A spectral-based method for identifying combustion states in coal-fired power plant furnaces

Abstract

The present invention discloses a method for identifying the combustion state of a coal-fired power plant furnace based on spectroscopy, which belongs to the field of furnace combustion technology in thermal power plants. The method comprises the following steps: using a spectrometer to measure the furnace flame radiation spectrum signal of 200nm to 1750nm; selecting a wavelength range to calculate the flame temperature T and the flame blackness ε as the first characteristic parameter of the furnace flame combustion; calculating the intensity ratio R of the Fe element radiation spectrum intensity to the average radiation intensity in the wavelength range selected in step 2 as the second characteristic parameter of the furnace flame combustion; calculating the total radiation intensity D as the third characteristic parameter of the furnace flame combustion; selecting a data set of three characteristic parameters as the input layer of a deep learning training model, and designing the output state; and inputting the three characteristic parameters of the spectral data into the trained model to identify the combustion state. The present invention can identify the furnace combustion state during the deep peak regulation period of the unit and guide the on-site operating conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of furnace combustion in thermal power plants, and in particular to a method for identifying the combustion state of a furnace in a coal-fired power plant based on spectrum. Background Art

[0002] Renewable energy sources are unable to provide stable power to the power grid due to environmental factors. Coal-fired power plants are required to maintain a balanced power supply load through deep peak shaving and rapid load changes. During deep peak shaving, the furnace temperature drops rapidly, making it difficult for pulverized coal to ignite and severely impacting the stability of the furnace flame. Accurate and real-time assessment of the furnace flame's combustion status is essential for adjusting combustion conditions.

[0003] Traditional spectral flame detection methods include ultraviolet (UV), infrared (IR), and visible light (VL) flame detection, each utilizing different spectral signals. UV flame detection utilizes the characteristic spectral signals generated by flame combustion within the UV spectrum; infrared flame detection utilizes the infrared radiation generated by flame combustion; and visible light flame detection utilizes the visible light signals generated by flame combustion.

[0004] In their research on flame combustion stability detection based on radiation spectroscopy, Huang Yaosong et al. used a photoelectric conversion system consisting of a phototransistor and a detection circuit to convert flame radiation light signals into electrical signals. They then extracted their time, frequency, and phase characteristics and used three algorithms: Mahalanobis distance, neural network, and least squares support vector machine (LSSVM). However, they did not extract features from the different spectral signals in each band.

[0005] Patent publication number CN115541803A in the prior art discloses a spectral analysis method for online detection of combustion flame stability in coal-fired power plant boilers, comprising the following steps: S1: using a spectrometer to collect and measure the 200-1100nm radiation spectrum; S2: measuring the parameters of average spectral radiation intensity, flame pulsation frequency, combustion temperature and average spectral emissivity based on the collected spectrum; S3: using the concentration of burned coal powder as a parameter to characterize the combustion stability index, and substituting the parameters measured in step S2 into it to establish an equation; S4: establishing a function of the combustion stability index CSI through a linear regression algorithm, thereby realizing online monitoring of the combustion stability index. However, the existing technology has the following defects: the measured band in the above technology is 200-1000nm, which fails to reflect the changes in the spectrum of the flame generated during boiler combustion in the near-infrared light band; the above technology uses the average spectral radiation intensity as a measurement parameter. In practice, depending on the different flame combustion states, the generated radiation intensity will only be reflected within a specific band, and the use of the average spectral radiation intensity will weaken this phenomenon; the above technology uses the burning coal powder concentration as a parameter to characterize the combustion stability index. In actual power plant operation, in response to environmental protection requirements, coal-fired units usually adopt a low-nitrogen combustion method. In order to reduce the flame core temperature to reduce NOx emissions, the flame near the burner is in an oxygen-deficient combustion state. The use of the burning coal powder concentration as a parameter to characterize the combustion stability index cannot reflect the impact of changes in oxygen content on it.

[0006] Patent publication number CN106897540A discloses a method for online monitoring and optimization of the combustion state of a boiler burner, comprising the following steps: arranging a spectroscopic flame detector at the boiler burner, collecting monitoring signals from the spectroscopic flame detector and the furnace pressure sensor, processing the signals on an intelligent processor, obtaining relevant characteristic parameters as input parameters, and using an artificial neural network algorithm to obtain parameters of the combustion state and furnace NOx emissions, determine the rationality of the combustion state, and transmit relevant adjustment instructions to the DCS system, thereby adjusting the boiler combustion and achieving more refined and intelligent control of the boiler. However, this prior art has the following defects: the characteristic parameters described in the above technology include the time mean, mean square error, standard deviation, power spectral density, power spectral entropy, and wavelet energy entropy of the infrared spectrum, visible light spectrum, ultraviolet spectrum, and furnace pressure signal of the combustion flame. However, in the actual measured furnace flame spectrum, the ultraviolet spectrum signal is extremely weak and cannot be used as a characteristic parameter of the furnace flame combustion. In addition, only certain bands in the infrared spectrum and visible light spectrum can well reflect the furnace flame combustion.

