Coal received basis elemental carbon content regression system and method of operation

Abstract

The application discloses a regression calculation method for received base element carbon content of coal, and belongs to the technical field of received base element carbon content measurement of coal, which comprises the following steps: obtaining coal quality original data after analysis and measurement, performing maximum minimum normalization processing to obtain normalized original data, training and authenticating a MISM received base element carbon content regression calculation model of coal by using the normalized original data, processing actual measured coal quality original data by using the trained and authenticated model, and calculating the received base element carbon content of coal. The application designs the MISM received base element carbon content regression calculation model of coal, can capture important features of coal quality data, establish the relationship between the coal quality data and the received base element carbon content of coal, and can obtain the received base element carbon content of coal through regression calculation only by measuring part of the coal quality original data, so that the detection cost of the received base element carbon content of coal is reduced, and the accuracy of the calculation result is ensured.

Description

Technical Field

[0001] This invention belongs to the field of measurement technology of carbon content of coal as received, specifically involving a regression calculation method for carbon content of coal as received. Background Technology

[0002] As an important method for calculating carbon emissions, the carbon nuclear algorithm provides a guarantee for the fairness and impartiality of carbon market transactions in the thermal power industry.

[0003] Carbon accounting requires enterprises to strictly adhere to relevant standards and methods in collecting, compiling, and calculating key data to ensure the accuracy of carbon emission results. According to the accounting requirements in the "Guidelines for Enterprise Greenhouse Gas Emission Accounting and Reporting (Power Generation Facilities)," the received-based elemental carbon content of coal is a key measured parameter in carbon accounting for thermal power plants. On the one hand, the received-based elemental carbon content directly affects the accuracy of the enterprise's carbon accounting and is a guarantee for the enterprise to bear reasonable compliance costs. On the other hand, the received-based elemental carbon content needs to be calculated from other coal quality analysis parameters and requires auxiliary parameters to verify data reliability. In addition to measuring elemental carbon content, the parameters that need to be measured include total moisture, air-dried moisture, air-dried total sulfur, air-dried ash, air-dried volatile matter, air-dried hydrogen content, air-dried fixed carbon, and bomb calorific value—a total of eight parameters. Measuring these parameters requires various corresponding instruments, the measurement process is complex and consumes a lot of manpower, resulting in significant testing costs for enterprises. If a company does not conduct actual elemental carbon measurements, it will need to use default values ​​to calculate carbon emissions, resulting in an overestimation of the company's compliance costs.

[0004] To ensure enterprises bear reasonable performance costs and reduce their testing costs, it is necessary to consider how to accurately estimate the elemental carbon content of coal based on received samples while minimizing the number of measured parameters. Based on existing measurement data, if a calculation model can be constructed to regress the elemental carbon content of coal based on received samples using eight parameters as inputs—total moisture, air-dried moisture, air-dried total sulfur, air-dried ash, air-dried volatile matter, air-dried hydrogen content, air-dried fixed carbon, and bomb calorific value—into the actual calculation, only some parameters need to be measured, such as total moisture, air-dried moisture, air-dried ash, air-dried volatile matter, air-dried fixed carbon, and bomb calorific value (six parameters in total). Parameters not measured are assigned default values, such as air-dried total sulfur and air-dried hydrogen content. This reduces the number of testing parameters and instruments required, thus lowering enterprise testing costs. Therefore, how to construct a regression model for the elemental carbon content of coal based on received samples using only a subset of input parameters is a key issue that needs to be addressed. Summary of the Invention

[0005] To address the aforementioned shortcomings in existing technologies, this invention provides a regression calculation method for the carbon content of coal as received as an element, and constructs a regression model for the carbon content of coal as received as an element, thus solving the problems of high carbon accounting costs and low calculation efficiency for enterprises.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: a regression calculation method for the carbon content of coal as a basic element, comprising the following steps:

[0007] S1: Obtain the raw coal quality data after analysis and measurement, perform maximum and minimum normalization processing to obtain normalized raw data, and divide the normalized raw data into training data and validation data;

[0008] S2: Using the training data as input data, the MISM coal received elemental carbon content regression calculation model is used to process the data to obtain the trained MISM coal received elemental carbon content regression calculation model.

[0009] S3: Using the validation data as input data, the trained MISM coal received elemental carbon content regression calculation model is used to process the data to obtain the certified MISM coal received elemental carbon content regression calculation model.

[0010] S4: Obtain the raw measured coal quality data and perform preprocessing operations to obtain preprocessed raw measured coal quality data;

[0011] S5: The pre-processed measured raw coal quality data is used as input data and processed using the certified MISM coal received-based elemental carbon content regression calculation model to obtain the coal received-based elemental carbon content.

[0012] The beneficial effects of this invention are as follows: This invention designs an ISSM mapping layer model and incorporates a Mamba module to build a MISM coal-based carbon content regression calculation model. This model has the ability to capture important features and can establish the relationship between measured data and coal-based carbon content by adjusting network structure parameters. This allows the coal-based carbon content to be obtained through regression calculation using only partial measured data, thereby reducing the detection cost of coal-based carbon content and reducing the measured parameters required for coal quality analysis for thermal power plants. At the same time, the regression calculation results ensure the accuracy of coal-based carbon content.

