METHOD FOR CONTROLLING A PROCESS WITHIN A SYSTEM, IN PARTICULAR A COMBUSTION PROCESS IN A BOILER OR FURNACE

Description

[0001] The invention relates to a method for controlling a combustion process in a boiler or furnace or a grinding process in a grinding device.

[0002] In such methods, the system's state variables are recorded, preferably by measuring process variables within the system. Furthermore, a process model is created that describes the effect of control actions on the system's state variables, for example, using computer-aided neural networks. The process within the system is controlled by executing control actions, taking into account the process model and specified control objectives.

[0003] In a known method of this type, difficult-to-measure or expensive-to-measure process variables are predicted using the process model in the neural network. To be able to track changes in the process, three steps are performed cyclically: a process analysis to identify an approach for the process model, training the neural network, and applying the process model for prediction. This process is time-consuming and labor-intensive.

[0004] US 5,259,064 A describes a method in which either a neural network used for predictions is continuously trained or two neural networks of the same structure are used, one for predictions and one for training.

[0005] US 5,943,660 A shows a method in which, starting from a current state, a linearized approach is used for a control action to achieve an optimization goal, whereby the two linearization coefficients for this are each determined from a neural network.

[0006] EP 0 609 999 A1 describes a method in which, on the one hand, the control signals for the actions required to achieve the current process variables in the system are determined using a first neural network. On the other hand, the control signals for the actions required to achieve the desired, possibly continuously corrected, process variables in the system are determined using a second neural network that corresponds to the first neural network. The difference between the two control signals is reduced by a control loop until the desired process variables are also the current process variables.

[0007] EP 1 364 163 B1 discloses a method for controlling a thermodynamic process, in particular a combustion process. The method involves measuring the system state, comparing it with optimization objectives, and implementing appropriate control actions within the system. A process model independent of the optimization objectives is determined, describing the impact of actions on the system state. Furthermore, a situation assessment independent of the process model uses quality functions to evaluate the system state with respect to the optimization objectives.

[0008] EP 1 396 770 B1 discloses a method for controlling a thermodynamic process, in which process variables in the system are measured, predictions are calculated in a neural network using a trained, current process model, compared with the optimization target and suitable actions are carried out in the system to control the process, wherein the process is automatically analyzed in parallel, at least one new process model is formed, trained and compared with the current process model with regard to the prediction.

[0009] The article Reza Mohammad ET AL: "On Fractional-Order PID Design" in "Applications of MATLAB in Science and Engineering", September 9, 2011, XP093018924, ISBN: 978-953-30-7708-6, DOI: 10.5772 / 22657 and the article Cooper Doug ET AL: "Feed Forward Uses Models Within the Controller Architecture", April 9, 2015, XP093019312, each disclose a description of a system based on a process model and a disturbance model.

[0010] The present invention is based on the object of improving the accuracy of the known methods.

[0011] The object is achieved according to the invention by a method having the features of claim 1. Further advantageous embodiments are the subject of the subclaims. The method according to the invention for controlling a combustion process in a boiler or furnace or a grinding process in a grinding device comprises the steps: Recording state variables (st) of the system; creating several independent disturbance models (SM) that describe the effects of disturbance-related system changes (vt) on the state variables (st) of the system, whereby when controlling (4) the process within the system the disturbance model (SM) that currently best represents the disturbance-related system changes (vt) is taken into account; creating several independent process models (PM) that describe the effects of control actions (at) on the state variables (st) of the system, whereby when controlling (4) the process within the system the process model (PM) that currently best represents the effects of control actions (at) on the state variables (st) of the system is taken into account;and controlling the process within the system by executing control actions (at) taking into account the process model (PM), the disturbance model (SM) and specified control objectives; ; wherein, when creating the plurality of independent disturbance models (SM), a distinction is made between the effects of short-term disturbance-related system changes (vt) on the state variables (st) of the system and the effects of longer-term disturbance-related system changes (vt) on the state variables (st) of the system, wherein the division into short-term disturbance-related system changes (vt) and longer-term disturbance-related system changes (vt) depends on the process to be controlled within the system and takes into account the length of the process to be controlled in the system; and wherein the plurality of independent process models (PM) take into account several assumed future effects of control actions (at) on the state variables (st) of the system.

[0012] The methods known from the state of the art use a process model that describes the effects of control actions on the state variables of the system and, at the same time, the effect of disturbance-induced system changes on the state variables of the system.

[0013] The present invention is based on the finding that the creation of disturbance models, which describe the effects of disturbance-related system changes on the state variables of the system, and the creation of separate process models, which describe the effects of actuating actions on the state variables of the system, significantly improves the accuracy of the inventive method for controlling a combustion process in a boiler or furnace or a grinding process in a grinding device, in contrast to a process model, which describes the effects of actuating actions on the state variables of the system and at the same time the effects of disturbance-related system changes on the state variables of the system.

[0014] When creating the disturbance models according to the invention, a distinction is made between the effects of short-term disturbance-related system changes on the state variables of the system and the effects of longer-term disturbance-related system changes on the state variables of the system.

[0015] The distinction between short-term disturbance-related changes and longer-term disturbance-related changes depends on the process to be controlled within the system. For example, in a combustion process in a power plant, short-term disturbance-related changes can easily occur over a period of several hours. The distinction between short-term disturbance-related changes and longer-term disturbance-related changes is made taking into account the duration of the process to be controlled within the system.

