JP2008180481A - Method and device for estimating gas concentration in coal-fired boiler - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method and device for estimating a gas concentration in a coal-fired boiler, capable of accurately estimating CO and NOx concentrations for various coal types, including a transition state. <P>SOLUTION: A neural network for obtaining information for a coal type used is established with a plant measurement value as input. Further, a neural network which learnt relations of process value with CO and NOx concentrations is established for each coal type. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a method and apparatus for estimating the concentration of gas components, particularly CO concentration and NOx concentration of a coal fired boiler which is one of the components of a thermal power plant, using a neural network.

  Coal-fired boilers use coal as fuel. In a thermal power plant having a coal-fired boiler, the concentrations of CO and NOx, which are environmental pollutants discharged from the boiler, must be set to a specified value or less. The production amount of CO and NOx is in a contradictory relationship. If the air (or oxygen) is excessive, the production amount of NOx increases. Conversely, if the air is insufficient, the production amount of CO increases. In recent coal-fired boilers, in order to reduce both CO and NOx and improve combustion efficiency, two-stage combustion is performed in which air is sent stepwise. In this combustion control, adjustment of the air amount, selection of the burner combustion pattern, etc. are performed to create an optimal combustion state. Adjustments for optimization of combustion control, for example, adjustment of control gain, planning of burner combustion pattern, etc. are performed offline beforehand.

  However, the combustion conditions adjusted in advance in this way are optimized for typical operation modes and are only rough operation plans. On the other hand, in response to operating conditions such as load requirement values, atmospheric conditions, and fuel characteristics that change from moment to moment, optimization of plant operation, that is, combustion efficiency is reduced while keeping CO and NOx concentrations within an allowable range. Maximization is required from an economic point of view.

  In order to realize this, it is necessary to be able to simulate on-line changes in CO and NOx concentrations with respect to control demand changes based on current operating conditions. That is, there is a function that evaluates the combustion efficiency and the amount of environmental load substances when the control demand is changed with respect to the current operating conditions obtained from the measurement data, and searches for the optimal control point in view of both points. is necessary.

  As a method for estimating the CO and NOx concentrations, there is a method of modeling the relationship between each operating condition and the change tendency of the gas concentration using actual machine data based on a learning type algorithm such as a neural network. In the case of this method, if there is actual machine data, an estimation model of the CO and NOx concentration can be easily created according to the characteristics of the actual machine, so this is one of the methods often applied. In addition, there is a method for estimating the CO and NOx concentration using a model (for example, see Patent Document 1).

JP 2006-132902 A (summary)

  However, the CO and NOx concentrations discharged from the boiler are different depending on the type of coal (hereinafter referred to as coal type) even under the same combustion conditions. As the operation mode of the power plant, the coal type sent to the mill (pulverizer) may change even during operation. In this case, the coal type before and after the change is mixed for a while after the coal type is changed. become. That is, there is a transitional state when switching from coal type A to coal type B.

  Since optimization of operation is required even under such operation conditions, the CO and NOx concentration estimation model is highly estimated not only for all types of coal, but also for conditions where multiple types of coal are mixed. Accuracy is required. For this reason, the neural network for estimating the CO and NOx concentrations corresponds to all coal types, and also considers a transitional state in which the coal type is changed, and a state in which at least two types of coal are mixed But you have to deal with it.

  However, there is a limit to making the neural network accurately correspond to all of these various states, and as a result, there is a problem that the estimation accuracy of the model is lowered.

  Furthermore, there is a problem that it is difficult to measure the change of coal types online during plant operation. The only information that can be obtained regarding the timing of the coal type change is analogy that another type of coal was stored in the coal storage facility and would soon be changed. As described above, it is difficult not only to measure the mixing ratio of coal types in a transitional state in which a plurality of coal types are mixed, but also to determine when switching to a new coal type.

  For this reason, when estimating CO and NOx concentration on-line using real machine data, there was a problem that it was difficult to obtain on-line information on the type of coal of fuel that was input.

  An object of the present invention is to provide a method and an apparatus for estimating the gas concentration of a coal-fired boiler capable of accurately estimating the CO and NOx concentrations including various transient states including various transient types.

  The present invention is a method for estimating the concentration of gas components discharged from a coal-fired boiler using a neural network, and is estimated using a plurality of neural networks prepared for each coal type and a neural network for determining the type of coal. It is characterized by doing.

  Further, the present invention is an apparatus for estimating the concentration of gas components discharged from a coal-fired boiler using a neural network, a plurality of neural networks prepared for each coal type, a neural net for determining the coal type ratio, An adder that weights and averages estimated values of gas concentrations output by a plurality of neural networks prepared for each coal type using the mixture ratio estimated by the neural net for determining the coal type ratio is provided. To do.

