Prediction and correction control method of calorific value of coal entering furnace combined with flame temperature

Abstract

The present invention relates to the field of soft measurement technology, and discloses a method for predicting and correcting the calorific value of incoming coal in combination with flame temperature, including providing an online soft measurement model of the calorific value of incoming coal based on LSTM, realizing dynamic modeling of the calorific value of incoming coal in a long time series, being able to predict the calorific value of coal in real time, and quickly tracking the change of coal quality in the furnace. On the other hand, a real-time correction method for the calorific value of incoming coal in combination with flame temperature is also provided, which fully mines the flame data information reflected by the combustion of fuel in the furnace, establishes the relationship between the change of the calorific value of the incoming coal and the distribution of the flame temperature field, and then corrects the calorific value of the incoming coal predicted by the LSTM model. The method proposed in the present invention performs dynamic modeling of the calorific value of the incoming coal for each layer of the pulverizing system, and can distinguish the change of the calorific value of the fuel of each mill, so that the online soft measurement result of the calorific value of the incoming coal is more accurate, so that it can fully adapt to the situation of significant changes in coal quality and coal blending in the current power plant.

Description

Technical Field

[0001] The invention relates to the technical field of soft measurement, and in particular to a method for predicting and correcting the calorific value of coal entering a furnace in combination with flame temperature. Background Art

[0002] At present, affected by the power market environment and the rise in coal prices, power generation companies have a tight supply of coal and a variety of types, resulting in a large difference between the coal entering the furnace and the original designed coal type. The quality of coal from different origins is uneven, and the power plant consumes a huge amount of coal. Affected by factors such as region, cost, and transportation, the power plant cannot burn a single type of coal in large quantities for a long time. In addition, power plants often burn low-calorific value and high-impedance inferior coal, and the blending cannot be guaranteed. The current control strategies are designed for a single type of coal and a single combustion method, which has great limitations. The uniformity of the blending makes the calorific value of the coal entering the furnace extremely unstable. When the unit is put into operation in AGC mode, the load command changes very frequently. If it cannot be adjusted in time when the calorific value of the coal fluctuates greatly, it is difficult to ensure the rate of load response and the control quality. It may even cause a large fluctuation in the main steam pressure, temperature and other important operating parameters of the unit, seriously affecting the safe and economic operation of the unit.

[0003] At present, the unit usually adopts the BTU correction method, which adapts to the change of coal quality of the unit by changing the coal supply according to the calorific value of the designed coal. However, the heat storage of the direct current boiler is small. When the calorific value of the coal seriously deviates from the designed coal, the BTU correction often causes the fluctuation of the boiler operating parameters, resulting in system instability. In addition, the BTU correction time lag is large, and it is impossible to adjust the parameters in time at the beginning of the coal quality change. It is difficult for the combustion adjustment to keep up with the load response in time, and it is easy to cause over-adjustment or under-adjustment of the coal quantity, resulting in poor quality of the coordinated control performance of the unit.

[0004] Therefore, being able to accurately and in real time monitor the calorific value of coal entering the furnace is of great significance to the combustion adjustment and operation regulation of the power plant. The existing technology has taken into account the operating conditions where multiple coal types may be blended. A single coal mill can independently set a calorific value correction coefficient, which is generally obtained by a simple calculation of the detected calorific value of the coal type in the combustion batch. In order to prevent over-adjustment of the coal quantity, the adjustment process of the BTU correction loop is relatively slow in actual production. The method of inputting the test calorific value through the coal quantity control of the coal mill can significantly shorten the adjustment process of the BTU calorific value correction. At present, there are two main methods for detecting the calorific value of coal entering the furnace: manual offline testing and online soft measurement.

[0005] The manual offline testing method is that the test personnel regularly take samples and prepare samples of coal at the boiler coal inlet every day, and then use instruments in the laboratory to analyze and test. However, the manual offline testing method is prone to sampling and sample preparation errors, and the test results usually have a long time lag, and the calorific value of the coal entering the furnace cannot be obtained in time. Therefore, this method is of little significance for the real-time operation adjustment of the power plant.

