Smoke concentration prediction method, device and medium

The prediction of smoke concentration through BP neural network is solved, and the measurement abnormality caused by sensor aging in the smoke monitoring system of coal-fired power plants is simplified, and the accounting and verification process is realized, which is accurate measurement and efficient monitoring of smoke emissions.

CN118335241BActive Publication Date: 2024-12-03湖南工商大学
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410429537.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-10
Publication Date
2024-12-03
Estimated Expiration
2044-04-10

AI Technical Summary

Technical Problem

The sensors in the smoke monitoring system of existing coal-fired power plants are prone to consumption or aging, resulting in abnormal measurements, affecting the accuracy of smoke emissions and the ecological environment, and the manual accounting and verification are complex, which increases the work burden.

Method used

The method of predicting smoke and dust concentration by using BP neural network is used to clean and normalize the coal quality and quantity data of coal entering the furnace, and build a prediction model, and use Adam optimizer to train the model to generate smoke and dust emission rules, simplify the calculation of coal-fired power plants. Verification process.

Benefits of technology

Accurate measurement of smoke and dust emissions is achieved, manual workload is reduced, measurement accuracy and efficiency is improved, and a reliable reference basis for CEMS smoke and dust monitoring data is provided.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118335241B_ABST
    Figure CN118335241B_ABST
Patent Text Reader

Abstract

The present invention discloses a smoke concentration prediction method, device and medium. The smoke emission of each batch of coal entering the furnace is calculated according to the hourly coal consumption of the unit as the label value of the model. The general law between the coal entering the furnace data and its corresponding smoke emission can be generated by training the model, and the relationship between the material and the smoke emission can be accurately identified, thereby reducing the workload of manual calculation and verification, and providing a reference basis for CEMS smoke monitoring data. At the same time, the Adam algorithm is used to optimize the BP neural network, which can automatically adjust the learning rate of each parameter and train the model quickly and efficiently. The present invention can solve the problems of measurement errors and complex manual calculation and verification in the prior art, thereby achieving the purpose of accurately measuring smoke emissions from coal-fired power plants and reducing manual workload.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of emission control technology, and more specifically, to a smoke concentration prediction method, device and medium. Background Art

[0002] A large amount of smoke is generated during the power generation process of coal-fired power plants, which affects the safe and stable operation of the units and causes serious damage to the ecological environment. Therefore, the country has introduced a series of ultra-low emission policies for coal-fired power plants and has carried out strict accounting and verification of smoke emissions. When CEMS continuously monitors smoke, due to the drastic and highly random changes in smoke concentration, it is necessary to use highly sensitive and high-speed response sensors to accurately measure the smoke concentration. However, in actual applications, the components in CEMS are prone to consumption or aging, which can cause measurement abnormalities and other problems. Summary of the invention

[0003] The present invention is provided to solve the above problems existing in the prior art. Therefore, a smoke concentration prediction method, device and medium are needed, which are intended to quickly and effectively update the neural network weights and threshold parameters, generate general rules between material data and smoke emissions, serve as a reference for CEMS smoke monitoring data, and simplify the accounting and verification process of coal-fired power plants to reduce the manual burden.

[0004] According to a first aspect of the present invention, a method for predicting smoke concentration is provided, the method comprising:

[0005] Cleaning the original data, including the coal quality and quantity data of the coal entering the furnace and the hourly emission data of smoke dust;

[0006] According to the hourly coal consumption of the unit, the smoke dust emission corresponding to each batch of coal entering the furnace is calculated in turn, and the coal quality data, quantity data and smoke dust emission of each batch of coal entering the furnace are normalized to obtain a data set;

[0007] A prediction model is constructed, and the prediction model is trained using the data set to achieve smoke concentration prediction with the trained prediction model.

[0008] Furthermore, the coal quality of the coal entering the furnace includes data on six properties: dry basis higher calorific value, volatile matter, air-dried coal moisture, air-dried coal ash, dry basis total sulfur, and analytical hydrogen. The quantity data is the quantity of the coal entering the furnace corresponding to the coal quality.