  In the present invention, a neural network is prepared for each coal type, and modeling for CO and NOx concentration estimation is performed only for that coal type to improve estimation accuracy. Furthermore, in consideration of the mixed state of coal types, a neural network for estimating the mixing ratio of each currently used coal type is also prepared. Based on the coal type ratio estimated by the neural network, the estimated values of the CO and NOx concentrations output by the neural network for each coal type are weighted and averaged to obtain a final estimated value.

  In the present invention, the neural network may be modeled specifically for each coal type, so that the estimation accuracy of each neural network can be improved, and the final CO and NOx concentration estimation accuracy can also be improved. Moreover, since the information on the coal type that is difficult to obtain from the measured values can be inferred by the neural network, accurate CO and NOx concentrations can be obtained even when the coal type is changed.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a diagram showing a configuration of an online gas concentration estimation apparatus 1 according to an embodiment of the present invention. The online gas concentration estimation apparatus 1 is a computer. The process value database (process value DB) 11 stores trend data of plant measurement values used for neural network estimation processing. The operation condition setting unit 13 sets input data corresponding to the operation condition in the gas concentration estimation unit 12 in the neural network estimation process. The gas concentration estimation unit 12 stores a neural network for estimating the CO and NOx concentrations. The output of the gas concentration estimator 12 is taken into a coal fired boiler control system and used for optimization of combustion control. A part of the output of the gas concentration estimation unit 12 is also taken into the display control unit 14 and displayed on the display device 2.

  FIG. 2 shows the configuration of the gas concentration estimation unit 12. The gas concentration estimation unit 12 includes a plurality of neural networks. As described above, a neural network specialized for each coal type is prepared in the present invention.

  In FIG. 2, a coal type A gas concentration estimation neural network 22 and a coal type B gas concentration estimation neural network 23 are neural networks corresponding to the respective coal types, and estimate CO and NOx concentrations according to input data. Output the value.

  The coal type ratio determining neural network 21 has a function of estimating the mixture ratio of coal types based on input data. The output of the coal type ratio determination neural network 21 corresponds to each neural network prepared for each of the above-described coal types, and outputs the mixture ratio of the respective coal types. This output value is set as a weighting coefficient in the adder 41. That is, the neural network for gas concentration estimation estimates the CO and NOx concentrations on the condition of each coal type, and the adder 41 performs weighted average processing according to the mixture ratio.

  The above is a rough description of the entire gas concentration estimation apparatus according to the present embodiment. Next, details of each neural network used in this apparatus will be described.

  First, the coal type ratio determination neural network 21 will be described with reference to FIG.

  The coal type ratio determination neural network uses the plant measurement data stored in the process value DB to determine the coal type of the pulverized coal currently used as fuel, including the mixed state. The output unit 31 of the coal type ratio determination neural network outputs the mixture ratios for the coal types A, B and C as numerical values. In FIG. 3, each output unit outputs a value of 0.80, 0.15, and 0.00, but this value is not used as it is as a mixture ratio, but the total is 1.0. Perform standardization. In this example, 0.84, 0.16, and 0.00 are obtained by the standardization process, and the current input fuel is mixed with 84% for coal type A and 16% for coal type B. It is determined that there is.

  In the input of the neural net 21 for determining the coal type ratio, a process value in which an influence due to the difference of the coal type appears is set. For example, CO concentration, NOx concentration, mill motor power, boiler absorption heat (calculated from the steam state at the boiler inlet / outlet), primary air flow (primary air fan power, etc. if not measured), secondary air flow, two-stage combustion Air flow ratio, load, fuel flow rate, etc.

  Next, a method for learning a coal type ratio determination neural network for realizing the above processing will be described.

  Actual data is used for the learning process of the neural net for determining the coal type ratio, but in addition to the process value that is the input value, the value of the mixture ratio of the coal type that is the output value is also required. However, as described above, since there is no measurement value for the coal type, actual measurement data cannot be used for output. For this reason, the neural net is trained by using the measured data in the period when the coal type is stored in the coal storage facility and sufficient time has passed and the current input fuel is expected to be only the same coal type. . For example, when it is predicted that the input fuel is only coal type A, the neural network learning sets 1.0 for the output unit corresponding to coal type A and 0.0 for the other output units. Learn the relationship with process values. Similarly, it learns also about coal type B and coal type C.

  The neural net for determining the coal type ratio learned by such a method outputs 1.0 for only the unit corresponding to the coal type for the process value when the input fuel is the same coal type, Other units output 0.0. Moreover, even when a plurality of coal types are mixed, the output value of the unit is increased according to the ratio of each coal type due to the interpolation function of the neural network.

  In this way, it is possible to evaluate information on the coal type of the input fuel online based on the process value.

  Next, a gas concentration estimation neural network will be described.

  The neural network for gas concentration estimation receives a process value corresponding to the combustion condition and outputs the CO and NOx concentrations at that time. The relationship between input and output is learned for each coal type.

