger the value, the stronger the nonlinear relationship.

[0038] To integrate indices with different dimensions on the same scale, extreme value normalization is used to unify Pearson and mutual information into the [0,1] interval, as follows: The comprehensive feature score is represented as follows: in, For the first Normalized linear correlation of each feature For the first Normalized nonlinear dependence of each feature The linear information weight parameter has a range of values. It can be adjusted based on experience or the performance of the validation set. The final feature importance score is used to sort and filter important features. The higher the overall score, the more significant the impact of the feature on coal consumption for power supply, and thus it will be given priority in the final model input.

[0039] It should be noted that the combined screening mechanism of Pearson and mutual information can not only eliminate invalid variables with no significant statistical relationship, but also capture the nonlinear variation law of coal consumption for power supply, thereby improving the stability and accuracy of feature selection.

[0040] In this embodiment of the invention, after completing steps A1-A5, step S200 also includes steps A6-A8; In A6: Input samples are constructed based on key operational features using the first feature engineering method; In A7: Each input sample consists of key operational feature values ​​of the current time and each time within the previous preset time length, and the power supply coal consumption value corresponding to the current time is used as the output label; In A8: Input samples are divided into training set, test set and validation set.

[0041] In one optional embodiment, the first feature engineering method can be wavelet transform, which performs multi-level wavelet decomposition on the time series data to obtain approximation coefficients and detail coefficients at different levels, extracts features from the wavelet coefficients of each level, and combines the extracted features into input samples. In another alternative embodiment, the first feature engineering method can also be autocorrelation feature engineering, calculating the autocorrelation coefficient, selecting significant lag periods based on the autocorrelation plot or partial autocorrelation plot, constructing lag features for each significant lag period, and using these lag features as part of the input sample, combining the lag features with other features to form a complete input sample. In this embodiment of the invention, the first feature engineering method includes a sliding window structure; Specifically, a sliding window structure is used to construct time-series samples, combining the features of "current time + previous N times" into a sequence input to capture the temporal dependence of coal consumption as operating conditions change. This structure enables the model to understand the impact of historical operating states on the current coal consumption level, laying the foundation for subsequent training of the TCN temporal convolutional network.

[0042] The standardized and filtered feature data is transformed into sequential samples to construct input-output pairs, such as: data from the previous 6 hours → current coal consumption / carbon factor. The samples are divided into training, validation, and test sets in a ratio of 6:2:2 for model training, parameter tuning, and generalization ability evaluation.

[0043] It should be noted that by capturing temporal dependencies through a sliding window, the model's ability to perceive dynamic changes is improved, thereby enhancing prediction accuracy and generalization performance.

[0044] In this embodiment of the invention, step S300 includes the following sub-steps B1-B2; In B1: Temporal convolutional network models include at least an input embedding layer, a causal convolutional layer, a dilated convolutional layer, residual connections and layer normalization structures, a regularization layer, and an output mapping layer; In B2: The temporal convolutional network model is trained using an optimizer, and the network parameters are updated by minimizing the error loss function between the predicted coal consumption for power supply and the actual value.

[0045] In one optional embodiment, the regularization layer can be L2 regularization. An L2 regularization term is added to the loss function, and regularization is added to the original loss function to obtain a new loss function. During model training, the model weights are updated by optimizing the new loss function to keep the weights at a small value. The regularization coefficient is adjusted by methods such as cross-validation to find the optimal regularization strength. In another alternative embodiment, regularization can also be batch normalization, which calculates the mean and variance of each feature, normalizes the input data using the calculated mean and variance, performs scaling and offset operations on the normalized data, adds a batch normalization layer after each convolutional layer or fully connected layer to normalize the input data, normalizes the input data based on statistical data during training, and normalizes the data using the mean and variance of the entire training set during inference. In this embodiment of the invention, regularization includes Dropout regularization; Specifically, the standardized data input is fed into a one-dimensional convolutional network (1DConv). This layer maps the original low-dimensional input to a higher-dimensional feature space, thereby improving feature representation while preserving the time-series structure. This process can be expressed as: in, The input sequence is normalized. These are the embedded temporal features.