[0007] Therefore, a reliable method for identifying the combustion state of the furnace in a coal-fired power plant is needed. By collecting the furnace flame radiation spectrum signal within 200-1750nm and analyzing and training it, the problem of identifying the furnace flame combustion state during deep peak regulation is solved. Summary of the Invention

[0008] The present invention aims to provide a method for identifying the combustion state of a coal-fired power plant furnace based on spectroscopy, which is characterized by comprising the following steps:

[0009] Step 1: Use a spectrometer to measure the furnace flame radiation spectrum signal from 200nm to 1750nm;

[0010] Step 2: Based on the Planck quantitative and two-color gray body characteristic judgment principles, the wavelength range is selected to calculate the flame temperature T and flame blackness ε as the first characteristic parameters of the furnace flame combustion;

[0011] Step 3: Calculate the intensity ratio R of the Fe element radiation spectrum intensity to the average radiation intensity in the band range selected in step 2 as the second characteristic parameter of the furnace flame combustion;

[0012] Step 4: Calculate the total radiation intensity D as the third characteristic parameter of the furnace flame combustion;

[0013] Step 5: Select the data sets of the first characteristic parameter, the second characteristic parameter, and the third characteristic parameter as the input layer of the deep learning training model, and design the output state according to different oxygen concentrations;

[0014] Step 6: Input the three characteristic parameters of the spectral data into the trained deep learning training model to identify the combustion state.

[0015] The wavelength ranges in step 2 are 800nm ​​to 1000nm and 1400nm to 1600nm.

[0016] The calculation methods of the flame temperature T and the flame blackness ε in step 2 are as follows:

[0017]

[0018] Where λ is the wavelength, Δλ is the wavelength change, c is the speed of light, h is Planck's constant, k is the Boltzmann constant, T is the flame temperature, ε(λ) is the flame blackness, I(λ,T) is the measured radiation intensity, and I b (λ,T) is the monochromatic blackbody radiation intensity of the radiating object.

[0019] The intensity ratio R in step 3 is calculated as follows:

[0020]

[0021] Where q Fe is the spectral radiation intensity emitted by Fe element, q min is the minimum value of the radiation intensity in the selected wavelength range, a, b are the minimum and maximum values ​​of the selected wavelength range respectively, q i is the radiation intensity corresponding to each wavelength.

[0022] The calculation method of the total radiation intensity D in step 4 is as follows:

[0023]

[0024] Where λ is the wavelength, a and b are the minimum and maximum values ​​of the selected wavelength range, respectively, and I(λ) is the measured spectral radiation intensity.

[0025] The output states in step 5 are oxygen-rich combustion, normal combustion, oxygen-deficient combustion, severe oxygen deficiency and extinction.

[0026] The deep learning training model in step 5 is a BP multi-layer neural network, and the activation function σ(x) used is:

[0027]

[0028] The back propagation function is:

[0029]

[0030]

[0031] Where y i To set the output value, is the output value of the output layer, p is the number of outputs of the output layer, E total is the variance of all set output values ​​and output layer output values, η is the learning rate, ω ij is the weight value of the network connection, is the updated weight value.

[0032] A spectral-based device for identifying the combustion state of a coal-fired power plant furnace, comprising:

[0033] The measurement module is used for measuring the furnace flame radiation spectrum signal of 200nm to 1750nm by spectrometer;

[0034] A calculation module is used to select a wavelength range to calculate the flame temperature T and flame blackness ε as the first characteristic parameter of the furnace flame combustion based on the Planck quantitative and two-color gray body characteristic judgment principles; calculate the intensity ratio R of the Fe element radiation spectrum intensity to the average radiation intensity in the wavelength range selected in step 2 as the second characteristic parameter of the furnace flame combustion; and calculate the total radiation intensity D as the third characteristic parameter of the furnace flame combustion;

[0035] The recognition module is used to select the data sets of the first characteristic parameter, the second characteristic parameter, and the third characteristic parameter as the input layer of the deep learning training model, and design the output state according to different oxygen concentrations; the three characteristic parameters of the spectral data are input into the trained deep learning training model to identify the combustion state.