  In the learning process, only the measured values of the plant are used. However, similar to the learning of the above-described coal type ratio determination neural network 21, learning is performed using data of a period in which the fuel is expected to be only the same coal type. In this way, a neural network for each coal type is constructed. Process values related to CO and NOx concentration as the input of the neural network, such as primary air flow rate (primary air fan power if non-measurement), secondary air flow rate, air flow ratio for two-stage combustion, load, Set fuel flow, burner operating conditions, atmospheric conditions (temperature, pressure, humidity), etc. The process value used as an input is common in the gas concentration estimation neural network prepared for each coal type.

  In the estimation process, not only plant measurement values but also signals from a coal fired boiler control system are used. This is because the control system takes in estimated values of CO and NOx concentrations when the control demand is changed in order to perform control optimization online. For example, the CO and NOx concentrations when the combustion conditions such as the air flow rate are changed with respect to the current operating conditions are estimated, and the optimum control point is searched from these concentrations and the estimated value of the plant efficiency.

  In the combustion control optimization process by the control system, the operating condition setting unit 13 shown in FIG. 1 changes the input value of the neural network according to the control demand. That is, among the process values set for the input of the neural network, the process value corresponding to the control condition is changed according to a command from the control system. For example, when analyzing the change tendency of the CO and NOx concentration when the primary air flow rate is changed, the operating condition setting unit 13 updates the value of the primary air flow rate among the inputs of the neural network.

  The data output from the gas concentration estimation unit 12 through the above processing is taken into the display control unit 14 and output to the display device 2. FIG. 4 is a display example of the display device 2. FIG. 4 shows the CO, NOx concentration, and the ratio of the coal type used as fuel when the coal type is switched as a trend. Regarding CO and NOx concentration, the emission ratio of each coal type is also trending.

  By the processing described above, the CO and NOx concentration estimation processing for optimizing the combustion control by the control system is realized.

It is a block diagram of the online gas concentration estimation apparatus which concerns on this invention. It is a block diagram of the gas concentration estimation part in the gas concentration estimation apparatus of this invention. It is a block diagram of the neural net | network for charcoal kind ratio determination in the gas concentration estimation apparatus of this invention. It is the figure which showed the example of a display of a display apparatus.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Online gas concentration estimation apparatus, 2 ... Display apparatus, 11 ... Process value database, 12 ... Gas concentration estimation part, 13 ... Operating condition setting part, 14 ... Display control part, 21 ... Neural network for charcoal type ratio determination, 22 A gas concentration estimation neural network for coal type A, 23 ... a gas concentration estimation neural network for coal type B, 31 ... an output unit of a coal type ratio determination neural network, 41 ... an adder.

Claims (7)

  A method for estimating the concentration of gas components discharged from a coal-fired boiler using a neural network, and estimating using a plurality of neural networks prepared for each type of coal and a neural network for determining the type of coal A method for estimating gas concentration in a coal fired boiler.   The method for estimating gas concentration in a coal fired boiler according to claim 1, wherein a mixing ratio of coal types of coal introduced as fuel is estimated using the neural network for determining the ratio of coal types, and the mixing ratio is used. In addition, the gas concentration estimation of the coal-fired boiler is characterized in that the estimated value of the gas concentration output from the plurality of neural networks prepared for each coal type is weighted and the concentration of the gas component at the time of changing the coal type is estimated. Method.   3. The method for estimating gas concentration of a coal fired boiler according to claim 2, wherein the learning of the neural net for determining the coal type ratio uses data of a period in which the coal type of coal input as fuel is the same type. , A method for estimating the gas concentration of a coal-fired boiler, wherein the tendency of the coal type is learned.   A device for estimating the concentration of gas components discharged from a coal-fired boiler using a neural network, a plurality of neural networks prepared for each coal type, a neural network for determining a coal type proportion, and the coal type proportion A coal-fired boiler comprising an adder that performs weighted averaging of estimated values of gas concentrations output from a plurality of neural networks prepared for each coal type using a mixing ratio estimated by a determination neural network Gas concentration estimation device.   5. The coal fired boiler gas concentration estimation apparatus according to claim 4, wherein a mixing ratio of coal types of coal introduced as fuel is estimated using the neural network for determining the ratio of coal types, and the mixing ratio is used. The estimated value of the gas concentration output from the plurality of neural networks prepared for each of the coal types by the adder is weighted and averaged to estimate the concentration of the gas component when the coal type is changed Gas concentration estimation device for fired boiler.   In the coal fired boiler gas concentration estimation apparatus according to claim 5, in the learning of the neural net for determining the coal type ratio, data of a period in which the coal types of coal supplied as fuel are the same type are used. An apparatus for estimating the gas concentration of a coal fired boiler, characterized by learning the tendency of the coal type.   The gas concentration estimation device for a coal fired boiler according to claim 4, further comprising a display device for displaying an estimated value of a coal type mixing ratio of coal introduced as fuel or an estimated value of a component of gas concentration. An apparatus for estimating the gas concentration of a coal-fired boiler.

Images