[0046] To ensure that the model predicts future values ​​using only historical and current information, a causal convolutional structure is introduced. Within each layer, the convolutional operation does not cross time steps to see future inputs, aligning with industrial forecasting logic.

[0047] By skipping connections to the inputs, the receptive field of the convolutional kernel is expanded, allowing the model to capture longer temporal dependencies. Dilated convolution introduces a dilation factor on top of standard convolution, enabling the kernel to compute over a larger range. The computation of each convolutional layer can be represented as: in, The dilation factor controls the distance between two adjacent convolution points in the convolution kernel. s The position in the sequence, This represents the size of the convolution kernel.

[0048] To accelerate convergence and prevent gradient vanishing, residual connections are added after each convolutional layer, and layer normalization is performed on the output to maintain numerical stability at each time step.

[0049] A Dropout layer is added after each residual block to effectively suppress the risk of overfitting by randomly deactivating some neurons. The Dropout ratio is dynamically searched and adaptively set by the SCSO algorithm during model training.

[0050] After multiple convolutions and feature extraction, the model finally maps the high-dimensional temporal features to the target variable, namely the predicted value of coal consumption for power supply at the current moment, through a fully connected network.

[0051] The model training process uses the mean squared error (MSE) as the objective function to measure the difference between the predicted value and the actual coal consumption, as shown below: Model training uses the Adam optimizer for gradient updates, and hyperparameters such as learning rate, number of convolutional channels, number of layers, kernel size, and dropout rate are jointly searched and adaptively adjusted by the Sand Cat Population Optimization (SCSO) algorithm to obtain the optimal model configuration.

[0052] It should be noted that the regularization and model building methods suppress overfitting by using Dropout random deactivation neurons, capture long-term temporal dependencies by combining causal convolution and dilated convolution, and accelerate convergence and stabilize training by using residual connections and layer normalization, which significantly improves the model's prediction accuracy and generalization ability, providing strong support for high-precision carbon emission prediction.

[0053] In this embodiment of the invention, step S400 includes the following sub-steps C1-C4; In C1: Set the sand cat population size and maximum number of iterations, and randomly initialize hyperparameter combinations for each sand cat to obtain the initial population position; In C2: The model is trained using each set of hyperparameters, and the prediction error on the validation set is used as the fitness function to evaluate the quality of each individual and obtain the current global best individual; In C3: In each iteration, the control parameters and rotation angle parameters of the population are updated to simulate the hearing acuity and spatial positioning ability of the sand cat. In C4: For each individual, a search direction angle is generated through a random walk mechanism.

[0054] In this embodiment of the invention, the sand cat swarm optimization algorithm belongs to an optimization method based on swarm intelligence, wherein individual solutions are modeled as numerical vectors. In a dimensional optimization problem, each sand cat represents a feasible solution, specifically represented as a... 3D array structure: Each element in the array All are floating-point numbers. Furthermore, they must satisfy the boundary constraints defined by the problem, meaning each variable should lie between its corresponding lower and upper bounds. In the algorithm initialization phase, at the beginning of the algorithm, the problem dimensions are first determined... and population size (in A random initial population matrix consisting of individual sand cats is generated, represented as follows: in, The population matrix of the sand cat group. For the first sand cat group Individual, For the first individual Each dimension.

[0055] The fitness cost of each sand cat is obtained by evaluating its corresponding fitness function, which quantifies the objective and constraints of the optimization problem. In the SCSO algorithm framework, this function guides the search process to find the optimal solution to the problem variables. Each sand cat calculates a fitness value based on its current position, serving as a measure of the quality of its solution. After each iteration, the algorithm selects the individual with the best fitness from the current population as the optimal solution for that iteration. If this solution is better than any solution recorded in previous iterations, it is updated as the global optimal solution. The remaining individuals then adjust their positions according to the direction of this optimal solution, moving towards a better region for the next round of searching.

[0056] In the algorithm, the optimal solution obtained in each iteration can be considered as the sand cat position closest to the "prey". To improve memory efficiency, if a better solution is found in a subsequent iteration, the original record is replaced; otherwise, the same or worse solution is not stored repeatedly, thus avoiding unnecessary resource consumption. This mechanism effectively improves the space utilization and running efficiency of the algorithm in the global search process.