[0013] Furthermore: the raw coal quality data after analysis and measurement include: total moisture content, air-dried moisture content, air-dried total sulfur content, air-dried ash content, air-dried volatile matter content, air-dried hydrogen content, air-dried fixed carbon content, bomb calorific value, and coal-as-received elemental carbon content;

[0014] The normalized raw data includes: normalized total moisture content, normalized air-dried basis moisture, normalized air-dried basis total sulfur, normalized air-dried basis ash, normalized air-dried basis volatile matter, normalized air-dried basis hydrogen content, normalized air-dried basis fixed carbon, normalized bomb calorific value, and normalized coal-as-it-as-it-based elemental carbon content.

[0015] The measured raw coal quality data of the pretreatment include: total moisture content of the pretreatment, moisture content of the pretreatment air-dried basis, ash content of the pretreatment air-dried basis, volatile matter of the pretreatment air-dried basis, fixed carbon of the pretreatment air-dried basis, and calorific value of the pretreatment bomb.

[0016] The beneficial effects of the above-mentioned further scheme are as follows: The present invention establishes the relationship between measured data and the carbon content of coal received by the coal by analyzing and measuring the original coal quality data and normalized original data. Then, only the pre-processed original data, that is, part of the measured data, is needed to complete the calculation of the carbon content of coal received by the coal.

[0017] Furthermore, the specific steps for processing using the MISM coal-based elemental carbon content regression calculation model are as follows:

[0018] A1: Use the input data as the input features of the Mamba mapping layer, and use the Mamba mapping layer to perform feature mapping to obtain the first mapped features;

[0019] A2: Add the first mapping feature to the input data to obtain the first summed feature;

[0020] A3: Based on the first summation feature, integrate activation is performed using a linear activation layer to obtain the integrated feature;

[0021] A4: Use the integrated features as input features of the Mamba mapping layer, and use the Mamba mapping layer to perform feature mapping to obtain the second mapped features;

[0022] A5: Add the second mapping feature and the first summation feature to obtain the second summation feature;

[0023] A6: Based on the second summation feature, the linear activation layer is used for integration activation to obtain the carbon content of the basic element of coal.

[0024] The beneficial effects of the above-mentioned further scheme are as follows: by establishing a regression calculation model for the carbon content of coal received by MISM and processing the data after coal quality analysis, the carbon content of coal received by MISM can be obtained. Compared with traditional methods, it has higher calculation efficiency and accuracy.

[0025] Furthermore, the mathematical expression of the regression calculation model for the carbon content of the basic element in MISM coal is as follows:

[0026] F6 = Linear[LeakyReLU(F5)]

[0027] F5 = F4 + F2

[0028] F4 = Mamba(F3)

[0029] F3 = Linear[LeakyReLU(F2)]

[0030] F2 = F1 + x in-norm

[0031] F1 = Mamba(x in-norm )

[0032]

[0033] Wherein, F6 represents the carbon content of the basic element of coal, F5 is the second additive feature, F4 is the second mapping feature, F3 is the integrated feature, F2 is the first additive feature, F1 is the first mapping feature, and x in-norm For the input data, Linear[leakyReLU()] is the linear activation layer function, Mamba() is the Mamba mapping layer function, v is the input of the LeakyReLU function, and α is a constant set.

[0034] The beneficial effects of the above-mentioned further solutions are as follows: This invention establishes a regression calculation model for the carbon content of coal received by the MISM coal through the Mamba mapping layer and the LeakyReLU activation function, which makes the model have better generalization and data processing capabilities, and can effectively extract and process complex features in coal data, making the calculation results of the carbon content of coal received by the MISM coal more accurate and stable.

[0035] Furthermore: The Mamba mapping layer performs feature mapping, and its specific implementation is as follows:

[0036] B1: Obtain the input features of the Mamba mapping layer;

[0037] B2: Based on the input features, normalize them using the root mean square of the features to obtain normalized input features;

[0038] B3: Based on the normalized input features, a linear layer is used for processing to obtain the third mapping feature;

[0039] B4: Based on the third mapping feature, process it using a 1D convolutional layer to obtain the convolutional feature;

[0040] B5: Based on the convolutional features, the activation layer is used to process them to obtain the first activation feature;

[0041] B6: Based on the first activation feature, the fourth mapping feature is obtained by processing it using a linear layer;

[0042] B7: Use the fourth mapping feature as the input feature of the ISSM mapping layer, and use the ISSM mapping layer to perform mapping to obtain the fifth mapping feature;

[0043] B8: Based on the fifth mapping layer, the activation layer is used for processing to obtain the second activation feature;

[0044] B9: Based on the input features, a linear layer is used for processing to obtain the sixth mapping feature;

[0045] B10: Based on the sixth mapping feature, the third activation feature is obtained by processing it using the activation layer;

[0046] B11: Multiply the third activation feature and the second activation feature to obtain a new feature;

[0047] B12: Based on the new features, a linear layer is used for processing to obtain the mapping features output by the Mamba mapping layer.

[0048] The beneficial effects of the above-mentioned further solutions are as follows: The Mamba mapping layer of the present invention combines feature normalization, multi-level feature extraction, activation function, feature fusion and ISSM mapping layer, which can improve the generalization ability and prediction performance of the model in complex machine learning tasks. At the same time, the Mamba mapping layer can also improve the performance of the MISM coal-fired basic element carbon content regression calculation model.