[0006] The online soft measurement method usually selects parameters related to boiler operation and combustion status as auxiliary variables, and uses mechanism modeling or artificial intelligence methods based on massive historical operation data to model and calculate the calorific value of coal entering the furnace, which belongs to the "black box" method. The general soft measurement method has certain feasibility and limitations. The accuracy of the measurement is directly related to the correctness of the established model. Moreover, after a period of operation, if the model is not updated and corrected, there will be problems with model inaccuracy. In addition, power plants nowadays mostly use coal blending and combustion, and the coal quality changes frequently. The calorific value of coal entering the furnace fluctuates greatly. The soft measurement method lacks the ability to adapt to changes in coal quality, and the prediction accuracy is difficult to guarantee. The soft measurement technology itself still has a lot of room for improvement. Depending on the selected feature parameters and the number of samples for training the model, the final soft measurement results may be significantly different. Summary of the invention

[0007] As a type of soft measurement technology, the present invention aims to overcome the shortcomings of the prior art and propose a method for predicting and correcting the calorific value of coal entering the furnace combined with flame temperature to solve the problem of inaccurate prediction of soft measurement models when online coal quality changes frequently and the calorific value of coal entering the furnace is unstable.

[0008] In order to achieve the above object, the present invention provides a method for predicting and correcting the calorific value of coal entering a furnace in combination with flame temperature, comprising the following steps:

[0009] S1. Select parameters related to the unit operation status from the power plant SIS system, including boiler load, main steam pressure, main steam flow, main steam temperature, reheat steam temperature, separator outlet temperature, feed water flow and secondary air volume;

[0010] Select the flue gas composition parameters related to combustion, including the NOX content of the flue gas at the SCR inlet, the CO content of the flue gas at the SCR inlet, the flue gas oxygen content, the carbon content of the fly ash, the carbon content of the slag, and the exhaust temperature;

[0011] Select parameters related to the pulverizing system, including coal feed rate, pulverizer current, pulverizer primary air volume, pulverizer primary air temperature, and pulverizer outlet temperature;

[0012] S2. Data preprocessing: After cleaning the data and removing outliers, normalization is performed to normalize the data to [0, 1], thereby eliminating the impact of different dimensions between data of different dimensions. The formula for data normalization is expressed by formula (1):

[0013]

[0014] Where: X norm is the normalized value, X is the original data, and X max is the maximum value of the data of each characteristic variable, X min is the maximum value of the data for each characteristic variable;

[0015] S3. Construct a long short-term memory neural network to establish a coal calorific value prediction model for each coal mill, and use the above data to train the constructed LSTM model;

[0016] S4. Based on massive historical data, a typical sample library of burner flame temperature and coal calorific value at each layer under certain load-pressure correspondence is established;

[0017] S5, inputting the data matrix to be tested within the continuous sampling time period into the trained coal calorific value prediction model, and obtaining the calorific value of the coal entering each coal mill in real time;

[0018] S6. Using the real-time measured flame temperature of each burner furnace, select the sample of the nearest sample from the typical sample library established in step S4 to correct the calorific value of the coal entering the furnace predicted by LSTM.

[0019] The method proposed in the present invention performs dynamic modeling of the calorific value of the coal entering the furnace for each layer of the pulverizing system, which can distinguish the changes in the calorific value of the fuel of each mill, making the online soft measurement results of the calorific value of the coal entering the furnace more accurate, and making it fully adaptable to the current situation of significant changes in coal quality and coal blending in power plants.

[0020] Furthermore, in the LSTM-based coal calorific value model established in step S3, the Adam algorithm is used to update the weight parameters of the neural network during the training process.