[0057] The fitness function of a sand cat population is expressed as: in, Let be the fitness function of the sand cat population.

[0058] The sand cat's prey-hunting behavior primarily relies on its ability to perceive low-frequency sound signals. In the algorithm, the solution represented by each sand cat can be expressed as a... D Dimensional vector: Each dimension corresponds to a variable parameter in the optimization problem. To reflect this perceptual characteristic in the mathematical model, a sensitivity range parameter is defined. This is used to simulate the response range of a sand cat to low-frequency signals.

[0059] As described by biological characteristics, sand cats can perceive sound signals with frequencies below 2Hz. In the algorithm implementation, this parameter... The value will decrease linearly from the initial value of 2 to 0 during the iteration process. This mechanism is achieved through parameters. The model is constructed with a default value of 2 to reflect the typical characteristics of the sand cat's auditory system. This parameter can be adjusted appropriately according to actual needs for different types of optimization problems, thereby controlling the speed and range of individual search behavior. The algorithm also incorporates the current iteration number. With maximum number of iterations The ratio is part of the dynamic adjustment mechanism. The above mathematical model is expressed as: In the sand cat swarm optimization algorithm, the core parameter controlling the transition between the search (exploration) and development phases is... .in, This indicates that the sand cat is currently in a global exploration phase, at which point it locates the possible location of prey by sensing low-frequency sound signals. This behavior simulates the mechanism by which sand cats rely on hearing to forage in their natural environment.

[0060] At the same time, use This represents a random number uniformly distributed within the interval [0,1], used to enhance population diversity and improve the algorithm's global search performance. These parameters work together to achieve an effective balance between exploration and exploitation in the algorithm.

[0061] in, For the general sensitivity range that linearly decreases from 2 to 0, while This refers to the sensitivity range of each cat. In addition, Operations used in the exploration or utilization phase, Used for guidance parameters This is to enable the transfer of control between these stages.

[0062] In the Sand Cat Swarm Optimization (SCSO) algorithm, the individual's position update mechanism is determined by the global optimum, the individual's current location, and its perception range. This strategy allows the sand cat to explore potentially better regions in the search space, thereby discovering new local optima. The updated position is typically within a reasonable range between the current coordinates and the optimal solution; this design helps to gradually approach the global optimum. This mechanism enables the sand cat to dynamically adjust its position in the search space, thereby exploring more potential high-quality solution regions and increasing the likelihood of discovering better prey locations.

[0063] To ensure low time and space complexity, a random factor is introduced during the position update process. This not only enhances population diversity but also prevents premature convergence in the search process. The mathematical form of this search behavior can be modeled as follows: in, This represents the position of the candidate solution that performs best in the current population. For its current position, This refers to the sensing range parameter.

[0064] In this embodiment of the invention, after completing steps C1-C4, step S400 further includes steps C5-C8; In C5: If the absolute value of the rotation angle parameter is less than or equal to 1, the local search strategy is used to guide the sand cat to accurately adjust the search direction in the high-frequency region; In C6: If the absolute value of the rotation angle parameter is greater than 1, the global search strategy is used to expand the search range and escape the local optimum. In C7: When a sand cat goes out of bounds, it is corrected back into the search space, the current best individual is updated, and the sand cat position matrix is ​​maintained; In C8: When the maximum number of iterations is reached, or the optimal solution does not improve for several consecutive generations, the iteration is terminated early, and the elite individual is output as the final optimal hyperparameter combination.

[0065] In this embodiment of the invention, when the parameter At this point, the algorithm enters the development phase, where the sand cat simulates attacking its prey. This process is first based on the current globally optimal position. With the individual's current location A new, random position is generated to guide the individual toward a potential high-quality solution.

[0066] At this stage, assuming the sand cat's perception range is a circular area, a roulette wheel selection mechanism is used to randomly assign an angle to each sand cat. This enables multi-directional local search in two-dimensional space. Introducing random positions helps the individual get closer to the target prey, while the selection of random angles enhances the algorithm's ability to escape local optima, improving the diversity and convergence quality of the search process.

[0067] The attack behavior is represented as: in, This is the optimal position (optimal solution). The location is random.