[0049] Furthermore, the mathematical expression for the Mamba mapping layer is as follows:

[0050] F 15 =Linear(F 14 )

[0051] F 14 =F 13 ×F 11

[0052] F 13 =SiLU(F 12 )

[0053] F 12 =Linear(x) in )

[0054] F 11 =SiLU(F 10 )

[0055] F 10 =ISSM(F9)

[0056] F9 = Linear(F8)

[0057] F8 = SiLU(F7)

[0058] F7 = 1DConv(F6)

[0059] F6 = Linear(x) RMS )

[0060]

[0061]

[0062]

[0063] Among them, F 15 F represents the mapping features output by the Mamba mapping layer. 14 For new features, F 13 As the third activation feature, F 12 For the sixth mapping feature, F 11 For the second activation feature, F 10 F9 is the fifth mapping layer, F8 is the fourth mapping layer, F8 is the first activation feature, F7 is the convolutional feature, F6 is the third mapping feature, and x RMS For the normalized input features, SiLU() is the Sigmoid-gated linear unit activation function, Linear() is the linear layer function, ISSM() is the ISSM mapping layer function, 1DConv() is the 1D convolutional layer function, and x in The input features of the Mamba mapping layer, The scaling result of the input features to the Mamba mapping layer, where params is an 8-dimensional trainable parameter matrix, and μ RMS The root mean square of the normalized feature, ε is a set constant, and x i represents the measured coal quality parameters, and w is the input to the Sigmoid-gated linear unit activation function.

[0064] The beneficial effects of the above-mentioned further scheme are as follows: the processing flow of the Mamba mapping layer can be clearly and concisely displayed through mathematical expressions; through the Mamba mapping layer, the MISM coal-fired basic element carbon content regression calculation model can process data more flexibly; at the same time, the Mamba mapping layer can be modified to meet different data characteristics, thereby enhancing the adaptability and practicality of the MISM coal-fired basic element carbon content regression calculation model.

[0065] Furthermore: the ISSM mapping layer performs mapping, and its specific implementation is as follows:

[0066] C1: Obtain the input features of the ISSM mapping layer and the previous state matrix of the ISSM mapping layer, and calculate the system matrix, control matrix, output matrix, and direct transfer matrix of the ISSM mapping layer respectively.

[0067] C2: Multiply the input features by the control matrix of the ISSM mapping layer to obtain the control features;

[0068] C3: Multiply the previous state matrix of the ISSM mapping layer and the system matrix of the ISSM mapping layer to obtain the state characteristics;

[0069] C4: Add the state features and control features to obtain the current state matrix of the ISSM mapping layer;

[0070] C5: Multiply the output matrix of the ISSM mapping layer with the current state matrix of the ISSM mapping layer to obtain the first feature;

[0071] C6: Multiply the direct transfer matrix of the ISSM mapping layer with the input matrix to obtain the second feature;

[0072] C7: Add the first feature and the second feature together to obtain the fifth mapping feature output by the ISSM mapping layer.

[0073] The beneficial effects of the above-mentioned further solutions are as follows: The present invention processes data features through the ISSM mapping layer, which can effectively integrate and transform features, enhance the model's dynamic adjustment capability and real-time responsiveness, enable the model to have better performance when processing data, and output mapping features more accurately.

[0074] Furthermore, the calculation methods for the system matrix, control matrix, output matrix, and direct transfer matrix of the ISSM mapping layer in C1 are as follows:

[0075] D1: Obtain the input features of the ISSM mapping layer;

[0076] D2: Set up a matrix of dimension M×N and initialize it using L2 norm normalization to obtain an initialized matrix, where M is the number of rows of the matrix and N is the number of columns of the matrix;

[0077] D3: Based on the input features of the ISSM mapping layer, four different linear layers are used for processing to obtain the seventh mapping feature, the eighth mapping feature, the ninth mapping feature, and the output matrix of the ISSM mapping layer.

[0078] D4: Based on the seventh mapping feature, the fourth activation feature is obtained by processing it using the activation layer;

[0079] D5: Multiply the fourth activation feature by the eighth mapping feature to obtain the control matrix of the ISSM mapping layer;

[0080] D6: Multiply the fourth activation feature by the initialized matrix to obtain the system features;

[0081] D7: Perform an exponential transformation on the system characteristics to obtain the system matrix of the ISSM mapping layer;

[0082] D8: Based on the ninth mapping feature, activate the activation layer to obtain the direct transfer matrix of the ISSM mapping layer.

[0083] The beneficial effects of the above-mentioned further scheme are as follows: by calculating the system matrix, control matrix, output matrix, and direct transfer matrix of the ISSM mapping layer, it is easier for the subsequent ISSM mapping layer to process the input data features and improve computational efficiency.

[0084] Furthermore, the mathematical expressions for the system matrix, control matrix, output matrix, and direct transfer matrix of the ISSM mapping layer are as follows:

[0085]

[0086] D = SiLU(F S7 )

[0087]

[0088] B = F S8 ×F S6

[0089] F S8 =SiLU(F S5 )

[0090] C = Linear4(F)

[0091] F S7 =Linear3(F)

[0092] F s6 =Linear2(F)

[0093] F S5 =Linear1(F)

[0094]

[0095] Where A is the system matrix of the ISSM mapping layer, B is the control matrix of the ISSM mapping layer, C is the output matrix of the ISSM mapping layer, D is the direct transfer matrix of the ISSM mapping layer, and F is the input feature of the ISSM mapping layer. S9 For system characteristics, F S8 As the fourth activation feature, F S7 For the ninth mapping feature, F S6 For the eighth mapping feature, F S5 The seventh mapping feature consists of four distinct linear layers: Linear4, Linear3, Linear2, and Linear1. SiLU() is the Sigmoid-gated linear unit activation function. Let be the initialized matrix, a be a matrix of dimension 16×N, and ‖ ‖2 be the L2 norm normalization.