[0009] Furthermore, the cleaning of the original data specifically includes:

[0010] When there are missing values ​​in the original data, if consecutive values ​​are missing, they are directly deleted. If a single data is missing, the missing value is interpolated by the nearest neighbor.

[0011] Furthermore, according to the hourly coal consumption of the unit, the smoke emission corresponding to each batch of coal entering the furnace is calculated in turn, including:

[0012] Calculate the hourly coal consumption of the unit based on the actual daily operating time and daily coal consumption of the unit;

[0013] Position the coal index at the first batch of coal entering the furnace, position the smoke monitoring value index at the first smoke monitoring value, and obtain the current coal quantity, which is the quantity of the first batch of coal entering the furnace obtained when the smoke emission corresponding to a batch of coal entering the furnace is 0;

[0014] If the current coal volume is greater than the hourly coal consumption, the smoke emission volume is added to the smoke monitoring value corresponding to the smoke monitoring value index, the smoke monitoring value index is increased by 1, and the current coal volume is subtracted from the hourly coal consumption;

[0015] If the current coal quantity is less than the hourly coal consumption, the smoke dust emission will be added to the current coal quantity / hourly coal consumption*smoke dust monitoring value and stored. After storage, it will be updated again to (hourly coal consumption-current coal quantity) / hourly coal consumption*smoke dust monitoring value. The smoke dust monitoring value index will be increased by 1. The current coal quantity will be updated to the quantity of coal entering the furnace corresponding to the coal index + the current coal quantity-hourly coal consumption. The coal index entering the furnace will be increased by 1.

[0016] Furthermore, according to the hourly coal consumption of the unit, the smoke dust emission corresponding to each batch of coal entering the furnace is calculated in sequence, including:

[0017] The monitoring time for obtaining smoke emission data;

[0018] When the running time is greater than the monitoring time:

[0019] In the case of unit shutdown, after calculating the coal consumption and smoke emissions of the day, the current coal quantity minus the fraction of the operating time multiplied by the hourly coal consumption is used as the updated current coal quantity;

[0020] When the unit is started, before calculating the coal consumption and smoke emission of the day, the current coal quantity minus the fraction of the running time multiplied by the hourly coal consumption is used as the updated current coal quantity;

[0021] When the running time is less than the monitoring time:

[0022] In the case of a unit shutdown, when processing the last monitoring hour of the day, the current coal quantity minus the fraction of the operating time multiplied by the hourly coal consumption is used as the updated current coal quantity.

[0023] When the unit is started, when processing the first monitoring hour of the day, the current coal quantity minus the decimal part of the operating time multiplied by the hourly coal consumption is used as the updated current coal quantity.

[0024] Furthermore, the prediction model is a BP neural network, which includes an input layer, at least one hidden layer and an output layer. The weight initialization method of the BP neural network adopts Glorot uniform distribution initialization, the threshold is initialized to 0, and the Relu function is selected as the activation function of the BP neural network.

[0025] Furthermore, the using the data set to train the prediction model specifically includes:

[0026] Splitting the data set into a training data set and a test data set according to a set ratio;

[0027] Input the training data set into the BP neural network, and calculate the input and output of the hidden layer and the output layer through forward propagation;

[0028] The error of the output layer prediction result is calculated according to the mse calculation formula, which is expressed as:

[0029]

[0030] Where mse represents mean square error, y is the true value, is the predicted value, and n is the total number of samples.

[0031] The error is back-propagated, and based on the Adam optimizer, the model parameters are optimized through the following formula to adjust the connection weights and thresholds between the hidden layer and the output layer:

[0032] m t =β1·m t-1 +(1-β1)·dk

[0033] v t =β2·v t-1 +(1-β2)·dk 2

[0034]

[0035]

[0036] Where m t-1 ,v t-1 is the first-order moment estimate before iteration, m t ,v t is the first-order moment estimate after iteration, β1, β2 are exponentially weighted average parameters, dk represents the gradient of weight w and threshold b, α is the learning rate, ε is a constant to prevent the denominator from being 0, w t-1 ,b t-1 is the weight and threshold to be trained, w t ,b tare the trained weights and thresholds;

[0037] Repeat the steps from forward propagation to back propagation until the preset number of training times is reached;

[0038] The trained BP neural network is evaluated using the test data set.

