EP1850069B1 - Method and Control Loop for Controlling a Combustion Process - Google Patents

Abstract

A method includes operating in a setpoint control mode and occasionally changing over to operate in disturbance control mode. Each operation mode includes selecting actions to control adjustment devices (9) that supply air and material to the combustion by evaluation of state variables. The setpoint control mode is operated to achieve an optimal setpoint for the state variables, stability of the combustion, or any combination. The disturbance control mode is operated to approach a state in which the state variables deviates in a targeted manner within predetermined limits from the setpoint. Independent claims are included for the following: (1) control loop for regulating combustion in an installation; (2) computer readable medium having computer executable instructions.

Description

The invention relates to a method for controlling a combustion process, with the features of the preamble of claim 1 and a control loop having the features of the preamble of claim 8.

In a known method of this type, either the setpoint values of the state variables are automatically regulated by comparing the actual values with the setpoint values and, if appropriate, performing actions, normally control interventions, or regulating the stability of the combustion process, by performing actions only in small scale.

The present invention has for its object to improve a method of the type mentioned. This object is achieved by a method having the features of claim 1 and by a control circuit having the features of claim 8. Further advantageous embodiments are the subject of the dependent claims.

By temporarily switching from the lead control to a fault control, according to which actions are selected to start states of the system in the plant in which the state variables within specified limits deliberately deviate from the optimum target value, additional information is obtained, the one allow for improved regulation. In particular, it can be avoided that the state of the system remains at a local minimum. Such actions would neither regulate to setpoint values which are currently targeting the set point, nor - as they seek greater changes in the state - control the stability of the combustion process be performed. Combinations of the two control cases are possible in the way of compromises.

Information can be procured regularly and in as wide a range as possible in a regular system of disruptions. In addition (or alternatively, if necessary), in an extraordinary disturbance regulation, certain regions of states may be more intensively tested.

The invention can be used in various stationary thermodynamic systems, in particular power plants, waste incineration plants and cement works.

Fig. 1

eine schematische Darstellung des zeitlichen Verlaufs einer Zustandsvariablen s(t) bis zu einem Zeitpunkt t₀ und die Vorhersagen für den weiteren Verlauf,

Fig. 2

eine schematische Darstellung des tatsächlichen zeitlichen Verlaufs einer Zustandsvariablen s(t) im Vergleich zu den zum Zeitpunkt t₀ getroffenen Vorhersagen,

Fig. 3

eine schematische Darstellung des zeitlichen Verlaufs einer Zustandsvariablen s(t) mit einer Aktion aⁱ zum Zeitpunkt t₀ und

Fig. 4

eine schematische Darstellung eine Anlage.

In the following the invention with reference to an embodiment shown in the drawing is explained in more detail. Show it

Fig. 1

a schematic representation of the time course of a state variable s (t) up to a time t ₀ and the predictions for the further course,

Fig. 2

a schematic representation of the actual time course of a state variable s (t) compared to the predictions made at time t ₀ ,

Fig. 3

a schematic representation of the time course of a state variable s (t) with an action a ⁱ at time t ₀ and

Fig. 4

a schematic representation of a plant.

A plant 1, for example a coal, oil or gas-fired power plant, a waste incineration plant or a cement plant, comprises a furnace 3, which is also to be understood as a grate, at least one observation device 5 which encloses the interior of the furnace 3 (or the grate). can capture, preferably further Sensors 7, at least one adjusting device 9, and a computer 11, to which the observation device (s) 5, further sensors 7 and adjusting device (s) 9 are connected.

The furnace 3 is fuel or other material to be reacted, referred to as Good G for short, for example, coal, oil, gas, refuse, lime or the like, and primary air (or oxygen) and secondary air (or oxygen), short as air L referred, supplied, this supply is controlled by the controllable by the computer 11 actuators 9. In the furnace 3, a combustion process takes place. The resulting flame body F (and, where appropriate, emissions of the walls of the furnace 3) is continuously detected by the observation devices 5. The observation devices 5 each include adjacent to a wall of the furnace 3 penetrating optical access, such as a lance or in the EP 1 621 813 A1 disclosed apparatus, nor a camera or the like, which operates in the optical region or adjacent regions of electromagnetic waves. Preferred is a temporally, spatially and spectrally high-resolution camera, as for example in the WO 02/070953 A1 is described.

The images of the flame body F (and the possible emissions of the walls of the furnace 3) are evaluated in the computer 11, for example, according to an eigenvalue method, which in the WO 2004/018940 A1 is described. In the EP 1 524 470 A1 A method is described how to obtain a few characteristic values from a spectrum. The data obtained from the images of the flame body F and the data of the other sensors 7, which measure, for example, the supply of the goods G and the air L, pollutant concentrations in the exhaust gases or the concentration of free lime (FCAO), as state variables s (t) which describe (time-dependent) the state of the system in Appendix 1 in general and the combustion process in particular, and are to be regarded as a vector.