[0096] The beneficial effects of the above-mentioned further scheme are as follows: by calculating the system matrix, control matrix, output matrix, and direct transfer matrix of the ISSM mapping layer using the above formulas, the practicality and reliability of the ISSM mapping layer can be improved, and the scientific nature of the scheme can be demonstrated.

[0097] Furthermore, the specific steps of S3 are as follows:

[0098] S301: Using the validation data as input data, the predicted normalized carbon content of coal received by the base element is calculated by the trained MISM coal received base element carbon content regression calculation model.

[0099] S302: Perform an inverse normalization operation on the predicted normalized carbon content of coal received by the base element to obtain the predicted carbon content of coal received by the base element.

[0100] S303: The average relative error between the calculated and predicted carbon content of coal received on the basis and the carbon content of coal received on the basis in the original coal quality data;

[0101] S304: Determine whether the average relative error is less than or equal to the set threshold. If yes, obtain the certified MISM coal-fired received element carbon content regression calculation model. Otherwise, optimize the MISM coal-fired received element carbon content regression calculation model and return to S2.

[0102] The beneficial effects of the above-mentioned further scheme are as follows: using the average relative error to determine the carbon content of MISM coal based on the elemental regression calculation model can ensure the reliability and accuracy of the model and improve the accuracy of the prediction results. Attached Figure Description

[0103] Figure 1A flowchart of a regression calculation method for the carbon content of basic elements in coal;

[0104] Figure 2 This is a flowchart of a regression calculation model for the carbon content of MISM coal received as a basic element in this invention.

[0105] Figure 3 This is a flowchart of a Mamba mapping layer calculation according to the present invention;

[0106] Figure 4 This is a flowchart of the ISSM mapping layer calculation process according to the present invention;

[0107] Figure 5 This is a flowchart illustrating the calculation of the parameter matrix of an ISSM mapping layer according to the present invention. Detailed Implementation

[0108] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0109] like Figure 1 The diagram shows a flowchart of a regression calculation method for the carbon content of coal as a basic element, which includes the following steps:

[0110] S1: Obtain the raw coal quality data after analysis and measurement, perform maximum and minimum normalization processing to obtain normalized raw data, and divide the normalized raw data into training data and validation data;

[0111] S2: Using the training data as input data, the MISM coal received elemental carbon content regression calculation model is used to process the data to obtain the trained MISM coal received elemental carbon content regression calculation model.

[0112] S3: Using the validation data as input data, the trained MISM coal received elemental carbon content regression calculation model is used to process the data to obtain the certified MISM coal received elemental carbon content regression calculation model.

[0113] S4: Obtain the raw measured coal quality data and perform preprocessing operations to obtain preprocessed raw measured coal quality data;

[0114] S5: The pre-processed measured raw coal quality data is used as input data and processed using the certified MISM coal received-based elemental carbon content regression calculation model to obtain the coal received-based elemental carbon content.

[0115] In one embodiment of the present invention, the specific method for obtaining raw coal quality data is as follows:

[0116] M1: Sample the coal entering the furnace to obtain sampled coal blocks;

[0117] M2: Further crush the sampled coal block to obtain a coal sample with a particle size of 3mm;

[0118] M3: Using specialized equipment to analyze the coal sample and obtain raw coal quality data.

[0119] The raw coal quality data after analysis and measurement include: total moisture content, air-dried moisture content, air-dried total sulfur content, air-dried ash content, air-dried volatile matter content, air-dried hydrogen content, air-dried fixed carbon content, bomb calorific value, and carbon content of coal as received.

[0120] The normalized raw data include: normalized total moisture content, normalized air-dried basis moisture, normalized air-dried basis total sulfur, normalized air-dried basis ash, normalized air-dried basis volatile matter, normalized air-dried basis hydrogen content, normalized air-dried basis fixed carbon, normalized bomb calorific value, and normalized coal-as-it-as-it-based elemental carbon content.

[0121] The measured raw data of pretreated coal include: total moisture content of pretreated coal, moisture content of pretreated air-dried coal, ash content of pretreated air-dried coal, volatile matter content of pretreated air-dried coal, fixed carbon content of pretreated air-dried coal, and calorific value of pretreated bomb.

[0122] The MISM coal received-based elemental carbon content regression calculation model was trained and validated using normalized raw data. The inputs to the MISM coal received-based elemental carbon content regression calculation model included: total moisture content, air-dried basis moisture, air-dried basis total sulfur, air-dried basis ash content, air-dried basis volatile matter, air-dried basis hydrogen content, air-dried basis fixed carbon, and bomb calorific value, totaling eight parameters. The coal received-based elemental carbon content was used to perform error analysis on the output of the MISM coal received-based elemental carbon content regression calculation model.

[0123] Based on the preprocessed measured raw coal quality data, the MISM coal received-based elemental carbon content regression calculation model is used for calculation. The input of the MISM coal received-based elemental carbon content regression calculation model includes: total moisture content, air-dried basis moisture, air-dried basis ash content, air-dried basis volatile matter, air-dried basis fixed carbon, and bomb calorific value, a total of 6 parameters. The air-dried basis total sulfur and air-dried basis hydrogen content are set to default values, which are set to 0.5 in this embodiment. At this time, the MISM coal received-based elemental carbon content regression calculation model outputs the calculation result of the coal received-based elemental carbon content.