[0039] Furthermore, the use of the test data set to evaluate the trained BP neural network specifically includes:

[0040] After the model training is completed, the test data set is input, and the predicted value is obtained through forward propagation calculation. The mean square error MSE, root mean square error RMSE, mean absolute error MAE and determination coefficient R2 are used to evaluate the prediction results. The calculation formula is as follows:

[0041]

[0042]

[0043]

[0044]

[0045] Among them, SSR represents the regression sum of squares, and SST represents the total deviation sum of squares.

[0046] According to a second aspect of the present invention, there is provided a smoke concentration prediction device, the device comprising:

[0047] A cleaning module is configured to clean the raw data, wherein the raw data includes coal quality and quantity data of coal entering the furnace and hourly smoke emission data;

[0048] The calculation module is configured to calculate the smoke dust emission corresponding to each batch of coal entering the furnace according to the hourly coal consumption of the unit, and normalize the coal quality data, quantity data and smoke dust emission of each batch of coal entering the furnace to obtain a data set;

[0049] The training module is configured to construct a prediction model, train the prediction model using the data set, and realize smoke concentration prediction with the trained prediction model.

[0050] According to a third aspect of the present invention, there is provided a readable storage medium storing one or more programs, wherein the one or more programs can be executed by one or more processors to implement the method as described above.

[0051] The present invention has at least the following beneficial effects:

[0052] (1) The present invention can solve the problems of measurement errors and complicated manual calculation and verification in the prior art, thereby achieving the purpose of accurately measuring smoke emissions from coal-fired power plants and reducing manual workload.

[0053] (2) The present invention calculates the smoke emission of each batch of coal entering the furnace according to the hourly coal consumption of the unit as the label value of the model. By training the model, the general law between the coal entering the furnace data and its corresponding smoke emission can be generated, and the relationship between the material and the smoke emission can be accurately identified, which reduces the workload of manual calculation and verification, and provides a reference basis for CEMS smoke monitoring data. At the same time, the Adam algorithm is used to optimize the BP neural network, which can automatically adjust the learning rate of each parameter and train the model quickly and efficiently. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A flow chart of a smoke concentration prediction method according to an embodiment of the present invention is shown.

[0055] Figure 2 A flowchart of smoke emission calculation according to an embodiment of the present invention is shown.

[0056] Figure 3 A schematic diagram of smoke emission prediction results according to an embodiment of the present invention is shown.

[0057] Figure 4 A structural diagram of a smoke concentration prediction device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and specific embodiments, but are not intended to limit the present invention. For the various steps described herein, if there is no necessity for a causal relationship between each other, the order in which they are described as examples herein should not be regarded as a limitation, and those skilled in the art should know that they can be adjusted in order, as long as the logic between them is not destroyed, resulting in the inability to implement the entire process.

[0059] The embodiment of the present invention provides a smoke concentration prediction method, specifically a smoke prediction method for a coal-fired power plant based on an Adam algorithm to optimize a BP neural network. Figure 1 As shown, the method comprises the following steps:

[0060] Step 1: Raw data preprocessing: Clean the coal quality and quantity data of the coal entering the coal-fired power plant and the hourly smoke emission data monitored by CEMS.

[0061] It should be noted that the original data in step 1 includes two parts. The first part includes data on six properties of the coal quality data of the incoming coal, including dry basis high calorific value, volatile matter, air-dried coal moisture, air-dried coal ash, dry basis total sulfur, and analytical hydrogen, as well as the quantity data of the incoming coal corresponding to the coal quality data of this batch. The second part is the hourly emission data of smoke monitored by CEMS. For the missing values ​​of the original data, the continuous missing values ​​are directly deleted. For the case of single missing data, the nearest neighbor interpolation method is used to process the missing values.

[0062] Step 2: Calculation of smoke emissions corresponding to each batch of coal entering the furnace: According to the hourly coal consumption of the unit, the smoke emissions corresponding to each batch of coal entering the furnace are calculated in turn. The coal quality data and its corresponding smoke emissions data are normalized and used as the input data of the model.