[0036] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The electronic device is characterized in that when the processor executes the computer program, each step of a method for identifying the combustion state of a coal-fired power plant furnace based on spectroscopy is implemented.

[0037] A storage medium stores a computer program thereon, wherein the computer program, when executed by a processor, implements each step of a method for identifying the combustion state of a coal-fired power plant furnace based on spectroscopy.

[0038] The beneficial effects of the present invention are:

[0039] The furnace combustion state identification model proposed in the present invention can identify the furnace combustion state during the deep peak regulation period of the unit and guide the on-site operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flow chart of the method for identifying the combustion state of a coal-fired power plant furnace based on spectroscopy according to the present invention;

[0041] Figure 2 Diagram of the deep learning training model used in this invention. DETAILED DESCRIPTION

[0042] The present invention proposes a method for identifying the combustion state of a coal-fired power plant furnace based on spectroscopy. The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0043] Figure 1 This is a flow chart of the method for identifying the combustion state of a coal-fired power plant furnace based on spectroscopy according to the present invention; the method comprises the following steps:

[0044] 1) Use a spectrometer to measure multiple groups of furnace flame radiation signals from 200 to 1750 nm.

[0045] 2) Select appropriate wavelength ranges within the wavelength range of 200-1750nm to calculate the flame temperature T and flame blackness ε as characteristic parameters of furnace flame combustion. The calculation method is as follows:

[0046] According to Planck's law, the radiation intensity I(λ,T) of an object with a wavelength of λ, a temperature of T, and a blackness of ε(λ) is:

[0047]

[0048] In formula (1), λ is the wavelength, Δλ is the wavelength change, c is the speed of light, h is the Planck constant, k is the Boltzmann constant, T is the flame temperature, and ε(λ) is the flame blackness. Generally, in industrial use, Planck's law can be replaced by Wien's law:

[0049]

[0050] In formula (2), I b (λ,T) is the monochromatic blackbody radiation intensity of the radiating object.

[0051] According to the gray body characteristic judgment principle based on the two-color method, the two monochromatic radiation intensities I(λ,T) and I(λ+Δλ,T) emitted from the same direction by the furnace flame obtained by the spectrometer are divided to obtain formula (3):

[0052]

[0053] In formula (3), ε(λ) is the blackness of the furnace flame at wavelength λ, and ε(λ+Δλ) is the blackness at wavelength λ+Δλ. When Δλ is small enough, ε(λ) / ε(λ+Δλ)≈1. The temperature distribution can be calculated based on the ratio of the monochromatic radiation intensities I(λ,T) and I(λ+Δλ,T) corresponding to λ and λ+Δλ:

[0054]

[0055] After obtaining the temperature T distribution, the flame blackness ε(λ) distribution can be calculated by using the ratio of the detected radiation intensity to the blackbody intensity at the same T:

[0056]

[0057] The distribution change of monochromatic blackness ε(λ) with wavelength can be used to determine whether the radiation satisfies the gray body characteristics. If the blackness changes with wavelength, the gray body assumption cannot be adopted.

[0058] 3) In the furnace flame spectrum, the spectral radiation intensities of Na, K, and Fe elements are all quite obvious and exhibit the following characteristics: when the furnace burns at low temperatures, the spectral radiation intensities of these atoms are low; when the furnace burns at high temperatures, the spectral radiation intensity gradually increases with increasing temperature. Experiments have shown that the spectral radiation intensity of Fe is much higher than that of Na and K. Therefore, the ratio R of the Fe element's radiation spectral intensity to the average radiation intensity in the selected wavelength range is calculated as another characteristic parameter of furnace flame combustion. The intensity ratio R is calculated as follows:

[0059]

[0060] In formula (6), q Fe is the radiation intensity emitted by Fe element, q min is the minimum value of the radiation intensity in the selected wavelength range, a and b are the maximum and minimum values ​​of the selected wavelength range respectively, and q i is the radiation intensity corresponding to each wavelength.

[0061] 4) Under different combustion conditions, the radiation intensity signal of the furnace flame changes significantly in the wavelength range of 800nm-1000nm and 1400nm-1600nm. The total radiation intensity D is obtained by integrating them respectively as another characteristic parameter of the furnace flame combustion. The total radiation intensity D is as shown in formula (7):

[0062]

[0063] Where λ is the wavelength and I(λ) is the measured spectral radiation intensity.

