SE501784C2 - Procedure for controlling a combustion process - Google Patents

Abstract

A method for controlling a combustion process in a combustion plant comprises sensing the content of NOx in the flue gases and affecting, on the basis thereof, the combustion by controlling a parameter to obtain the desired NOx content, wherein interference is caused in said at least one combustion process parameter, preferably the excess of O2 in the flue gases and/or the flow rate of a chemical supplied, e.g. urea, and generating a signal which gives successive values of the parameter and which is low-pass-filtered; measuring the NOx content and the O2 content of the flue gases by means of transducers so as to generate an NOx content signal and an O2 content signal; performing the NOx and O2 content measurements such that the signals obtained are referable to flue-gas measuring points located at any rate in the immediate vicinity of one another; low-pass-filtering the NOx and O2 content signals obtained and sampling these low-pass-filtered signals to obtain NOx and O2 content values successive in time; calculating zero-%-O2-compensated NOx comp content values by using the O2 content values obtained and then utilising O2 content values which are related in time to the corresponding NOx content values; high-pass-filtering the NOx comp content values and the successive parameter values; calculating, on the basis of the high-pass-filtered values, the relationship between NOx comp and the parameter; and controlling the parameter on the basis thereof.

Description

5Û1 784 10 15 20 25 30 35 Object of the invention An object of the present invention is to provide a method of the kind mentioned in the introduction, whereby the relationship between a combustion process parameter and NOX can be determined quickly and with good accuracy without operating mode needs to be changed and without other combustion process parameters have a significant impact.

Another purpose is to designate a procedure that is useful in real time and thus provides the opportunity to control on-line, ie can be integrated into the boiler's control system five.

Summary of the invention The above objects are achieved according to the invention with a process having the claims set out in the appended claims the features.

The method thus comprises accomplishing disturbances in the parameter in question and generates a signal which contain successive values of the parameter and by means sensors measure the varying NOX and O2 levels in the smoke the gases, so that a NOX content signal and an O 2 content signal alstras. If the parameter is the excess oxygen in the flue gases, the 02 content signal will also be the parameter signal.

If the parameter is a flow of a chemical fed into the boiler potassium, such as urea, senses the flow to produce a parameter signal.

The obtained NOX and O2 content signals (and in where applicable, also other utilized parameter signal) low-pass filtered and sampled, preferably in conjunction with A-D conversion, to obtain chronologically successive NOX and O2 content values (and, where applicable, other utilized parameter signal).

In order to provide good accuracy, the NOx content value is recalculated so that NOX composite content values were compensated to zero% 02 obtained, using associated 02 content values.

It is essential in this case that it is a question of values such as refers to processes that occur in any case mainly simultaneously 10 15 20 25 30 35 501 784 3 poem in the process. In order to ensure the right timeliness relation, the previously mentioned N0x- and The 02 content measurements so that the signals obtained are attributable to flue gas measuring points, which are located in any case in the immediate vicinity of each other, plus transport times between the measuring points and the sensors and the time constants of the sensors are also taken into account in the interconnection of related NOX and O2 content values in connection with compensation calculations na. In doing so, however, one should make sure that it does not exist a prompt relationship between the NOX and 02 content values.

In order to eliminate elements of trends, transients, etc the obtained NOx comp content values are high-pass filtered and the successive parameter values (which are thus the sampled ones) 02 content values, when the parameter is the excess oxygen in the smoke the gases, and sampled flow values, when the parameter is one chemical flow), after which the relationship between N0xkomp and the parameter in question can be calculated. This is done preferably using recursive process identification, whereby the derivative of the relationship can be determined by response test. Since this can be done on-line and in real-time, there are opportunities to quickly detect process changes and control the process with its guidance, too automatically by integration into the boiler control system.

Utilization of recursive process identification has proved to mean that there is no requirement for absolute temporal process concurrency for the parameter values or NOX values, which are linked to each other below the connection calculation. This is of course a clear advantage then the parameter is a chemical flow, which is sensed on one other than the place where 02 and NOX are sensed.

Then the parameter is the excess oxygen 02 in the flue gases however, time compensation must be provided if necessary such a way that the respective O 2 content value is linked to one In any case, the Nüxkomp content value is not in time after the latter, ie it must be ensured that 02 is seen control NOK and not the other way around. 501 784 10 15 20 25 30 35 4 The controlled disturbances that are preferably used according to the invention is of the type PRBS (Pseudo Random Binary Sequence). This has been shown to give higher frequency components. of the values used for the relationship calculation, which means that the identification of the connection can implemented in a shorter time and with higher accuracy. The algorithm used for the connection calculation needs thus less data for the calculations. The results of the calculations are obtained faster and a sudden change of the process can be detected and remedied with shorter time delay.