[0021] Furthermore, in step S4, the research object is a boiler with front and rear counter-combustion mode, and the furnace flame temperature collected and used comes from the fire detection temperature probe; three layers of vortex burners are arranged on the front and rear walls of the furnace, each layer has 6 burners, and a total of 36 burners; correspondingly, the fire detection temperature probes are arranged on the burners on the front and rear walls of the furnace, and each layer has 6 flame temperature measurement points.

[0022] Furthermore, in step S5, when predicting the calorific value of the coal entering the furnace in real time online, only the nearest future time step is predicted, and then the predicted calorific value of the coal entering the furnace is added to the training set after the prediction, and the prediction is gradually rolled forward, thereby realizing dynamic modeling and continuous updating and optimization of the model.

[0023] Furthermore, in step S6, the calorific value of the incoming coal predicted by LSTM is corrected using the flame temperature of each layer.

[0024] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of a method for predicting and correcting the calorific value of incoming coal in one embodiment of the present invention;

[0026] Figure 2 LSTM structure principle diagram in one embodiment of the present invention;

[0027] Figure 3 It is a schematic diagram of the installation structure of a fire detection temperature probe in one embodiment of the present invention;

[0028] Figure 4 The present invention is a flow chart of a method for correcting the calorific value of coal entering a furnace in one embodiment of the present invention. DETAILED DESCRIPTION

[0029] The specific embodiments of the present invention are described in detail below. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0030] In the present invention, unless otherwise stated, the directional words such as "upper" and "lower" usually refer to the orientation in the assembled state. "Inside" and "outside" refer to the inside and outside relative to the outline of each component itself.

[0031] The present invention provides a method for predicting and correcting the calorific value of coal entering the furnace in combination with flame temperature, such as Figure 1-Figure 4 As shown, the following steps are included:

[0032] S1. Select parameters related to the unit operation status from the power plant SIS system, including boiler load, main steam pressure, main steam flow, main steam temperature, reheat steam temperature, separator outlet temperature, feed water flow and secondary air volume;

[0033] Select the flue gas composition parameters related to combustion, including the NOX content of the flue gas at the SCR inlet, the CO content of the flue gas at the SCR inlet, the flue gas oxygen content, the carbon content of the fly ash, the carbon content of the slag, and the exhaust temperature;

[0034] Select parameters related to the pulverizing system, including coal feed rate, pulverizer current, pulverizer primary air volume, pulverizer primary air temperature, and pulverizer outlet temperature;

[0035] S2. Data preprocessing: After cleaning the data and removing outliers, normalization is performed to normalize the data to [0, 1], thereby eliminating the impact of different dimensions between data of different dimensions. The formula for data normalization is expressed by formula (1):

[0036]

[0037] Where: X norm is the normalized value, X is the original data, and X max is the maximum value of the data of each characteristic variable, X min is the maximum value of the data for each characteristic variable;

[0038] S3. Construct a long short-term memory neural network to establish a coal calorific value prediction model for each coal mill, and use the above data to train the constructed LSTM model;

[0039] S4. Based on massive historical data, a typical sample library of the flame temperature and coal calorific value of each layer of burners under a certain load-pressure is established; taking the A-layer pulverizing system as an example, there are N data in total, and the typical sample library established is

[0040]

[0041] S5, inputting the data matrix to be tested within the continuous sampling time period into the trained coal calorific value prediction model, and obtaining the calorific value of the coal entering each coal mill in real time;

[0042] S6. Using the real-time measured flame temperature of each burner furnace, select the sample of the nearest sample from the typical sample library established in step S4 to correct the calorific value of the coal entering the furnace predicted by LSTM.