[0124] In one embodiment of the present invention, in step S1, the raw coal quality data after analysis and measurement are subjected to maximum and minimum normalization processing, and the calculation method is as follows:

[0125]

[0126] Where, x norm To normalize each parameter in the original data, x i For each parameter in the original data, x imax x represents the maximum value of each parameter in the original data. imin The minimum value of each parameter in the original data is used; after obtaining the normalized original data, the training data and validation data are divided in a 4:1 ratio.

[0127] In one embodiment of the present invention, the data is processed using a regression calculation model of the carbon content of coal received by MISM, such as... Figure 2 The diagram shows the flowchart of the MISM coal received-based elemental carbon content regression calculation model. The specific steps are as follows:

[0128] A1: Using the input data as the input features of the Mamba mapping layer, the Mamba mapping layer is used to perform feature mapping to obtain the first mapped feature, the expression of which is as follows:

[0129] F1 = Mamba(x in-norm )

[0130] Where F1 is the first mapping feature, Mamba() is the Mamba mapping layer function, and x in-norm Input data;

[0131] A2: Add the first mapping feature to the input data to obtain the first summed feature. The calculation method is as follows:

[0132] F2 = F1 + x in-norm

[0133] Among them, F2 is the first summation feature;

[0134] A3: Based on the first summed feature, a linear activation layer is used for integrated activation to obtain the integrated feature, calculated as follows:

[0135] F3 = Linear[LeakyReLU(F2)]

[0136]

[0137] Where F3 is the integrated feature, Linear[LeakyReLU()] is the linear activation layer function, v is the input of the LeakyReLU function, and α is a constant set. In this embodiment, α is set to 0.01.

[0138] A4: Using the integrated features as input features to the Mamba mapping layer, the Mamba mapping layer is used to perform feature mapping to obtain the second mapped features. The calculation method is as follows:

[0139] F4 = Mamba(F3)

[0140] Among them, F4 is the second mapping feature;

[0141] A5: Add the second mapping feature and the first summed feature to obtain the second summed feature. The calculation method is as follows:

[0142] F5 = F4 + F2

[0143] Among them, F5 is the second summation feature;

[0144] A6: Based on the second summation characteristic, the linear activation layer is used for integration activation to obtain the carbon content of the basic element of coal. The calculation method is as follows:

[0145] F6 = Linear[LeakyReLU(F5)]

[0146] F6 represents the carbon content of the basic element of coal.

[0147] In one embodiment of the present invention, the specific implementation methods for obtaining the first mapped feature using the Mamba mapping layer in A1 and obtaining the second mapped feature using the Mamba mapping layer in A4 are the same, such as... Figure 3 The diagram shown is a flowchart of the Mamba mapping layer computation process, and its specific implementation is as follows:

[0148] B1: Obtain the input features of the Mamba mapping layer. In this embodiment, the dimension of the input features is 8.

[0149] B2: Based on the input features, normalize them using the root mean square of the features to obtain the normalized input features. The calculation method is as follows:

[0150]

[0151]

[0152] Where, x RMS For the normalized input features, x in The input features of the Mamba mapping layer, The scaling result of the input features to the Mamba mapping layer, where params is an 8-dimensional trainable parameter matrix, and μ RMS The root mean square of the normalized feature, ε is a set constant, and x iThese are the measured coal quality parameters; in this embodiment, the initial parameters of the trainable parameter matrix are set to 1.0, and ε is set to 10. -5 ;

[0153] B3: Based on the normalized input features, a linear layer is used for processing to obtain the third mapping feature, calculated as follows:

[0154] F6 = Linear(x) RMS )

[0155] Where F6 is the third mapping feature, Linear() is the linear layer function, and the mapping dimension is set to 16;

[0156] B4: Based on the third mapping feature, a 1D convolutional layer is used to process the data to obtain the convolutional feature, calculated as follows:

[0157] F7 = 1DConv(F6)

[0158] Where F7 is the convolutional feature, 1DConv() is the 1D convolutional layer function, the kernel size in the 1D convolutional layer is set to 1×3, and the padding size is set to 1;

[0159] B5: Based on the convolutional features, the first activation feature is obtained by processing it using an activation layer. The calculation method is as follows:

[0160] F8 = SiLU(F7)

[0161]

[0162] Where F8 is the first activation feature, SiLU() is the Sigmoid-gated linear unit activation function, and w is the input of the Sigmoid-gated linear unit activation function;

[0163] B6: Based on the first activation feature, a linear layer is used for processing to obtain the fourth mapping feature, calculated as follows:

[0164] F9 = Linear(F8)

[0165] F9 is the fourth mapping layer, Linear() is the linear layer function, and the mapping dimension is set to 16.

[0166] B7: Using the fourth mapping feature as the input feature of the ISSM mapping layer, the ISSM mapping layer is used to perform mapping to obtain the fifth mapping feature, calculated as follows:

[0167] F 10 =ISSM(F9)

[0168] Among them, F 10This is the fifth mapping layer. ISSM() is the ISSM mapping layer function, and the mapping dimension is set to 16.