[0068] In the Sand Cat Group Optimization (SCSO) algorithm, the balance between exploration and exploitation is achieved through two key adaptive parameters. and These two parameters enable the algorithm to dynamically switch between global search and local refinement, thereby improving overall optimization performance. Specifically, the parameters... The value depends on Its range of variation varies with It gradually contracts as its size decreases. When When decreasing linearly from the initial value of 2 to 0, The range of values ​​for also gradually shrinks from [−4,4] to zero. This design ensures that the algorithm has strong exploration capabilities in the early stages, while in the later stages it tends to develop high-quality solution regions.

[0069] Specifically, It is in the interval [−2] ,2 The value is randomly generated within the range [−1, 1]. When its value falls within the range [−1, 1], it indicates that the sand cat will move to some intermediate point between its current position and the optimal prey position, which corresponds to the development stage of the algorithm. Conversely, if... This triggers the exploration mechanism, guiding individuals to search for potential solutions in a broader search space.

[0070] This parameter setting strategy not only achieves a smooth transition between exploration and exploitation, but also automatically adjusts the weight of the two stages based on problem characteristics, enhancing the algorithm's adaptability and robustness. The mathematical description of this behavior is as follows: During the algorithm's exploration phase, each sand cat employs a different perception radius, enabling it to conduct diverse searches across a vast solution space and effectively avoid getting trapped in local optima. This mechanism not only enhances global exploration capabilities but also provides higher-quality candidate solutions for subsequent development phases, thus being a crucial factor in improving utilization efficiency.

[0071] The location update strategy describes the movement rules for the Sand Cat during the exploration and development phases. When the control parameters... When an individual enters development mode, its position update direction moves towards the current optimal solution, simulating attack behavior against prey; conversely, if... Then the individual performs an exploration operation, searching for new potential feasible regions in the entire solution space.

[0072] It should be noted that the algorithm's exploration-expansion balance mechanism enables it to exhibit good convergence speed and accuracy when dealing with high-dimensional complex problems or multi-objective optimization tasks. By continuously searching for new local optima globally, SCSO not only improves search efficiency but also enhances its adaptability and stability to complex problems.

[0073] In this embodiment of the invention, step S500 includes the following sub-steps D1-D5; In D1: The optimized temporal convolutional network model is trained using the training set; In D2: The validation set is used to monitor the loss changes during training, and an early stopping mechanism is introduced to prevent the model from overfitting; In D3: The power supply coal consumption prediction performance of the trained temporal convolutional network model is evaluated using the test set; In D4: The trained temporal convolutional network model is used to predict the coal consumption for power supply, and the unit coal consumption value for power supply is obtained. In D5: The unit coal consumption for power generation is combined with the carbon content of the coal and the incomplete combustion coefficient. Through the preset carbon and carbon dioxide conversion relationship, the carbon dioxide emissions corresponding to the unit power generation are calculated as the carbon factor.

[0074] In this embodiment of the invention, during training, the loss change on the validation set is monitored to avoid overfitting, and an EarlyStopping mechanism is introduced to improve the model's generalization ability. Finally, the model's performance on the power supply coal consumption prediction task is evaluated using a test set, with metrics including RMSE, MAE, and MAPE.

[0075] RMSE is the square root of the average of the squared errors between predicted and actual values. It measures the average deviation between predicted and actual values. A larger value indicates more or more severe large errors in the model, while a smaller value indicates a smaller overall deviation and more accurate predictions. It is expressed as: in, This is the actual value. These are predicted values.

[0076] MAE is the average absolute error between predicted and actual values. It measures the average absolute deviation between predicted and actual values. A larger MAE indicates a larger deviation in each prediction, while a smaller MAE indicates that the model's predictions are very close to the actual values. The formula is: in, This is the actual value. These are predicted values.

[0077] MAPE is the average percentage of the absolute error between predicted and true values. It measures the relative proportion of error to the true value, usually expressed as a percentage. A higher percentage indicates a larger relative error, while a lower percentage indicates higher relative accuracy of the model's predictions. It is represented as: in, This is the actual value; These are predicted values.