Through the furnace 3 as (control) route, the observation device (s) 5 and the other sensors 7, the computer 11 and the adjusting devices 9, a control loop is defined. A conventional control loop can also be provided only with furnace 3, sensors 7, computer 11 and adjusting devices 9 and without the monitoring device (s) 5, the control of which takes into account only a few state variables s _t (ie is low-dimensional) and then by the inclusion of the observation device (en) 5 is optimized. The system in Appendix 1, for example, can be regulated to specific setpoint values or to a stable process (ie a quiet, quasi-stationary operation of system 1). In both cases, the state described by the actual values of the state variables s (t) is evaluated and, if appropriate, suitable positioning actions (control actions), briefly referred to as actions a ⁱ , are selected, which are to be executed by the actuating devices 9. In addition to the supply of good G and air L further activities of adjusting devices 9 and optionally also a sampling can be an action a ⁱ in the sense of the invention. Also disorders can be treated as unwanted actions a ⁱ . There are conceivable combinations of the two aforementioned regulatory cases conceivable, which then represent compromises.

The evaluation of the state and the selection of suitable actions a ⁱ can be carried out, for example, according to a method as described in the WO 02/077527 A1 is described. At least one neural network is implemented in the computer 11, which stores as a process model the reactions of the states of the system to actions a ⁱ , ie the (non-linear) links between the values of the state variables s (t) at a time t = t ₀ and the actions a ⁱ then taken a on the one hand and the resulting values of the state variables s (t) at a later (ie later by a specific time interval) time t = t ₁ (or t ₁ , t ₂ , t ₃ ...) on the other hand, and indeed at as many times t in the past. In this sense, also disturbances as (unwanted) actions a ⁱ can be included in the process model. An evaluation of the situation independent of the process model, ie the stored links, designed in the manner of a simplified quality is, evaluated for a certain time t, the values of the state variables s (t) with respect to predetermined optimization goals r ^j , ie how close the state of the system at the time t is the optimal state. By evaluating a state predicted with the process model as a function of a specific action a ⁱ at a future point in time, the suitability of the particular action a ⁱ for approximating the optimization target r ^j can be determined.

Preferably, three (or four) process models are stored in the computer 11 (each in its own neural network), one each of which has a short (t ₁ -t ₀ ), one (or two) mean (t ₂ -t ₀ ) and contains a longer (t ₂ -t ₀ ) time interval learned links. Accordingly, short-term, medium-term and longer-term forecasts are possible. The said time intervals vary depending on the system 1 between about a few seconds and a few hours. The state variables s (t) should and can usually vary within certain limits, ie within an interval, for example between a lower limit s _l and an upper limit S _h around an optimum setpoint S _o . The values s _l , s _h and s _o can be time-dependent. The short-term, medium-term and longer-term predictions are used to determine the difference from s (t) to the optimal target value s _o (for example, the optimization target r ^j would be that s (t) - s _o = 0 or at least minimal) and the Compliance with these limits (thresholds s _l , s _h ) estimate and recognize the probable requirement of actions a '. The temporal evolution of a state variable s (t) up to the time t = t ₀ as well as the short-term prediction for t = t ₁ , the medium-term prediction for t = t ₂ and the long-term prediction for t = t ₃ are simplified in Fig. 1 shown. In Fig. 2 Then, the actual development of s (t) compared to the predictions is shown, with no action a ^{i for} better comparability.

In order to improve the accuracy, not only the process models are continuously supplemented by the actual developments of the state variables s (t) in response to actions a ', but there is a competition of several process models with regard to the quality of the predictions. For this purpose, alternative process models, for example with other topologies, are set up in the background and trained whose predictions are compared with the currently used process models to replace the latter if necessary, as for example in the EP 1 396 770 A1 is described.

According to the invention, a change from the normal control, the so-called command control, to a fault control (and back) is possible, in which the computer 11 emits test signals, so that - regardless of the optimization goals r ^j - different actions a 'are made to different Directions specifically adjacent (ie adjacent to each current state with respect to the state variable s (t) adjacent states) and preferably - by successively stringing the start - to reach even more remote states. However, in order not to obstruct or even disturb the operation of the system 1, only states within the limits (limit values s _l , s _h ) of the state variable s (t) are selected as destination, ie only actions are selected on the basis of which the state variables s (t) are expected to remain within their limits.

The computer 11 starts a "regular" fault control regularly, for example, about every seven days, but at the latest every four weeks, with as many as possible, preferably as far as possible evenly distributed within the limits states are approached. If the same problem often occurs in the control, the computer 11 starts an "extraordinary" fault control. Such a problem exists, for example, when the state variables s (t) often tend toward a limit (threshold values s _l , s _h ), ie the mean drifts, and / or frequent actions a 'are necessary to compensate for deviations, and / or other incompatibilities of the rules to target values (optimization goals r ^j ) and to a stable process occur. In the case of the extraordinary disturbance control, it is possible to approach states which are tuned to the triggering problem, for example, depending on the solution strategy, in the direction of the problem or chosen exactly the opposite.