[0169] B8: Based on the fifth mapping layer, the activation layer is used for processing to obtain the second activation feature, calculated as follows:

[0170] F 11 =SiLU(F 10 )

[0171] Among them, F 11 As the second activation feature, SiLU() is the activation function of the Sigmoid-gated linear unit;

[0172] B9: Based on the input features, a linear layer is used for processing to obtain the sixth mapping feature, calculated as follows:

[0173] F 12 =Linear(x) in )

[0174] Among them, F 12 For the sixth mapping feature, x in The input features are those of the ISSM mapping layer;

[0175] B10: Based on the sixth mapping feature, the third activation feature is obtained by processing it using the activation layer. The calculation method is as follows:

[0176] F 13 =SiLU(F 12 )

[0177] Among them, F 13 As the third activation feature, SiLU() is the activation function of the Sigmoid-gated linear unit;

[0178] B11: Multiply the third activation feature and the second activation feature to obtain a new feature. The calculation method is as follows:

[0179] F 14 =F 13 ×F 11

[0180] Among them, F 14 For new features;

[0181] B12: Based on the new features, a linear layer is used for processing to obtain the mapping features output by the Mamba mapping layer. The calculation method is as follows:

[0182] F 15 =Linear(F 14 )

[0183] Among them, F 15The Linear() function is used to map the features output by the Mamba mapping layer, with the mapping dimension set to 8.

[0184] In one embodiment of the present invention, mapping is performed using an ISSM mapping layer, such as... Figure 4 The diagram shown is a flowchart of the ISSM mapping layer calculation process, and its specific implementation is as follows:

[0185] C1: Obtain the input features of the ISSM mapping layer and the previous state matrix of the ISSM mapping layer, and calculate the system matrix, control matrix, output matrix, and direct transfer matrix of the ISSM mapping layer respectively.

[0186] C2: Multiply the input features by the control matrix of the ISSM mapping layer to obtain the control features;

[0187] C3: Multiply the previous state matrix of the ISSM mapping layer and the system matrix of the ISSM mapping layer to obtain the state characteristics;

[0188] C4: Add the state features and control features to obtain the current state matrix of the ISSM mapping layer;

[0189] C5: Multiply the output matrix of the ISSM mapping layer with the current state matrix of the ISSM mapping layer to obtain the first feature;

[0190] C6: Multiply the direct transfer matrix of the ISSM mapping layer with the input matrix to obtain the second feature;

[0191] C7: Add the first feature and the second feature together to obtain the fifth mapping feature output by the ISSM mapping layer.

[0192] In C1, the previous state matrix of the ISSM mapping layer is initially a matrix of all zeros, and its dimension is the same as that of the system matrix of the ISSM mapping layer. The system matrix, control matrix, output matrix, and direct transfer matrix of the ISSM mapping layer are the parameter matrices of the ISSM mapping layer, such as... Figure 5 The diagram shown is a flowchart of the calculation process for the parameter matrix of the ISSM mapping layer. Its specific implementation is as follows:

[0193] D1: Obtain the input features F of the ISSM mapping layer;

[0194] D2: Set a matrix a of dimension M×N and initialize it using L2 norm normalization to obtain the initialized matrix. Where M is the number of rows in matrix a, set to 8, and N is the number of columns in matrix a, set to 128, calculated as follows:

[0195]

[0196] in, Let be the initialized matrix, a be a matrix of dimension 16×N, and ‖ ‖2 be the L2 norm normalization;

[0197] D3: Based on the input features of the ISSM mapping layer, four different linear layers are used for processing to obtain the seventh, eighth, and ninth mapping features, as well as the output matrix C of the ISSM mapping layer. The calculation method is as follows:

[0198] C = Linear4(F)

[0199] F S7 =Linear3(F)

[0200] F S6 =Linear2(F)

[0201] F S5 =Linear1(F)

[0202] Where C is the output matrix of the ISSM mapping layer, F S7 For the ninth mapping feature, F S6 For the eighth mapping feature, F S5 The seventh mapping feature is F, which is the input feature of the ISSM mapping layer. Linear4, Linear3, Linear2, and Linear1 are four different linear layers.

[0203] D4: Based on the seventh mapping feature, the fourth activation feature is obtained by processing it using the activation layer. The calculation method is as follows:

[0204] F S8 =SiLU(F S5 )

[0205] Among them, F S8 The fourth activation feature is SiLU(), which is the activation function of the Sigmoid-gated linear unit.

[0206] D5: Multiply the fourth activation feature by the eighth mapping feature to obtain the control matrix B of the ISSM mapping layer, calculated as follows:

[0207] B = F S8 ×F S6

[0208] Where B is the control matrix of the ISSM mapping layer;

[0209] D6: Multiply the fourth activation feature by the initialized matrix a to obtain the system features, calculated as follows:

[0210]

[0211] Among them, F S9 System characteristics;

[0212] D7: Perform an exponential transformation on the system characteristics to obtain the system matrix A of the ISSM mapping layer. The calculation method is as follows:

[0213]

[0214] Where A is the system matrix of the ISSM mapping layer;

[0215] In one embodiment of the present invention, the MISM coal-based received elemental carbon content regression calculation model is used to perform regression calculation to obtain the normalized coal-based received elemental carbon content. The obtained normalized coal-based received elemental carbon content is then denormalized to obtain the coal-based received elemental carbon content. The calculation method is as follows:

[0216] C regression =C regression-norm ×(C imax -C imin )+C imin

[0217] Among them, C regression The carbon content of coal is the basic element. regression-norm To normalize the carbon content of coal based on the fundamental element, C imin To determine the minimum carbon content of the basic element received by the coal, C imax This represents the maximum measured carbon content of the basic element in coal.