[0064] 5) Take the above feature parameters as the input layer {x i |i=1,2,...,5} input deep learning algorithm, the furnace flame will show different combustion states under different oxygen concentrations, so 5 output states {y i |i=1,2,...,5}, which are respectively: oxygen-rich combustion, normal combustion, oxygen-deficient combustion, severe oxygen deficiency and extinction.

[0065] Figure 2 This is a diagram of the deep learning training model used in the present invention. The deep learning algorithm used in this embodiment is the BP multi-layer neural network algorithm. The BP multi-layer neural network algorithm is a very flexible classification and recognition method. The characteristic parameters of the furnace flame spectrum signal are used as the input of the BP neural network to obtain the flame combustion state, and different output states {y i |i=1,2,...,5} are each set to 1.

[0066] The BP multi-layer neural network is a feedforward network consisting of an input layer, one or more hidden layers, an output layer, and a soft-max layer to obtain the final prediction result. The specific operation of the network is shown below.

[0067] Activation: The commonly used activation functions are ReLU activation function and sigmoid activation function. The image of sigmoid activation function is a nonlinear curve. Therefore, it can better approximate the nonlinear relationship. The function is shown in formula (8):

[0068]

[0069] The feature parameters are taken as the input layer {x i |i=1,2,...,m} input network, hidden layer parameters {h ij |i=1,2,...,m;j=1,2,...,n} is the value of the input layer after the sigmoid activation function transformation. This process can be expressed as

[0070]

[0071] where h ij is the hidden layer parameter, ω ij is the weight value of the network connection, b ij is the deviation value.

[0072] In the hidden layer, each neuron works like the input layer, using the sigmoid activation function to transform the value from the previous layer. For example, the hidden layer parameter h1j is transformed into the input of h2j in the next step, and h2j is transformed into the input of h3j in the next step, and so on until the output. The number of hidden neurons is set to 7, which is determined by cross-validation. The output value of the final output layer is

[0073] Back propagation: The output error (the difference between the output value of the output layer and the set value) is calculated by back propagating along the original path, and then back propagated through the hidden layer to the input layer. During the back propagation process, the error is distributed to each unit in each layer, and the error signal of each unit in each layer is obtained, which is used as the basis for correcting the weight value of each unit. This calculation process is completed using the gradient descent method. After continuously adjusting the weight value of each layer of neurons, the error signal is reduced to a minimum. The calculation formula of the gradient descent method is as follows (10) (11):

[0074]

[0075] Where y i To set the output, is the output value of the output layer, p is the number of output values ​​of the output layer, E totalis the variance of all set output values ​​and output layer output values, η is the learning rate, ω ij is the weight value of the network connection, is the updated weight value.

[0076] 6) The trained BP multi-layer neural network model is retained, and the characteristic parameters based on the furnace flame combustion spectrum signal are input to identify each combustion state.

[0077] This embodiment further provides a spectral-based device for identifying the combustion state of a coal-fired power plant furnace, comprising:

[0078] The measurement module is used for measuring the furnace flame radiation spectrum signal of 200nm to 1750nm by spectrometer;

[0079] A calculation module is used to select a wavelength range to calculate the flame temperature T and flame blackness ε as the first characteristic parameter of the furnace flame combustion based on the Planck quantitative and two-color gray body characteristic judgment principles; calculate the intensity ratio R of the Fe element radiation spectrum intensity to the average radiation intensity in the wavelength range selected in step 2 as the second characteristic parameter of the furnace flame combustion; and calculate the total radiation intensity D as the third characteristic parameter of the furnace flame combustion;

[0080] The recognition module is used to select the data sets of the first characteristic parameter, the second characteristic parameter, and the third characteristic parameter as the input layer of the deep learning training model, and design the output state according to different oxygen concentrations; the three characteristic parameters of the spectral data are input into the trained deep learning training model to identify the combustion state.

[0081] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, each step of the spectral-based method for identifying the combustion state of a coal-fired power plant furnace is implemented.

[0082] This embodiment further provides a storage medium storing a computer program. When the computer program is executed by a processor, each step of the method for identifying the combustion state of a coal-fired power plant furnace based on spectroscopy is implemented.

[0083] This detection method collects the furnace flame radiation spectrum signal within 200-1750nm, and extracts the flame temperature, flame blackness, intensity ratio and the total radiation intensity of two different wavelengths as characteristic parameters. It uses a deep learning algorithm for training to identify five combustion states: oxygen-rich combustion, normal combustion, oxygen-deficient combustion, severe oxygen deficiency and extinction. The resulting method is a reliable method for identifying the combustion state of the furnace of a coal-fired power plant, which can guide on-site operating conditions during deep peak regulation.