The recursive process identification utilizes one large number of successive, weighted values, with the latest values are given the highest weight. The time intervals between the sample is suitably of the order of seconds or seconds. customers, and the algorithm-driven model of process identification calculation can be updated after a few new samples, step by step after each new sample. It is understood that this provides possibility of determining the relationship's derivatives with very short intervals, for example of the size of tens of seconds so that the process can be followed almost continuously ligt.

According to an advantageous embodiment of the method according to the invention, when the current parameter is a chemical flow, that between the flow and N0x and of the cost of the chemical and of cost in the form of NOX fee quickly and efficiently direct the chemical flow towards a flow that provides a minimization having regard to the estimated of the total cost.

The invention has in practice been shown to provide very accurate accurate measurement results, which make it possible to with very good prediction induce parameter changes to affect The NOK content in the flue gases, and this without the process in otherwise disturbed.

The invention will be described in the following further by way of exemplary embodiments with reference to display to attached drawing. 10 15 20 25 30 35 501 784 5 Brief description of the drawing Fig. 1 is a diagram showing how one combustion system comprising a boiler with fluidized bed modified to incorporate an embodiment of present invention; Fig. 2 is a diagram generally illustrating the the treatment in connection with the embodiment according to Fig. 1; Fig. 3 is a diagram illustrating utilized pro- cess identification; and Fig. 4 is a diagram illustrating utilization of another embodiment of the invention in terms of cost minimization by urea injection in order to reduce NOX.

Description of embodiments The incineration plant schematically shown in Fig. 1 which has been modified to include a embodiment of the invention, comprises a boiler 1 with a circulating fluidized bed, which is supplied with primary air 3 and secondary air 5 from a total air flow 7. The flue gases passes an electric filter 9, before passing out through one chimney 11. Before the electric filter 9, a meter for 02 which gives a value of combustion-02. This issue used in the boiler control, as will be described further below.

Two flue gas sensors 13, 15 are connected to the chimney via a common suction line 17. The sensor 13 gives a signal corresponding to the excess oxygen OZM (Environment 02) in the flue gases and the sensor 15 gives a signal corresponding to the content of N03 in the flue gases. The sensor signals are low-pass filtered and analog-digital is converted, during simultaneous sampling, into one block 19, the corresponding output signals of which are fed to a computer 21 for further treatment, as will also be written in more detail below. 501 784 10 15 20 25 30 35 6 The sensor signals are delayed in time partly due to the transport time through the suction line 17, partly due to the respective res own dynamics, which can be described with a time constant.

In the current example, the time constant has been selected as it time it takes for the output signal to change from a zero level to a certain level, eg about 63% of the final level.

The one illustrated in Fig. 1, in accordance with an embodiment modified pan of the invention has a control system, which is designed so that the air to the boiler is determined to its setpoint by the plant's vapor pressure regulator.

A low vapor pressure calls for a higher total air flow pannan. The distribution of air between primary and secondary air then takes place according to quotas entered in the control system.

The measured combustion O 2 content is then used as setpoint to affect the fuel flow to the boiler.

The air control system is supplemented with a PRBS feature. It is easy for the secondary flow setpoint to affect caused by the interference signal. The new setpoint will be equal to that old setpoint * (1 + k * PRBS). Because PRBS is either zero or high, the new setpoint means that secondary the air is made to increase temporarily when the PRBS signal determines this.

So that the control system now does not counteract the desired one the secondary air flow increase preferably takes place simultaneously influence of the total air flow regulator. This mode is called mode of interference 1.

The interference signal now also affects the total air regulation in such a way that the actual value of the total air regulator is reduced approximately by the amount of air taken out via the there air. The sudden reduction in the temperature of the primary air ficial actual value causes the total air regulator to increase its progress. In this way, the primary air flow becomes large seen unaffected by the disturbance while the secondary air flow varies.

A second mode, interference mode 0, means the same thing as disturbance mode 1, but with the difference that the total air is unaffected by the disturbances. This means that the ratio between 10 15 20 25 30 35 501 784 7 secondary and primary air vary with the disturbances.

The disturbances that have been used in the project are of the type PRBS. This is a standard type of interference signal that varies seemingly randomly between two levels. The signal has appropriate statistical properties, as it contains a wide range of frequency components.

The highest frequency component is adjusted so that it have time to influence the process. In this application, they consider it appropriate to have a shorter period of time of order minute, especially 60 seconds. In addition, more it a set of frequency components that are slow more than this fastest period. Frequency spectrum for sig- the channel is a line spectrum with a relatively constant height in front to the cut-off frequency.

In addition, the PRBS signal has such properties that it is as long time high as low considered for a longer time.

The PRBS signal can be easily introduced into the controllable control systems as a current signal in the appropriate inter- vall.