In the drawings, for example, a case is shown that s (t) constantly moves above the optimum target value s _o ( Fig. 2 ) and also in the forecasts ( Fig. 1 ), in particular the longer-term prediction over the time interval t ₃ -t ₀ , tends towards the upper limit value s _h . In the context of the guidance regulation, an action a 'would be selected at t = t ₀ , or t = t _{1 ,} bringing s (t) closer to the optimum setpoint value s _o . By contrast, in the disturbance control, for example, an action a 'is also selected which brings s (t) to the lower limit s _l . This is with an action a 'at the time t = t ₀ in Fig. 3 shown.

LIST OF REFERENCE NUMBERS

1

investment

3

oven

5

observer

7

sensor

9

locking device

11

computer

a ⁱ

action

F

flame body

G

Well

L

air

r ^j

optimization goal

s (t)

state variable

S _h

upper limit

S _l

lower limit

S _o

optimal target value

t ₀ , t ₁ , t ₂ , t ₃

time

Claims (10)

1. A method for controlling a combustion process in an installation (1), in particular in a power-generating plant, a waste incinerator or a cement plant, in which, with air (L) being supplied, material (G) is converted by means of the combustion process with at least one flame body (F) being formed, wherein the state variables (s(t)) which describe the state of the system in the installation (1) are determined by using at least one observation device (5) that images the flame body (F) and also by using other sensors (7) and are evaluated in a computer (11), whereupon, if necessary, suitable actions (aⁱ) are selected in order to control adjustment devices (9) for at least the supply of material (G) and/or air (L), and wherein setpoint control is carried out to achieve setpoints (s_o) of the state variables (s(t)) and/or stability of the combustion process, characterized in that occasionally a changeover is made from setpoint control to disturbance control and, according to the latter, actions (aⁱ) are selected in order to approach states in the system in the installation (1) at which the state variables (s(t)) deviate in a targeted manner within predetermined limits (s_t, s_h).from the optimal setpoint (s_o)
2. A method according to Claim 1, characterized in that in the case of disturbance control, for each current state, states being adjacent with respect to the state variables s(t), are approached.
3. A method according to any of the preceding claims, characterized in that a changeover is regularly made from setpoint control to ordinary disturbance control and back again.
4. A method according to Claim 3, characterized in that in the case of ordinary disturbance control, as far as possible uniformly distributed states are approached within the predetermined limits (s₁, s_h).
5. A method according to any of the preceding claims, characterized in that when the same problems frequently occur in the control process, in particular a frequent tendency of the state variables (s(t)) to tend towards a limit (s₁, s_h), and/or when there is a frequent need for actions (aⁱ) to compensate for tendentially the same deviations, and/or when other inconsistencies occur in the control processes aimed at achieving setpoints (s_o) of the state variables (s(t)) and also a stable combustion process, a changeover is made from setpoint control to extraordinary disturbance control.
6. A method according to Claim 5, characterized in that in the case of extraordinary disturbance control, states are approached that are matched to the triggering problems.
7. A method according to any of the preceding claims, characterized in that several process models are used to evaluate the state variables (s(t)) and to select the actions (a'), in order to obtain short-term, medium-term and long-term predictions for the state variables (s(t)).
8. A control loop for implementing a method according to one of the preceding claims, in an installation (1), in particular in a power-generating plant, a waste incinerator or a cement plant, having a (controlled) system (3) for converting material (G) by means of the combustion process, with air (L) being supplied, and at least one flame body (F) being formed, and having at least one observation device (5) imaging the flame body (F) and having further sensors (7) to determine the state variables (s(t)) describing the state of the system in the installation (1), and having a computer (11) to evaluate the state variables (s(t)) and, if necessary, to select suitable actions (aⁱ), and having adjusting devices (9) that can be controlled by the actions (aⁱ) to regulate at least the supply of material (G) and/or air (L), during setpoint control the computer (11) controlling to achieve setpoints (s_o) of the state variables (s(t)) and/or stability of the combustion process, characterized in that the computer (11) changes occasionally from setpoint control to disturbance control and, according to the latter, selects actions (aⁱ) in order to approach states of the system in the installation (1) at which the state variables (s(t)) deviate in a targeted manner, within predetermined limits (s₁, s_h), from the optimal setpoint (s_o).
9. A control loop according to Claim 8, characterized in that in the computer (11) there is implemented at least one neuronal network which in each case stores a process model for evaluating the state variables (s(t)) and selecting the actions (aⁱ).
10. A control loop according to Claim 9, characterized in that the computer (11) contains several neuronal networks with process models for short-term, medium-term and long-term predictions of the state variables (s(t)), and/or with process models which compete with each other as regards the quality of the predictions.

Images