[0218] In one embodiment of the present invention, the specific steps of S3 are as follows:

[0219] S301: Using the validation data as input data, the predicted normalized carbon content of coal received by the base element is calculated by the trained MISM coal received base element carbon content regression calculation model.

[0220] S302: Perform an inverse normalization operation on the predicted normalized carbon content of coal received by the base element to obtain the predicted carbon content of coal received by the base element.

[0221] S303: The average relative error between the calculated and predicted carbon content of coal received on the basis and the carbon content of coal received on the basis in the original coal quality data;

[0222] S304: Determine if the average relative error is less than or equal to the set threshold. If yes, obtain the certified MISM coal-fired carbon content regression calculation model. Otherwise, the accuracy of the MISM coal-fired carbon content regression calculation model is insufficient. Optimize the MISM coal-fired carbon content regression calculation model and return to S2. The set threshold is 0.02. When the average relative error is greater than the set threshold, increase the mapping dimension in the linear activation layer of the MISM coal-fired carbon content regression calculation model. When the initial mapping dimension of the MISM coal-fired carbon content regression calculation model is 128, optimize its model structure by increasing the mapping dimension by 32 each time. It is recommended that the maximum mapping dimension should not exceed 512 to prevent the model from becoming too large and affecting computational efficiency and accuracy.

[0223] The beneficial effects of this invention are as follows: This invention designs an ISSM mapping layer model and incorporates a Mamba module to build a MISM coal-based carbon content regression calculation model. This model has the ability to capture important features and can establish the relationship between measured data and coal-based carbon content by adjusting network structure parameters. This allows the coal-based carbon content to be obtained through regression calculation using only partial measured data, thereby reducing the detection cost of coal-based carbon content and reducing the measured parameters required for coal quality analysis for thermal power plants. At the same time, the regression calculation results ensure the accuracy of coal-based carbon content.

Claims

1. A regression calculation method for the carbon content of coal as a basic element, characterized in that, Includes the following steps: S1: Obtain the raw coal quality data after analysis and measurement, perform maximum and minimum normalization processing to obtain normalized raw data, and divide the normalized raw data into training data and validation data; S2: Using the training data as input data, the MISM coal received elemental carbon content regression calculation model is used to process the data to obtain the trained MISM coal received elemental carbon content regression calculation model. S3: Using the validation data as input data, the trained MISM coal received elemental carbon content regression calculation model is used to process the data to obtain the certified MISM coal received elemental carbon content regression calculation model. S4: Obtain the raw measured coal quality data and perform preprocessing operations to obtain preprocessed raw measured coal quality data; S5: The pre-processed measured raw coal quality data is used as input data and processed using the certified MISM coal received-based elemental carbon content regression calculation model to obtain the coal received-based elemental carbon content. The specific steps for processing using the MISM coal-based elemental carbon content regression calculation model are as follows: A1: Use the input data as the input features of the Mamba mapping layer, and use the Mamba mapping layer to perform feature mapping to obtain the first mapped features; A2: Add the first mapping feature to the input data to obtain the first summed feature; A3: Based on the first summation feature, integrate activation is performed using a linear activation layer to obtain the integrated feature; A4: Use the integrated features as input features of the Mamba mapping layer, and use the Mamba mapping layer to perform feature mapping to obtain the second mapped features; A5: Add the second mapping feature and the first summation feature to obtain the second summation feature; A6: Based on the second summation feature, the linear activation layer is used for integration activation to obtain the carbon content of the basic element of coal. The mathematical expression for the regression calculation model of the basic element carbon content of MISM coal is as follows: in, The carbon content of the basic element of coal. For the second additive feature, For the second mapping feature, To integrate features, As the first additive feature, For the first mapping feature, For input data, For linear activation layer functions, For Mamba mapping layer functions, for The input of the function, This is a constant that is set.

2. The regression calculation method for the carbon content of coal as a basic element according to claim 1, characterized in that, The raw coal quality data after analysis and measurement include: total moisture content, air-dried moisture content, air-dried total sulfur content, air-dried ash content, air-dried volatile matter content, air-dried hydrogen content, air-dried fixed carbon content, bomb calorific value, and coal-as-received elemental carbon content. The normalized raw data includes: normalized total moisture content, normalized air-dried basis moisture, normalized air-dried basis total sulfur, normalized air-dried basis ash, normalized air-dried basis volatile matter, normalized air-dried basis hydrogen content, normalized air-dried basis fixed carbon, normalized bomb calorific value, and normalized coal-as-it-as-it-based elemental carbon content. The measured raw coal quality data of the pretreatment include: total moisture content of the pretreatment, moisture content of the pretreatment air-dried basis, ash content of the pretreatment air-dried basis, volatile matter of the pretreatment air-dried basis, fixed carbon of the pretreatment air-dried basis, and calorific value of the pretreatment bomb.