[0078] The calculated carbon factor includes, based on the principle of material balance, the amount of carbon generated when 1 kg of carbon is burned. kg of CO2. For coal-fired power units, their carbon emissions can be estimated using the following formula: in, Represents the amount of CO2 generated by the fixed combustion of coal (in tons); Represents raw coal consumption (tons); Represents the carbon content of coal; This represents the heat loss coefficient due to incomplete combustion. (Different coal types have different coefficients.) The following empirical values ​​can be used as a reference: bituminous coal / lignite: 0.01, lean coal: 0.015, anthracite: 0.025, low-quality anthracite: 0.04.

[0079] This study obtains the unit coal consumption value for power supply based on a data-driven prediction model. (Unit: g / kWh), the carbon intensity per unit of electricity is expressed as: in, Carbon dioxide emissions per unit of electricity.

[0080] The final result It represents the CO2 emissions per kilowatt-hour of electricity generated, and is used to measure the carbon emission level in the electricity production process.

[0081] The above is an illustrative scheme of a data-driven calculation model construction method for carbon emissions from coal-fired power units according to this embodiment. It should be noted that the technical solution of this data-driven calculation model construction system for carbon emissions from coal-fired power units belongs to the same concept as the technical solution of the aforementioned data-driven calculation model construction method for carbon emissions from coal-fired power units. Details not described in detail in the technical solution of the data-driven calculation model construction system for carbon emissions from coal-fired power units in this embodiment can be found in the description of the technical solution of the aforementioned data-driven calculation model construction method for carbon emissions from coal-fired power units.

[0082] The carbon emission data-driven calculation model construction system for coal-fired power units in this embodiment includes: The preprocessing module is used to acquire historical operating parameter data of the coal-fired power unit, preprocess the historical operating parameter data, and obtain the first historical operating parameter data. The feature extraction module is used to perform correlation analysis on the first historical operating parameter data, extract key operating features, construct the key operating features, and obtain input samples. The model building module is used to take the input sample as input and build a temporal convolutional network model using the first historical running parameter data; The model optimization module is used to perform global search and adaptive optimization on the temporal convolutional network model using optimization algorithms to obtain the optimized temporal convolutional network model. The calculation module is used to predict the coal consumption for power generation using an optimized temporal convolutional network model. Based on the predicted coal consumption for power generation, the carbon factor is calculated, thereby driving the construction of the calculation model using carbon emission data from coal-fired units.

[0083] This embodiment also provides a computer device suitable for constructing a data-driven calculation model for carbon emissions from coal-fired power units, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a data-driven calculation model construction method for carbon emissions from coal-fired power units, as proposed in the above embodiments.

[0084] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a data-driven calculation model construction method for carbon emissions from coal-fired power units as proposed in the above embodiments.

[0085] The storage medium proposed in this embodiment and the method for constructing a data-driven calculation model for carbon emissions from coal-fired power units proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0086] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computing device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0087] Example 2 differs from the first example. It provides a verification test of a method for constructing a data-driven calculation model for carbon emissions from coal-fired power units, and verifies and explains the technical effects used in this method.

[0088] The unit collects the following operating variables: boiler thermal status: Main_Steam_Temp (main steam temperature), Main_Steam_Pressure (main steam pressure), Reheat_Steam_Temp (reheat steam temperature); fuel input: Coal_Flow (fuel flow rate); combustion status: Oxygen_Content (furnace oxygen content); load level: Generator_Output (power generation); auxiliary equipment operation: Fan_Speed ​​(fan speed), FW_Delta_T (feedwater temperature difference). Preprocessing includes removing outliers (such as excessively high main steam temperature or excessively low oxygen content) based on the project scope; and performing normalization on each variable to ensure that parameters of different dimensions can be input into the model. Feature selection and construction include the prediction target: Y = Coal_Consumption; For each feature When performing Pearson calculations on parameters such as Main_Steam_Temp, Coal_Flow, and Oxygen_Content with coal consumption: For the first The actual measured value at any given time, such as the main steam temperature; This represents the actual coal consumption at the corresponding moment; , This represents the average value over the time window.

[0089] The implications of mutual information in various scenarios are as follows: For example, the probability that the main steam temperature falls within a certain temperature range; Coal consumption distribution; The probability of temperature range and coal consumption range occurring simultaneously, etc.