[0063] In a specific embodiment, step 2 first calculates the hourly coal consumption of the unit according to the actual daily operation time and daily coal consumption of the unit, and obtains the daily monitoring time of the CEMS system. Secondly, the calculation is initialized, the coal index is positioned at the first batch of coal entering the furnace, the smoke monitoring value index is positioned at the first smoke monitoring value, and the smoke emission corresponding to the batch of coal entering the furnace is set to 0, and the number of the first batch of coal entering the furnace is obtained, which is recorded as the current coal quantity. Third, if the current coal quantity is greater than the hourly coal consumption, the smoke emission is added to the smoke monitoring value corresponding to the smoke monitoring value index, the smoke monitoring value index is increased by 1, and the current coal quantity is subtracted from the hourly coal consumption; if the current coal quantity is less than the hourly coal consumption, the smoke emission is added to the current coal quantity / hourly coal consumption*smoke monitoring value and stored, and after storage, it is updated again to (hourly coal consumption-current coal quantity) / hourly coal consumption*smoke monitoring value, and the smoke monitoring value index is increased by 1. The current coal quantity is updated to the number of coal entering the furnace corresponding to the coal index + the current coal quantity-hourly coal consumption, and the coal index is increased by 1. Repeat the above steps until the daily coal consumption data table is traversed.

[0064] For example, the specific calculation process of step 2 can be found in Figure 2 shown.

[0065] In a specific embodiment, the coal quality data, quantity and smoke emission of each batch of coal entering the furnace are normalized to obtain an enhanced data set (X, Y), where X is the coal quality index and quantity data of the coal entering the furnace, which are used as the feature value of the prediction model, and Y is the smoke emission corresponding to the batch of coal entering the furnace, which is used as the label value of the prediction model. The data set is represented as follows:

[0066]

[0067] Step 3: BPNN prediction: Split the input data into training data and test data, determine the network structure and train it, predict the test data, and obtain the test error and prediction results.

[0068] In a specific embodiment, in step 3, the data set is split into a training data set (X_train, Y_train) and a test data set (X_test, Y_test) in a ratio of 8:2, and a BP neural network structure of an input layer, a first hidden layer, a second hidden layer, and an output layer is constructed. The weight initialization method uses Glorot uniform distribution initialization, the threshold is initialized to 0, and the Relu function is selected as the activation function of the BP neural network. The function expression is as follows:

[0069] y=max(0,x)

[0070] Among them, y represents output, x represents input, and max represents the maximum value function.

[0071] Use the training data set to train the BP neural network. Input the training data set into the network, calculate the input and output of each unit in the hidden layer and the output layer through forward propagation; calculate the error of the output layer prediction result according to the mse calculation formula; back-propagate the error, use the Adam optimizer to optimize the model parameters, adjust the connection weights and thresholds between the hidden layer and the output layer; repeat the steps from forward propagation to back-propagation until the preset number of training times is reached.

[0072] The mse calculation formula is as follows:

[0073]

[0074] Where y is the true value, is the predicted value.

[0075] For the weight w t and threshold b t , the calculation formula of Adam optimizer is as follows:

[0076]

[0077]

[0078]

[0079] Where m t-1 ,v t-1 is the first-order moment estimate before iteration, m t ,v t is the first-order moment estimate after iteration, β1, β2 are exponentially weighted average parameters, dk represents the gradient of weight w and threshold b, α is the learning rate, ε is a decimal to prevent the denominator from being 0, w t-1 ,b t-1 is the weight and threshold to be trained, w t ,b tare the trained weights and thresholds.

[0080] Exemplarily, the Adam algorithm is used to update the weights and thresholds of the model, which can calculate the individual adaptive learning rates of different parameters based on the first-order moment and second-order moment of the gradient. The calculation steps are as follows:

[0081] Compute the first moment estimate, m t =β1·m t-1 +(1-β1)·dk, where m t-1 、m t The first-order moment estimate before and after the iteration, β1 is the exponentially weighted average parameter, and dk represents the gradient of the weight w and the threshold b.

[0082] Compute the second moment estimate, v t =β2·v t-1 +(1-β2)·dk 2 , where v t-1 、v t The first-order moment estimate before and after the iteration, β2 is the exponentially weighted average parameter, dk 2 Represents the squared gradient of the weight w and the threshold b.