[0043] The first aspect of the present invention provides an online soft measurement model of the calorific value of the coal entering the furnace based on LSTM, which realizes dynamic modeling of the calorific value of the coal entering the furnace in a long time series, can predict the calorific value of the coal in real time, and quickly track the changes in the coal quality in the furnace. Another aspect of the present invention also provides a real-time correction method for the calorific value of the coal entering the furnace in combination with the flame temperature, which fully mines the flame data information reflected by the combustion of the fuel in the furnace, establishes the relationship between the change in the calorific value of the coal entering the furnace and the distribution of the flame temperature field, and then corrects the calorific value of the coal entering the furnace predicted by the LSTM model. The method proposed in the present invention performs dynamic modeling of the calorific value of the coal entering the furnace for each layer of the pulverizing system, which can distinguish the changes in the calorific value of the fuel of each mill, making the online soft measurement results of the calorific value of the coal entering the furnace more accurate, so that it can fully adapt to the current situation of significant changes in coal quality and coal blending in power plants.

[0044] In an optional implementation, the LSTM-based coal calorific value model established in step S3 uses the Adam algorithm to update the weight parameters of the neural network during the training process.

[0045] In an optional embodiment, in step S4, the research object is a boiler with front and rear counter-combustion mode, and the furnace flame temperature collected and used comes from the fire detection temperature probe; three layers of vortex burners are arranged on the front and rear walls of the furnace, each layer has 6 burners, and a total of 36 burners; correspondingly, the fire detection temperature probes are arranged on the burners on the front and rear walls of the furnace, and each layer has 6 flame temperature measurement points.

[0046] In an optional embodiment, in step S5, when predicting the calorific value of the coal entering the furnace online in real time, only the nearest future time step is predicted to obtain the calorific value of the coal entering the furnace, and then the calorific value of the coal entering the furnace is added to the training set after the prediction, and the prediction is gradually rolled forward, thereby realizing dynamic modeling and continuous updating and optimization of the model.

[0047] In an optional implementation, in step S6, the calorific value of the incoming coal predicted by LSTM is corrected using the flame temperature of each layer. Taking layer A as an example, there are 6 flame temperature measurement points, and the specific correction process is as follows:

[0048] Based on the Euclidean distance, the k samples closest to the real-time furnace flame temperature in the typical sample library are obtained, and the coal calorific value corresponding to these k samples is obtained, and re-arranged into Q A (1) , Q A (2) , ..., Q A (k) ;

[0049] Calculate the calorific value Q of coal obtained by LSTM prediction model 预测The relationship between the average calorific value of the coal of the nearest k samples is: 预测 -average(Q A (1) +Q A (2) +...+Q A (k) )|≤ε, which is less than a certain constant ε, then no correction is performed; otherwise, correction is performed based on the relationship between the real-time flame temperature and the temperature of the nearest k samples, thereby obtaining the final calorific value of the coal entering the furnace;

[0050] Q=λ·Q 预测

[0051] λ is the correction coefficient, which is the average value of the ratio of the average value of each flame temperature of the nearest k samples to the real-time value. The calculation method is expressed by formula (2):

[0052]

[0053] As an embodiment of the present invention, the adopted long short-term memory network LSTM structure is described as follows.

[0054] LSTM is a recurrent neural network that has good learning performance for long time series data. LSTM uses its unique gating system to remove or add information to the cell state, selectively retaining the data of the hidden layer at the previous moment, thus effectively avoiding the problems that recurrent neural networks encounter when facing longer sequence data.

[0055] LSTM network structure unit such as Figure 2 As shown in Figure 1, an LSTM consists of several hidden layers. Each LSTM unit contains three gates that control the state and an internal memory unit. The three gates are the forget gate, input gate, and output gate. The input gate selectively records the new input information to the internal memory unit, the forget gate selectively forgets the information of the unit state, and the output gate selectively outputs the information of the unit state.

[0056] In the figure t represents the output of the input gate unit, f t Represents the output of the forget gate unit, o t represents the output of the output gate unit, is a candidate memory unit, c t represents the internal memory unit, h t is the output of the hidden unit. t is the input of the unit, h t-1 is the output of the previous hidden unit.

[0057] When there is no useful information in the input sequence, the input gate i t The value of is close to 0, and the forget gate f t The value will be close to 1, which means that useful information in the past is saved and used.