[0090] The key features ultimately selected by the model typically include: Main_Steam_Temp, Coal_Flow, Oxygen_Content, Reheat_Steam_Temp, Generator_Output, etc.

[0091] The feature construction part combines the parameters from the past N time steps into a window sequence, for example: → ; Example of constructed input sequence: The ratio of training set, validation set, and test set is 6:2:2.

[0092] The TCN model construction includes: Input embedding layer: The input matrix processed by Conv1D consists of the above-mentioned unit parameters, which are used to extract local time variation patterns such as "steam temperature rises → coal consumption decreases".

[0093] Causal convolution: ensures that predicted coal consumption depends only on historical parameters (such as past Coal Flow and oxygen changes).

[0094] Dilated convolution: Captures time-delay behavior during runtime, for example: Load increases (Generator_Output) → Fuel flow increases after a few minutes (Coal_Flow) → Coal consumption changes.

[0095] Residual connectivity and normalization: Improve the training stability of the model in complex multi-parameter scenarios.

[0096] Output layer: Outputs the predicted coal consumption value at the current moment.

[0097] SCSO optimization includes: Each "sandcat individual" corresponds to a set of TCN hyperparameters, such as: kernel size: determines whether the dynamic chain of load-fuel-temperature can be captured; expansion factor: corresponds to the time delay of energy transfer in the unit; dropout: suppresses the influence of monitoring noise such as oxygen fluctuations; learning rate: controls the convergence speed of the model.

[0098] The fitness function is used to verify the coal consumption error. When a certain parameter combination fails to effectively learn the pattern such as "Oxygen_Content decreases → coal consumption increases", the fitness is automatically reduced.

[0099] Evaluation metrics include: RMSE: measures the overall error in coal consumption; MAE: mean deviation; MAPE: error percentage, suitable for scenarios sensitive to carbon factor conversion; EarlyStopping is used in training to prevent overfitting.

[0100] The calculation of the electrocarbon factor includes: TCN prediction. (Unit: g / kWh), the carbon factor calculated based on coal quality parameters is expressed as: in, TCN-predicted unit coal consumption; Carbon content; : Incomplete combustion coefficient, related to Oxygen_Content.

[0101] By collecting and preprocessing multi-source operating parameters, combined with feature selection and time series modeling, the power supply coal consumption can be accurately predicted and the carbon factor can be calculated, which significantly improves the accuracy and real-time performance of carbon emission accounting and provides strong support for low-carbon dispatch and management in the power industry.

[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not 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 solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for constructing a data-driven calculation model for carbon emissions from coal-fired power units, characterized in that, include: Historical operating parameter data of the coal-fired power unit is obtained, and the historical operating parameter data is preprocessed to obtain the first historical operating parameter data; Correlation analysis is performed on the first historical operating parameter data to extract key operating features. The key operating features are then constructed to obtain input samples. Using the input sample as input, a temporal convolutional network model is constructed using the first historical running parameter data; The time convolutional network model is subjected to global search and adaptive optimization using an optimization algorithm to obtain the optimized time convolutional network model. An optimized temporal convolutional network model is used to predict coal consumption for power generation. Based on the predicted coal consumption for power generation, the carbon factor is calculated, thereby enabling the construction of the calculation model driven by carbon emission data from coal-fired units.

2. The method for constructing a data-driven calculation model for carbon emissions from coal-fired power units as described in claim 1, characterized in that, The key operational features extracted include: Set up a candidate operating feature set, calculate the correlation coefficient between each candidate operating feature and power supply coal consumption, and use it as the linear correlation index of the corresponding feature; Calculate the mutual information value between each candidate operating feature and power supply coal consumption, as a nonlinear dependence index of the corresponding feature; Normalize the linear correlation index and the nonlinear dependence index; The normalized linear correlation index and nonlinear dependence index are weighted and summed according to preset weight parameters to obtain a comprehensive score for each candidate operational feature. The candidate operational features are ranked according to the comprehensive score to obtain the key operational features.