[0083] The Adam algorithm combines the adaptive learning rate optimization algorithm (Adagrad) and the root mean square transfer algorithm (RMSProp). Compared with the traditional gradient descent method, the Adam algorithm can better adapt to the update speed of different parameters by adaptively adjusting the learning rate, thereby accelerating the convergence speed. In addition, the Adam algorithm has good robustness and can adapt to different types of training data. Compared with other optimization algorithms, such as Adagrad and RMSProp, the Adam optimizer performs better when dealing with non-stationary objective functions.

[0084] Combine the first-order moment estimation and the second-order moment estimation to update the weights and thresholds. where w t 、b t are the adjusted weights and thresholds, w t-1 、b t-1 The weights and thresholds to be trained, α is the learning rate, and ε is a constant added for numerical stability.

[0085] After the network training is completed, input the test data set X test , the predicted value is obtained by forward propagation calculation, and the mean square error MSE, root mean square error RMSE, mean absolute error MAE and R 2 To evaluate the prediction results. The calculation formula is as follows:

[0086]

[0087]

[0088]

[0089]

[0090] Among them, SSR represents the regression sum of squares, and SST represents the total sum of squares.

[0091] The prediction results are shown in Table 1:

[0092] Table 1 Smoke prediction evaluation index

[0093] MSE RMSE MAE <![CDATA[R 2 ]]> 53.653274 7.324840 6.145478 0.653019

[0094] As can be seen from Table 1, through the comprehensive evaluation of multiple regression evaluation indicators, the prediction method has high accuracy, can identify the general law between the incoming coal and the smoke emission, can be used in the actual production process of coal-fired power plants, and provide a reference for smoke monitoring data in view of the problem that CEMS is prone to errors. Using this invention method, the coal quality and quantity data of the incoming coal are measured and input in real time, which can identify whether the CEMS is abnormal and reflect the quality of the incoming coal detection data to a certain extent, thereby reducing the workload of manual calculation and verification.

[0095] Based on the trained model, the prediction results of smoke emissions are as follows: Figure 3 As shown, according to Figure 3 It is shown that the present invention has a higher accuracy in predicting smoke emission.

[0096] The embodiment of the present invention also provides a smoke concentration prediction device, such as Figure 4 As shown, the device 400 includes:

[0097] A cleaning module 401 is configured to clean the raw data, wherein the raw data includes the coal quality and quantity data of the coal entering the furnace and the hourly emission data of smoke dust;

[0098] The calculation module 402 is configured to calculate the smoke dust emission corresponding to each batch of coal fed into the furnace according to the hourly coal consumption of the unit, and normalize the coal quality data, quantity data and smoke dust emission of each batch of coal fed into the furnace to obtain a data set;

[0099] The training module 403 is configured to construct a prediction model, and train the prediction model using the data set, so as to achieve smoke concentration prediction with the trained prediction model.

[0100] In some embodiments, the coal quality of the coal entering the furnace includes data on six properties: dry basis high calorific value, volatile matter, air-dried coal moisture, air-dried coal ash, dry basis total sulfur, and analytical hydrogen. The quantity data is the quantity of coal entering the furnace corresponding to the coal quality.

[0101] In some embodiments, the cleaning module is further configured to:

[0102] When there are missing values ​​in the original data, if consecutive values ​​are missing, they are directly deleted. If a single data is missing, the missing value is interpolated by the nearest neighbor.

[0103] In some embodiments, the computing module is further configured to:

[0104] Calculate the hourly coal consumption of the unit based on the actual daily operating time and daily coal consumption of the unit;

[0105] Position the coal index at the first batch of coal entering the furnace, position the smoke monitoring value index at the first smoke monitoring value, and obtain the current coal quantity, which is the quantity of the first batch of coal entering the furnace obtained when the smoke emission corresponding to a batch of coal entering the furnace is 0;

[0106] If the current coal volume is greater than the hourly coal consumption, the smoke emission volume is added to the smoke monitoring value corresponding to the smoke monitoring value index, the smoke monitoring value index is increased by 1, and the current coal volume is subtracted from the hourly coal consumption;

[0107] If the current coal quantity is less than the hourly coal consumption, the smoke dust emission will be added to the current coal quantity / hourly coal consumption*smoke dust monitoring value and stored. After storage, it will be updated again to (hourly coal consumption-current coal quantity) / hourly coal consumption*smoke dust monitoring value. The smoke dust monitoring value index will be increased by 1. The current coal quantity will be updated to the quantity of coal entering the furnace corresponding to the coal index + the current coal quantity-hourly coal consumption. The coal index entering the furnace will be increased by 1.