Claims

1. A method for identifying the combustion state of a coal-fired power plant furnace based on spectroscopy, characterized in that: The following steps are involved: Step 1: Use a spectrometer to measure the furnace flame radiation spectrum signal from 200nm to 1750nm; Step 2: Based on the Planck quantitative and two-color gray body characteristic judgment principles, the wavelength range is selected to calculate the flame temperature T and flame blackness ε as the first characteristic parameters of the furnace flame combustion; Step 3: Calculate the intensity ratio R of the Fe element radiation spectrum intensity to the average radiation intensity in the band range selected in step 2 as the second characteristic parameter of the furnace flame combustion; Step 4: Calculate the total radiation intensity D as the third characteristic parameter of the furnace flame combustion; Step 5: Select the data sets of the first characteristic parameter, the second characteristic parameter, and the third characteristic parameter as the input layer of the deep learning training model, and design the output state according to different oxygen concentrations; Step 6: Input the three characteristic parameters of the spectral data into the trained deep learning training model to identify the combustion state.

2. The method for identifying the combustion state of a coal-fired power plant furnace based on spectrum according to claim 1, characterized in that: The wavelength ranges in step 2 are 800nm ​​to 1000nm and 1400nm to 1600nm.

3. The method for identifying the combustion state of a coal-fired power plant furnace based on spectrum according to claim 1, characterized in that: The calculation methods of the flame temperature T and the flame blackness ε in step 2 are as follows: Where λ is the wavelength, Δλ is the wavelength change, c is the speed of light, h is Planck's constant, k is the Boltzmann constant, T is the flame temperature, ε(λ) is the flame blackness, I(λ,T) is the measured radiation intensity, and I b (λ,T) is the monochromatic blackbody radiation intensity of the radiating object.

4. The method for identifying the combustion state of a coal-fired power plant furnace based on spectrum according to claim 1, characterized in that: The intensity ratio R in step 3 is calculated as follows: R=(q Fe -q min ) / q ave Where q Fe is the spectral radiation intensity emitted by Fe element, q min is the minimum value of the radiation intensity in the selected wavelength range, a, b are the minimum and maximum values ​​of the selected wavelength range respectively, q ave is the average radiation intensity in the selected wavelength range, q i is the radiation intensity corresponding to each wavelength.

5. The method for identifying the combustion state of a coal-fired power plant furnace based on spectrum according to claim 1, characterized in that: The calculation method of the total radiation intensity D in step 4 is as follows: Where λ is the wavelength, a and b are the minimum and maximum values ​​of the selected wavelength range, respectively, and I(λ) is the measured spectral radiation intensity.

6. The method for identifying the combustion state of a coal-fired power plant furnace based on spectrum according to claim 1, characterized in that: The output states in step 5 are oxygen-rich combustion, normal combustion, oxygen-deficient combustion, severe oxygen deficiency and extinction.

7. The method for identifying the combustion state of a coal-fired power plant furnace based on spectrum according to claim 1, characterized in that: The deep learning training model in step 5 is a BP multi-layer neural network, and the activation function σ(x) used is: The back propagation function is: Where y i To set the output value, is the output value of the output layer, p is the number of outputs of the output layer, E total is the variance of all set output values ​​and output layer output values, η is the learning rate, ω ij is the weight value of the network connection, is the updated weight value.

8. An identification device for the method for identifying the combustion state of a coal-fired power plant furnace based on spectrum according to claim 1, characterized in that: include: The measurement module is used for measuring the furnace flame radiation spectrum signal of 200nm to 1750nm by spectrometer; A calculation module is used to select a wavelength range to calculate the flame temperature T and flame blackness ε as the first characteristic parameters of the furnace flame combustion based on the Planck quantitative and two-color gray body characteristic judgment principles; Calculate the intensity ratio R of the Fe element radiation spectrum intensity to the average radiation intensity in the band range selected in step 2 as the second characteristic parameter of the furnace flame combustion; calculate the total radiation intensity D as the third characteristic parameter of the furnace flame combustion; The recognition module is used to select the data sets of the first characteristic parameter, the second characteristic parameter, and the third characteristic parameter as the input layer of the deep learning training model, and design the output state according to different oxygen concentrations; the three characteristic parameters of the spectral data are input into the trained deep learning training model to identify the combustion state.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, each step of the method for identifying the combustion state of a coal-fired power plant furnace based on spectrum as claimed in any one of claims 1 to 7 is implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the method for identifying the combustion state of a coal-fired power plant furnace based on spectrum as claimed in any one of claims 1 to 7 is implemented.

Images