3. The regression calculation method for the carbon content of coal as a basic element according to claim 1, characterized in that, The Mamba mapping layer performs feature mapping, and its specific implementation is as follows: B1: Obtain the input features of the Mamba mapping layer; B2: Based on the input features, normalize them using the root mean square of the features to obtain normalized input features; B3: Based on the normalized input features, a linear layer is used for processing to obtain the third mapping feature; B4: Based on the third mapping feature, process it using a 1D convolutional layer to obtain the convolutional feature; B5: Based on the convolutional features, the activation layer is used to process them to obtain the first activation feature; B6: Based on the first activation feature, the fourth mapping feature is obtained by processing it using a linear layer; B7: Use the fourth mapping feature as the input feature of the ISSM mapping layer, and use the ISSM mapping layer to perform mapping to obtain the fifth mapping feature; B8: Based on the fifth mapping layer, the activation layer is used for processing to obtain the second activation feature; B9: Based on the input features, a linear layer is used for processing to obtain the sixth mapping feature; B10: Based on the sixth mapping feature, the third activation feature is obtained by processing it using the activation layer; B11: Multiply the third activation feature and the second activation feature to obtain a new feature; B12: Based on the new features, a linear layer is used for processing to obtain the mapping features output by the Mamba mapping layer.

4. The regression calculation method for the carbon content of coal as a basic element according to claim 3, characterized in that, The mathematical expression for the Mamba mapping layer is as follows: in, The mapping features output by the Mamba mapping layer. As a new feature, This is the third activation feature. The sixth mapping feature, This is the second activation feature. This is the fifth mapping layer. This is the fourth mapping layer. As the first activation feature, For convolutional features, For the third mapping feature, For normalized input features, It is the activation function of the Sigmoid-gated linear unit. For linear layer functions, For ISSM mapping layer functions, For 1D convolutional layer functions, The input features of the Mamba mapping layer, The scaled result of the input features to the Mamba mapping layer, The training parameter matrix has a dimension of 8. The root mean square of the normalized feature. A constant set, These are the measured coal quality parameters. This is the input to the Sigmoid-gated linear unit activation function.

5. The regression calculation method for the carbon content of coal as a basic element according to claim 3, characterized in that, The ISSM mapping layer performs mapping, and its specific implementation is as follows: C1: Obtain the input features of the ISSM mapping layer and the previous state matrix of the ISSM mapping layer, and calculate the system matrix, control matrix, output matrix, and direct transfer matrix of the ISSM mapping layer respectively. C2: Multiply the input features by the control matrix of the ISSM mapping layer to obtain the control features; C3: Multiply the previous state matrix of the ISSM mapping layer and the system matrix of the ISSM mapping layer to obtain the state characteristics; C4: Add the state features and control features to obtain the current state matrix of the ISSM mapping layer; C5: Multiply the output matrix of the ISSM mapping layer with the current state matrix of the ISSM mapping layer to obtain the first feature; C6: Multiply the direct transfer matrix of the ISSM mapping layer with the input matrix to obtain the second feature; C7: Add the first feature and the second feature together to obtain the fifth mapping feature output by the ISSM mapping layer.

6. The regression calculation method for the carbon content of coal as a basic element according to claim 5, characterized in that, The calculation methods for the system matrix, control matrix, output matrix, and direct transfer matrix of the ISSM mapping layer in C1 are as follows: D1: Obtain the input features of the ISSM mapping layer; D2: Set up a matrix of dimension M×N and initialize it using L2 norm normalization to obtain an initialized matrix, where M is the number of rows of the matrix and N is the number of columns of the matrix; D3: Based on the input features of the ISSM mapping layer, four different linear layers are used for processing to obtain the seventh mapping feature, the eighth mapping feature, the ninth mapping feature, and the output matrix of the ISSM mapping layer. D4: Based on the seventh mapping feature, the fourth activation feature is obtained by processing it using the activation layer; D5: Multiply the fourth activation feature by the eighth mapping feature to obtain the control matrix of the ISSM mapping layer; D6: Multiply the fourth activation feature by the initialized matrix to obtain the system features; D7: Perform an exponential transformation on the system characteristics to obtain the system matrix of the ISSM mapping layer; D8: Based on the ninth mapping feature, activate the activation layer to obtain the direct transfer matrix of the ISSM mapping layer.

7. The regression calculation method for the carbon content of coal as a basic element according to claim 6, characterized in that, The mathematical expressions for the system matrix, control matrix, output matrix, and direct transfer matrix of the ISSM mapping layer are as follows: in, The system matrix of the ISSM mapping layer. This is the control matrix for the ISSM mapping layer. This is the output matrix of the ISSM mapping layer. For the direct transfer matrix of the ISSM mapping layer, The input features of the ISSM mapping layer, As a system characteristic, This is the fourth activation feature. The ninth mapping feature, This is the eighth mapping feature. This is the seventh mapping feature. , , as well as For four different linear layers, It is the activation function of the Sigmoid-gated linear unit. For the initialized matrix, It is a matrix of dimension 16×N. Normalize for L2 norm.

8. The regression calculation method for the carbon content of coal as a basic element according to claim 1, characterized in that, The specific steps of S3 are as follows: S301: Using the validation data as input data, the predicted normalized carbon content of coal received by the base element is calculated by the trained MISM coal received base element carbon content regression calculation model. S302: Perform an inverse normalization operation on the predicted normalized carbon content of coal received by the base element to obtain the predicted carbon content of coal received by the base element. S303: The average relative error between the calculated and predicted carbon content of coal received on the basis and the carbon content of coal received on the basis in the original coal quality data; S304: Determine whether the average relative error is less than or equal to the set threshold. If yes, obtain the certified MISM coal-fired received element carbon content regression calculation model. Otherwise, optimize the MISM coal-fired received element carbon content regression calculation model and return to S2.

Images