[0108] In some embodiments, the computing module is further configured to:

[0109] The monitoring time for obtaining smoke emission data;

[0110] When the running time is greater than the monitoring time:

[0111] In the case of unit shutdown, after calculating the coal consumption and smoke emissions of the day, the current coal quantity minus the fraction of the operating time multiplied by the hourly coal consumption is used as the updated current coal quantity;

[0112] When the unit is started, before calculating the coal consumption and smoke emission of the day, the current coal quantity minus the fraction of the running time multiplied by the hourly coal consumption is used as the updated current coal quantity;

[0113] When the running time is less than the monitoring time:

[0114] In the case of a unit shutdown, when processing the last monitoring hour of the day, the current coal quantity minus the fraction of the operating time multiplied by the hourly coal consumption is used as the updated current coal quantity.

[0115] When the unit is started, when processing the first monitoring hour of the day, the current coal quantity minus the decimal part of the operating time multiplied by the hourly coal consumption is used as the updated current coal quantity.

[0116] In some embodiments, the prediction model is a BP neural network, which includes an input layer, at least one hidden layer and an output layer. The weight initialization method of the BP neural network adopts Glorot uniform distribution initialization, the threshold is initialized to 0, and the Relu function is selected as the activation function of the BP neural network.

[0117] In some embodiments, the training module is further configured to:

[0118] Splitting the data set into a training data set and a test data set according to a set ratio;

[0119] Input the training data set into the BP neural network, and calculate the input and output of the hidden layer and the output layer through forward propagation;

[0120] The error of the output layer prediction result is calculated according to the mse calculation formula, which is expressed as:

[0121]

[0122] Where mse represents mean square error, y is the true value, is the predicted value, and n is the total number of samples.

[0123] The error is back-propagated, and based on the Adam optimizer, the model parameters are optimized through the following formula to adjust the connection weights and thresholds between the hidden layer and the output layer:

[0124] m t =β1·m t-1 +(1-β1)·dk

[0125] v t =β2·v t-1 +(1-β2)·dk 2

[0126]

[0127]

[0128] Where m t-1 ,vt-1 is the first-order moment estimate before iteration, m t ,v t is the first-order moment estimate after iteration, β1, β2 are exponentially weighted average parameters, dk represents the gradient of weight w and threshold b, α is the learning rate, ε is a constant to prevent the denominator from being 0, w t-1 ,b t-1 is the weight and threshold to be trained, w t ,b t are the trained weights and thresholds;

[0129] Repeat the steps from forward propagation to back propagation until the preset number of training times is reached;

[0130] The trained BP neural network is evaluated using the test data set.

[0131] In some embodiments, the training module is further configured to:

[0132] After the model training is completed, the test data set is input, and the predicted value is obtained through forward propagation calculation. The mean square error MSE, root mean square error RMSE, mean absolute error MAE and determination coefficient R2 are used to evaluate the prediction results. The calculation formula is as follows:

[0133]

[0134]

[0135]

[0136]

[0137] Among them, SSR represents the regression sum of squares, and SST represents the total sum of squares.

[0138] It should be noted that the various device structures described in this embodiment belong to the same technical concept as the previously described method, and achieve the same technical effect through the same principle, which will not be repeated here.

[0139] An embodiment of the present invention further provides a readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the method described in any of the above embodiments.

[0140] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present invention with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. The elements in the claims will be interpreted broadly based on the language adopted in the claims, and are not limited to the examples described in this specification or during the implementation of this application, and the examples will be interpreted as non-exclusive. Therefore, this specification and examples are intended to be considered as examples only, and the true scope and spirit are indicated by the following claims and the full scope of their equivalents.