[0058] When the input sequence contains important information, then the input gate i t The value of is close to 1, and the forget gate f t The value will be close to 0, which means forgetting past information and selecting and recording current useful information.

[0059] Forget gate controls input x t and the previous hidden layer output h t-1 The degree of being forgotten is calculated as follows:

[0060] f t =σ(x t W xf +h t-1 W hf +b f )

[0061] Where: W xf , W hf x t and h t-1 The weight of the forget gate, b f is the bias of the forget gate, and σ(·) represents the sigmoid function.

[0062] The input gate controls the input x t And the degree to which the current calculated state is updated to the memory unit, the formula is as follows:

[0063] i t =σ(x t W xi +h t-1 W hi +b i )

[0064] Where: W xi , W hi x t and h t-1 The weight of the input gate, b i is the bias term of the input gate.

[0065] The information of the internal memory unit is updated as follows:

[0066]

[0067] Where: W xc , W hc They are x t and ht-1 The weight of the internal computing unit, b c is the bias term of the internal memory unit.

[0068] Output gate controls input x t And the current output depends on the degree of the current memory cell:

[0069] o t =σ(x t W xo +h t-1 W ho +b o )

[0070] h t =o t ·tanh(c t )

[0071] Where: W xo , W ho x t and h t-1 The weight of the output gate, b o is the bias of the output gate.

[0072] The LSTM model used in the present invention contains 1 hidden layer, and the number of hidden layer units is 128.

[0073] The root mean square error (RMSE) and mean absolute percentage error (MAPE) are used to evaluate the model, which are defined as follows:

[0074]

[0075] Where: n is the number of samples to be predicted, y i represents the predicted value of the i-th sample, Represents the actual value of the i-th sample.

[0076] like Figure 3 As shown, as another embodiment of the present invention, a flame detection probe capable of measuring temperature is installed on the burner, and is mainly composed of an infrared temperature measuring device, a signal processing unit and an analog display unit. The infrared temperature measuring device includes an infrared detector, a thermal insulation coating and a cooling unit. The device can output a flame temperature signal of 800°C-1800°C with an accuracy of ±1%. The flame temperature can be monitored and long-term historical data can be stored on the host computer.

[0077] The combustion of coal in the boiler is a complex reaction process, and the heat transfer in the furnace has great inertia and delay. Therefore, the general operating parameters of the boiler cannot quickly and timely respond to the changes in the combustion caused by the change in coal quality in the furnace. The furnace flame temperature is an important parameter in the combustion process in the reaction furnace, which can characterize the combustion state of the boiler. At the same time, the size of the flame temperature of the burners on different layers in the furnace space distribution reflects the different energy released by the calorific value of the coal entering the furnace in different pulverizing systems. Generally speaking, the higher the calorific value of the coal entering the furnace, the greater the heat per unit space, and the better the combustion condition of the boiler.

[0078] The present invention uses a fire detection infrared temperature measuring device installed on the burner to obtain a flame temperature that can quickly respond to changes in combustion in the furnace. The front and rear walls of the furnace are each arranged with three layers of vortex burners, each layer has 6 burners, and a total of 36 burners. Correspondingly, the fire detection temperature probes are arranged on the burners on the front and rear walls of the furnace, and each layer has 6 flame temperature measurement points.

[0079] Based on this, Figure 4 As shown, as an embodiment of the present invention, a real-time correction method for the calorific value of coal entering the furnace combined with the flame temperature is proposed.

[0080] Taking layer A as an example, the specific correction process is as follows:

[0081] (1) Based on the Euclidean distance, obtain the k samples closest to the real-time furnace flame temperature in the typical sample library, obtain the coal calorific value corresponding to these k samples, and re-rank them into Q A (1) , Q A (2) , …, Q A (k) .

[0082] The present invention takes k=5.

[0083] (2) Calculate the coal calorific value Q obtained by the LSTM prediction model 预测 The relationship between the average calorific value of the coal of the nearest k samples is, if |Q 预测 -average(Q A (1) +Q A (2) +…+Q A (k) )|≤ε, then no correction is performed. Here ε=15.