The idea of using controlled disturbances is that provide higher frequency components of 02 and thus also at NOX. This means that it is possible to make an shortening the relationship between these in a shorter time and with higher accuracy.

The technology used with PRBS interference has shown do not otherwise affect the operation of the boiler. Disorders can be connected for long periods of the day, even at night. time and when the plant is running unmanned.

The preferred signal processing that takes place outbound from the OZM and NOx signals obtained from sensors 13 and 15 The figures are illustrated in Figures 2 and 3.

The signals are filtered using hardware as one first step. This type of filter is an LP filter. Where- after, the signal is sampled via the system's A / D conversion lare. Sampling is done with 1 Hz and low-pass filtering with 0.4 Hz switching frequency. 501 784 10 15 20 25 30 35 8 The signals thus obtained NOX and OZM now represents the levels in ppm and% respectively. To not get with the effect of dilution, nox is now compensated to NOxkomp whereby time differences between the measured must be taken into account. Thus, all NOx aggregates are recalculated pile to the corresponding value at 0% oxygen according to the known connection: N0xcomp (t) = NOx (t) / l. - 02M (t-dt) / 21.) where t = 1, 2, 3 ... (time in seconds) dt = time compensation.

The compensation must be done in such a way that not there is some prompt connection between OZM and NOxkomp.

Nor may there be temporal events occurs at N0xkomp before 02M.

The compensation can be exemplified in the following way.

Assume that NOX and OZM are stored as 1 second sample in each vector in a file. These figures are available now the dynamics of the sensors. Assume that the time delay of the OZM sensor (Transport time + Time constant) = DT02 is for example 3 seconds shorter than DTNOX (equivalent to NOK). Then shall the rule given below is used for calculation of N0xkomp.

N0x sample t = 1, 2, 3 .... s X X X X X X X X X X X X X X X X X X Lv X X X X X X X X X X X X X X X X X X 02M sample t = 1, 2, 3 .... s X X X X X X X X X X X X X X X X X X Noxkomp t - 1, 2, 3, .... s NOxcomp (t) - NOx (t) / (1 - 02M (t + 2) / 21 for t = 1, 2, 3 .... s X 10 15 20 25 30 35 501 784 9 Now that 02M is ahead of N0xkomp in terms of time of the properties of the donors, the vectors O2M (t) and NO comp (t) is used directly as input signal and output signal, respectively for later calculation of the transfer function (connection) and the derivative dNOxkomp / d02M calculated by step test.

Should the N0x sensor be the fastest, ie DTNOX be e.g. 2 seconds shorter than DT02, the order will be it reverse in time. Then instead N0xcomp (t) = N0x (t) / (1 - 02M (t-2) / 21) for t - 3, 4, 5 ...

In this case, additional time compensation is required prior to the latter calculation so that NOxkomp compensates race in time. The input signal in this case becomes 02M (t) and the output the signal N0xkomp (t + 2). With these new vectors, the tification and step test are done just as above.

Should the sensors be just as fast, however a time shift with at least one step take place, so as not to there is some prompt connection between Nøxkomp and OZM, thus: NOx comp (t) I NOx (t) / (1 - 02M (t + 1) / 21).

In this case, no additional time compensation is required. for the later calculation of the relationship.

The signals obtained in this way must now be -filtered to eliminate elements of trends and trans- sienter. Trends or low-frequency characteristics of sig- namely, have a detrimental effect on the calculations. Same thing applies to transients that occur. These contain no important information for the subsequent identity tification.

The signals thus processed have now oscillated which are mainly around an average value. Sig- the channels can now be used for the recursive identification. 501 784 10 15 20 25 30 35 10 In the calculations, a forgetfulness factor is taken into account. This in- implies that the most recent sampled values are most important for the calculations while previously obtained samples are weighted less and less. Transferred to a time window so means the oblivion factor to a large number of samples, eg 2000 samples form the basis of the calculations. It in this way the resulting model is updated after each new sample.

This means every second in the preferred case.

The preferred method used to calculate it relationship between the oxygen content of the flue gases and NOX uses process identification. The method is easiest explained with reference to Fig. 3. The process is perceived as a black box with an input signal and an output signal. Signal The samples are sampled by means of a computer, in which a general model is programmed. The output of the model is compared with the process sense output. If the difference is zero, the model is and the process in accordance with each other. A deviation leads immediately to the correction of the model parameters.

It should be noted that the process and the model are the same input signal.

The model thus obtained can then be used to determine the dynamics of the process. Step response test can be carried out at appropriate time intervals and thus can the derivative is calculated. All this happens without the process is disturbed.

A significant advantage of process identification is that it can be made into a continuous process. It means that the coupling between 02 and NOX is calculated on-line and in real time with recursive methods. Changes that occur to due to variations in fuel chemistry and / or others combustion technical changes can thus be detected immediately and appropriate measures are taken with a view to: for example, keep NOX within prescribed values.