[0141] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art can use other embodiments when reading the above description. In addition, in the above-mentioned specific embodiments, various features can be grouped together to simplify the present invention. This should not be interpreted as a feature of an invention that is not claimed for protection being necessary for any claim. On the contrary, the subject matter of the present invention may be less than all the features of the embodiments of a specific invention. Thus, the following claims are incorporated into the specific embodiments as examples or embodiments, wherein each claim is independently used as a separate embodiment, and it is considered that these embodiments can be combined with each other in various combinations or arrangements. The scope of the present invention should be determined with reference to the full scope of the equivalent forms of the attached claims and these claims.

Claims

1. A smoke concentration prediction method, characterized in that: The method comprises: Cleaning the original data, including the coal quality and quantity data of the coal entering the furnace and the hourly emission data of smoke dust; According to the hourly coal consumption of the unit, the smoke dust emission corresponding to each batch of coal entering the furnace is calculated in turn, and the coal quality data, quantity data and smoke dust emission of each batch of coal entering the furnace are normalized to obtain a data set; Constructing a prediction model, and using the data set to train the prediction model, so as to achieve smoke concentration prediction with the trained prediction model; The prediction model is a BP neural network, which includes an input layer, at least one hidden layer and an output layer. The weight initialization method of the BP neural network adopts Glorot uniform distribution initialization, the threshold is initialized to 0, and the Relu function is selected as the activation function of the BP neural network; The using the data set to train the prediction model specifically includes: Splitting the data set into a training data set and a test data set according to a set ratio; Input the training data set into the BP neural network, and calculate the input and output of the hidden layer and the output layer through forward propagation; The error of the output layer prediction result is calculated according to the mse calculation formula, which is expressed as: Where mse represents mean square error, y is the true value, is the predicted value, n is the total number of samples; The error is back-propagated, and based on the Adam optimizer, the model parameters are optimized through the following formula to adjust the connection weights and thresholds between the hidden layer and the output layer: m t =β1·m t-1 +(1-β1)·dk v t =β2·v t-1 +(1-β2)·dk 2 Where m t-1 ,v t-1 is the first-order moment estimate before iteration, m t ,v t is the first-order moment estimate after iteration, β1, β2 are exponentially weighted average parameters, dk represents the gradient of weight w and threshold b, α is the learning rate, ε is a constant to prevent the denominator from being 0, w t-1 ,b t-1 is the weight and threshold to be trained, w t ,b t are the trained weights and thresholds; Repeat the steps from forward propagation to back propagation until the preset number of training times is reached; Use the test data set to evaluate the trained BP neural network; The use of the test data set to evaluate the trained BP neural network specifically includes: After the model training is completed, the test data set is input, and the predicted value is obtained through forward propagation calculation. The mean square error MSE, root mean square error RMSE, mean absolute error MAE and determination coefficient R2 are used to evaluate the prediction results. The calculation formula is as follows: Among them, SSR represents the regression sum of squares, and SST represents the total sum of squares.

2. The method according to claim 1, characterized in that: The coal quality of the coal entering the furnace includes data on six properties: dry basis high calorific value, volatile matter, air-dried coal moisture, air-dried coal ash, dry basis total sulfur, and analytical hydrogen. The quantity data is the quantity of coal entering the furnace corresponding to the coal quality.

3. The method according to claim 2, characterized in that The cleaning of the original data specifically includes: When there are missing values ​​in the original data, if consecutive values ​​are missing, they are directly deleted. If a single data is missing, the missing value is interpolated by the nearest neighbor.