[0084] Otherwise, correction is performed based on the relationship between the real-time flame temperature and the temperatures of the k nearest samples to obtain the final calorific value of the coal entering the furnace.

[0085] (3) The specific correction is to multiply the calorific value of the coal entering the furnace predicted by LSTM by a correction coefficient λ to obtain the corrected calorific value of the coal entering the furnace.

[0086] Q=λ·Q 预测

[0087] λ is the correction coefficient. Specifically, it is the average value of the sum of the ratios of the average value of each flame temperature of the nearest k samples to the real-time value, and is calculated as follows:

[0088]

[0089] The preferred embodiments of the present invention are described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.

[0090] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0091] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A method for predicting and correcting the calorific value of coal entering a furnace in combination with flame temperature, characterized in that: The following steps are involved: S1. Select parameters related to the unit operation status from the power plant SIS system, including boiler load, main steam pressure, main steam flow, main steam temperature, reheat steam temperature, separator outlet temperature, feed water flow and secondary air volume; Select the flue gas composition parameters related to combustion, including the NOX content of the flue gas at the SCR inlet, the CO content of the flue gas at the SCR inlet, the flue gas oxygen content, the carbon content of the fly ash, the carbon content of the slag, and the exhaust temperature; Select parameters related to the pulverizing system, including coal feed rate, pulverizer current, pulverizer primary air volume, pulverizer primary air temperature, and pulverizer outlet temperature; S2. Data preprocessing: After cleaning the data and removing outliers, normalization is performed to normalize the data to [0, 1], thereby eliminating the impact of different dimensions between data of different dimensions. The formula for data normalization is expressed by formula (1): Where: X norm is the normalized value, X is the original data, and X max is the maximum value of the data of each characteristic variable, X min is the maximum value of the data for each characteristic variable; S3. Construct a long short-term memory neural network to establish a coal calorific value prediction model for each coal mill, and use the above data to train the constructed LSTM model; S4. Based on massive historical data, a typical sample library of burner flame temperature and coal calorific value at each layer under certain load-pressure correspondence is established; S5, inputting the data matrix to be tested within the continuous sampling time period into the trained coal calorific value prediction model, and obtaining the calorific value of the coal entering each coal mill in real time; S6. Using the real-time measured flame temperature of each burner furnace, select the sample of the nearest sample from the typical sample library established in step S4 to correct the calorific value of the coal entering the furnace predicted by LSTM.

2. The method for predicting and correcting the calorific value of coal entering the furnace in combination with flame temperature according to claim 1, characterized in that: In step S3, the LSTM-based coal calorific value model is established, and the Adam algorithm is used to update the weight parameters of the neural network during the training process.

3. The method for predicting and correcting the calorific value of coal entering the furnace in combination with flame temperature according to claim 1, characterized in that: In step S4, the research object is a boiler with front and rear counter-combustion mode, and the furnace flame temperature collected and used comes from the fire detection temperature probe; three layers of vortex burners are arranged on the front and rear walls of the furnace respectively, with 6 burners on each layer, for a total of 36 burners; correspondingly, the fire detection temperature probes are arranged on the burners on the front and rear walls of the furnace, and each layer has 6 flame temperature measurement points.

4. The method for predicting and correcting the calorific value of coal entering the furnace in combination with flame temperature according to claim 1, characterized in that: In step S5, when predicting the calorific value of the coal entering the furnace in real time online, only the nearest future time step is predicted, and then the predicted calorific value of the coal entering the furnace is added to the training set after the prediction, and the prediction is gradually rolled forward, thereby realizing dynamic modeling and continuous updating and optimization of the model.

5. The method for predicting and correcting the calorific value of coal entering the furnace in combination with flame temperature according to claim 1, characterized in that: In step S6, the calorific value of the incoming coal predicted by LSTM is corrected using the flame temperature of each layer.

Images