3. The method for constructing a data-driven calculation model for carbon emissions from coal-fired power units as described in claim 2, characterized in that, The construction of the key operational features includes: Based on key operational features, input samples are constructed using the first feature engineering method. Each input sample consists of key operational feature values ​​for the current time and each time within the previous preset time period, and the power supply coal consumption value corresponding to the current time is used as the output label; The input samples are divided into training set, test set and validation set.

4. The method for constructing a data-driven calculation model for carbon emissions from coal-fired power units as described in claim 3, characterized in that, Building a temporal convolutional network model includes: A temporal convolutional network model includes at least an input embedding layer, a causal convolutional layer, a dilated convolutional layer, residual connections and layer normalization structures, a regularization layer, and an output mapping layer. An optimizer is used to train the temporal convolutional network model, and the network parameters are updated by minimizing the error loss function between the predicted coal consumption for power supply and the actual value.

5. A method for constructing a data-driven calculation model for carbon emissions from coal-fired power units as described in claim 1 or 4, characterized in that, The global search and adaptive optimization of the temporal convolutional network model includes: Set the sand cat population size and maximum number of iterations, and randomly initialize hyperparameter combinations for each sand cat to obtain the initial population location; The model is trained using each set of hyperparameters, and the prediction error on the validation set is used as the fitness function to evaluate the quality of each individual and obtain the current global best individual. In each iteration, the control parameters and rotation angle parameters of the population are updated to simulate the hearing acuity and spatial positioning ability of the sand cat. For each individual, a search direction angle is generated through a random walk mechanism.

6. The method for constructing a data-driven calculation model for carbon emissions from coal-fired power units as described in claim 5, characterized in that, Also includes: If the absolute value of the rotation angle parameter is less than or equal to 1, a local search strategy is used to guide the sand cat to accurately adjust its search direction in the high-frequency region. If the absolute value of the rotation angle parameter is greater than 1, then the global search strategy is used to expand the search range and escape the local optimum. If a sand cat goes out of bounds, it is corrected back into the search space, the current best individual is updated, and the sand cat position matrix is ​​maintained. When the maximum number of iterations is reached, or the optimal solution does not improve for several consecutive generations, the iteration is terminated early, and the elite individual is output as the final optimal hyperparameter combination.

7. The method for constructing a data-driven calculation model for carbon emissions from coal-fired power units as described in claim 6, characterized in that, The calculated electrocarbon factor includes: The optimized temporal convolutional network model is trained using the training set; The validation set is used to monitor the loss changes during training, and an early stopping mechanism is introduced to prevent model overfitting. The power supply coal consumption prediction performance of the trained temporal convolutional network model was evaluated using a test set; The coal consumption for power supply is predicted using a trained temporal convolutional network model to obtain the unit coal consumption value for power supply. The unit coal consumption for power generation is combined with the carbon content of the coal and the incomplete combustion coefficient. Through the preset carbon and carbon dioxide conversion relationship, the carbon dioxide emissions corresponding to the unit power generation are calculated as the carbon factor.

8. A system for constructing a data-driven calculation model for carbon emissions from coal-fired power units, using the data-driven calculation model construction method for carbon emissions from coal-fired power units as described in any one of claims 1 to 7, characterized in that, include: The preprocessing module is used to acquire historical operating parameter data of the coal-fired power unit, preprocess the historical operating parameter data, and obtain the first historical operating parameter data. The feature extraction module is used to perform correlation analysis on the first historical operating parameter data, extract key operating features, construct the key operating features, and obtain input samples. The model building module is used to take the input sample as input and build a temporal convolutional network model using the first historical running parameter data; The model optimization module is used to perform global search and adaptive optimization on the temporal convolutional network model using optimization algorithms to obtain the optimized temporal convolutional network model. The calculation module is used to predict the coal consumption for power generation using an optimized temporal convolutional network model. Based on the predicted coal consumption for power generation, the carbon factor is calculated, thereby driving the construction of the calculation model using carbon emission data from coal-fired units.

9. A computer device, characterized in that, include: The system includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for constructing a data-driven calculation model for carbon emissions from a coal-fired power unit as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the steps of the method for constructing a data-driven calculation model for carbon emissions from a coal-fired power unit as described in any one of claims 1 to 7.

Images