4. The method according to claim 1, characterized in that According to the hourly coal consumption of the unit, the smoke dust emission corresponding to each batch of coal entering the furnace is calculated in sequence, including: Calculate the hourly coal consumption of the unit based on the actual daily operating time and daily coal consumption of the unit; Position the coal index at the first batch of coal entering the furnace, position the smoke monitoring value index at the first smoke monitoring value, and obtain the current coal quantity, which is the quantity of the first batch of coal entering the furnace obtained when the smoke emission corresponding to a batch of coal entering the furnace is 0; If the current coal volume is greater than the hourly coal consumption, the smoke emission volume is added to the smoke monitoring value corresponding to the smoke monitoring value index, the smoke monitoring value index is increased by 1, and the current coal volume is subtracted from the hourly coal consumption; If the current coal quantity is less than the hourly coal consumption, the smoke dust emission will be added to the current coal quantity / hourly coal consumption*smoke dust monitoring value and stored. After storage, it will be updated again to (hourly coal consumption-current coal quantity) / hourly coal consumption*smoke dust monitoring value. The smoke dust monitoring value index will be increased by 1. The current coal quantity will be updated to the quantity of coal entering the furnace corresponding to the coal index + the current coal quantity-hourly coal consumption. The coal index entering the furnace will be increased by 1.

5. The method according to claim 4, characterized in that According to the hourly coal consumption of the unit, the smoke emission corresponding to each batch of coal entering the furnace is calculated in turn, including: The monitoring time for obtaining smoke emission data; When the running time is greater than the monitoring time: In the case of unit shutdown, after calculating the coal consumption and smoke emissions of the day, the current coal quantity minus the fraction of the operating time multiplied by the hourly coal consumption is used as the updated current coal quantity; When the unit is started, before calculating the coal consumption and smoke emission of the day, the current coal quantity minus the fraction of the running time multiplied by the hourly coal consumption is used as the updated current coal quantity; When the running time is less than the monitoring time: In the case of a unit shutdown, when processing the last monitoring hour of the day, the current coal quantity minus the fraction of the operating time multiplied by the hourly coal consumption is used as the updated current coal quantity. When the unit is started, when processing the first monitoring hour of the day, the current coal quantity minus the decimal part of the operating time multiplied by the hourly coal consumption is used as the updated current coal quantity.

6. A smoke concentration prediction device, characterized in that: The device comprises: A cleaning module is configured to clean the raw data, wherein the raw data includes coal quality and quantity data of coal entering the furnace and hourly smoke emission data; The calculation module is configured to calculate the smoke dust emission corresponding to each batch of coal entering the furnace according to the hourly coal consumption of the unit, and normalize the coal quality data, quantity data and smoke dust emission of each batch of coal entering the furnace to obtain a data set; A training module is configured to construct a prediction model, train the prediction model using the data set, and implement smoke concentration prediction with the trained prediction model; The prediction model is a BP neural network, which includes an input layer, at least one hidden layer and an output layer. The weight initialization method of the BP neural network adopts Glorot uniform distribution initialization, the threshold is initialized to 0, and the Relu function is selected as the activation function of the BP neural network; The training module is further configured to: Splitting the data set into a training data set and a test data set according to a set ratio; Input the training data set into the BP neural network, and calculate the input and output of the hidden layer and the output layer through forward propagation; The error of the output layer prediction result is calculated according to the mse calculation formula, which is expressed as: Where mse represents mean square error, y is the true value, is the predicted value, n is the total number of samples; The error is back-propagated, and based on the Adam optimizer, the model parameters are optimized through the following formula to adjust the connection weights and thresholds between the hidden layer and the output layer: m t =β1·m t-1 +(1-β1)·dk v t =β2·v t-1 +(1-β2)·dk 2 Where m t-1 ,v t-1 is the first-order moment estimate before iteration, m t ,v t is the first-order moment estimate after iteration, β1, β2 are exponentially weighted average parameters, dk represents the gradient of weight w and threshold b, α is the learning rate, ε is a constant to prevent the denominator from being 0, w t-1 ,b t-1 is the weight and threshold to be trained, w t ,b t are the trained weights and thresholds; Repeat the steps from forward propagation to back propagation until the preset number of training times is reached; Use the test data set to evaluate the trained BP neural network The training module is further configured to: After the model training is completed, the test data set is input, and the predicted value is obtained through forward propagation calculation, and the mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE) and determination coefficient (R) are used. 2 To evaluate the prediction results, the calculation formula is as follows: Among them, SSR represents the regression sum of squares, and SST represents the total sum of squares. 7 . A non-transitory computer-readable storage medium storing instructions, which, when executed by a processor, perform the method according to claim 1 .