For example, the control system's 02 controller can be affected under the guidance of the derivative for the relationship between 02M and NOx comp in such a way that the prevailing NO3 content in the flue gases change in a controlled manner in the desired 10 15 20 25 30 35 B01 784 11 direction.

The preferred model structure used for pro- cess identification is called ARX. The analysis work can be done using the commercially available computer software MATLAB.

Details of the process identification can be found in written in Lennart Ljung, System identification toolbox for use with Mathlab, Users guide April 6, 1988, The Math- Works, Inc.

Fig. 4 illustrates how the present invention an advantageous way can be used for optimization purposes in in connection with the injection of a chemical, in particular urea, in purpose to reduce the NOX content in flue gases emitted. Before- the assumptions are that urea injection can reduce the NOK content, thereby reducing environmental tax A for NOX, but that in case the cost B for the urea injection simultaneously increases with increased urea flow U. The total cost A + B has one minimum corresponding to a certain optimal urea flow Uopt, which one thus wants to be able to decide and steer towards. For this requires knowledge of the relationship between NO3 and urea flow U. This relationship can be calculated according to the invention, in particular in the manner described by reference to Figures 1 to 3, however, filtered and sampled urea flow values after high-pass filtering replaces the OZM values in the actual connection calculation ningen. The PRBS disorder must of course also relate to the urea flow and not the air flow.

Claims (11)

501 784 10 15 20 25 30 35 12 PATENT REQUIREMENTS A method for controlling a combustion process in an incineration plant, for example a cogeneration plant, in which the content of NOX in flue gases generated during combustion is sensed and, consequently, the combustion is affected by controlling at least one parameter in order to maintain the NOX. the content at the desired level, characterized in that a) causes disturbances in the at least one combustion process parameter, preferably the excess of O 2 in the flue gases and / or the flow of a chemical fed into the combustion chamber, e.g. urea, and generates a signal representing successive values of the combustion process parameter, b) by means of sensors measure the NOX content and the O 2 content in the flue gases, whereby a NOX content signal and an O 2 content signal are generated, c) perform the NOX and O 2 content measurements so that the obtained signals are attributable to flue gas measuring points, which are in any case in the immediate vicinity of each other, d) low-pass filters the obtained NOX and O2 content signals, e) samples the low-pass filtered NOx and Oz content signals to obtain temporally successive NOX and O content values, f) calculates to zero% 02 compensated NOx compost values using the obtained O2 content values, 2 g) uses 02 content values for compensation calculation which are time-related to the corresponding NOx content values, h) high-pass filters, preferably in terms of software in a computer, the obtained NOx composite values and said successive parameter values, so that elements of trends, transients, etc. are eliminated, i) outgoing from the high-pass filtered values calculates the relationship between NOxkomp and said parameter, and j) with the guidance of said relationship performs the control of the parameter. A method according to claim 1, wherein the relationship between NOxkomp and said parameters is calculated on-line by recursive process identification, wherein the derivative of the relationship is determined by step response test. Method according to Claim 2, in that a large number of successive values are used for the recursive process identification - k ä n n e t e c k - whereby recently obtained values are given the highest weight while previously obtained values are given less and less weight the older they are. Method according to Claim 2 or 3, in that the model identification of the process identification is updated after a few new samples, preferably after each new sample. Method according to one of Claims 2 to 4, characterized in that the time intervals between the samples are of the order of seconds. Method according to one of the preceding claims, characterized in that the sampling of the low-pass filtered signals takes place by means of A-D conversion. Method according to one of the preceding claims, characterized in that the disturbances are caused by means of a so-called PRBS disturbance signal. Method according to one of the preceding claims, characterized in that in step g) the temporal relation comprises considering the time delay between the measuring point and the transducer and the transient time constant of the transducers. Method according to any one of the preceding claims, wherein the controlled parameter is the O 2 content in the flue gases, characterized in that said successive parameter values are said O 2 content values temporally related to the Nøxcomph content values. A method according to claim 9, wherein the combustion process takes place in a fluidized bed plant, preferably a circulating fluidized bed, wherein air is supplied in the form of primary air and secondary air, characteristic - that the combustion process disturbances are effected by influencing the setpoint value of the secondary air flow under the simultaneous influence of the total air regulation, so that the primary air flow is at least substantially unaffected. ll. Method according to one of Claims 1 to 9, in which the combustion process parameter is the flow of a chemical fed into the combustion chamber, in particular urea, characterized in that, taking into account the calculated relationship between N0x comp and the sampled chemical flow, the cost for the chemical and of cost in the form of NOX fee, the chemical flow directs towards a flow that minimizes the total cost.