[0016] The inventive consideration of the disturbance models when controlling the process within the system has the advantage that the execution of point actions can be dispensed with if, according to the disturbance model created, it can be assumed that the effects of the disturbance-related system changes on the state variables of the system are limited in time and / or the effects of disturbance-related system changes on the state variables of the system are within a certain range.

[0017] The inventive consideration of the process models, the disturbance models and the specified control objectives when controlling the process within the system thus significantly increases the accuracy of the method for controlling the process within the system.

[0018] According to one variant of the invention, the state variables (st) are recorded using system sensors or manual and / or automatic sample evaluations. Recording of the state variables (st) using sensors is preferred, as this can be done continuously. However, if state variables (st) cannot be recorded directly using sensors, manually and / or automatically taken samples must be evaluated and made available to the method according to the invention.

[0019] According to a preferred variant of the invention, during the creation of the disturbance models (SM), past disturbance-related system changes (vt) are taken into account for integration in a temporal context. Taking past disturbance-related system changes into account when creating the disturbance models (SM) results in a continuous improvement of the disturbance models (SM).

[0020] According to a further variant of the invention, the disturbance models (SM) are continuously adapted based on the effects of disturbance-related system changes (vt) on the system's state variables (st). By continuously adapting the disturbance models based on the effects of disturbance-related system changes on the system's state variables, the method according to the invention always has a current disturbance model available for controlling the process within the system.

[0021] In conjunction with the consideration of past fault-related system changes when creating the fault models, an up-to-date fault model is available at all times and an optimized fault model with regard to past fault-related system changes.

[0022] According to a variant of the method according to the invention, the disturbance models (SM) are created by means of a test run of the system and / or using expert knowledge. The disturbance models (SM) initially created by means of a test run of the system and / or using expert knowledge can be adapted for integration in a temporal context by continuously considering the effects of disturbance-related system changes (vt) on the state variants (st) of the system and / or by considering past disturbance-related system changes (vt).

[0023] In a further variant of the system according to the invention, the disturbance models (SM) are created by means of a computer-aided neural network and, in particular, are continuously adapted (trained).

[0024] The neural network is trained in particular by means of known evolutionary strategies, genetic algorithms, genetic programming or evolutionary programming, preferably within the framework of an autonomous selection and / or optimization process.

[0025] According to the invention, several independent disturbance models (SM) are created, and when controlling the process within the system, the disturbance model (SM) that best represents the current disturbance-induced system changes (vt) is taken into account. Preferably, several computer-aided neural networks are created for the independent disturbance models (SM). When controlling the process within the system, the disturbance model (SM) that delivers the best results for the current situation is taken into account. Depending on the current process situation, the disturbance models (SM) are compared with each other, and a disturbance model (SM) is selected for control.

[0026] In a particularly preferred variant of the invention, each disturbance model (SM) is created to predict a specific group of disturbance-related system changes (vt). The disturbance-related system changes (vt) are divided into groups (clusters) by the disturbance models (SM), and each disturbance model is specifically designed to predict a specific group of disturbance-related system changes (vt). When controlling the process within the system, the disturbance model (SM) that delivers the best results for the current disturbance-related system changes (vt) is taken into account.

[0027] If disturbance-related system changes (vt) occur during the control of the process within the system that are not captured by any group of one of the disturbance models (SM), a new disturbance model (SM) is created for the new group of disturbance-related system changes (vt) and subsequently improved (trained) based on measurements and the execution of control actions (at).

[0028] It is useful to assign a unique identification code (genetic code) to each disturbance model (SM) or to each group (cluster) belonging to the disturbance models in order to simplify the differentiation of the individual disturbance models (SM) or groups (clusters) of disturbance-induced system changes (vt).

[0029] Furthermore, the quality of the individual disturbance models (SM) or the groups (clusters) of disturbance-related system changes (vt) can preferably be determined and / or adjusted.

[0030] According to a further preferred variant of the invention, the disturbance models (SM) consider several future-oriented assumed effects of disturbance-induced system changes (vt) on the state variables (st) of the system. The disturbance models (SM) generate a prediction for the future-oriented assumed effect of disturbance-induced system changes (vt) on the state variables (st) of the system and consider this in the current situation of the process within the system.

[0031] According to a variant of the method according to the invention, during the creation of the process models (PM), past effects of control actions (at) on the system's state variables (st) are taken into account for integration in a temporal context. Thus, during the creation of the process models (PM), the past effects of control actions (at) on the system's state variables (st) are taken into account, so that the process models (PM) are continuously improved taking the past into account.

[0032] In a further variant of the invention, the process models (PM) are continuously adapted based on the effects of control actions (at) on the system's state variables (st). The process models (PM) are thus continuously improved. In particular, by additionally considering past effects of control actions (at) on the system's state variables (st), improved and constantly updated process models (PM) are created.

[0033] According to a variant of the invention, the process models (PM) are created by means of a test run of the system with at least exemplary execution of possible control actions (at) and / or by means of expert knowledge. Process models created in this way can subsequently be continuously adapted based on the effect of control actions (at) on the state variables (st) of the system, and / or past effects of control actions (at) on the state variables (st) of the system can be taken into account when creating the process models (PM).

[0034] According to an expedient variant of the invention, the process models (PM) are created by means of a computer-aided neural network and, in particular, are continuously adapted (trained).

[0035] The neural network is trained in particular by means of known evolutionary strategies, genetic algorithms, genetic programming or evolutionary programming, preferably within the framework of an autonomous selection and / or optimization process.

[0036] According to the invention, several independent process models (PM) are created, and when controlling the process within the system, the process model (PM) that currently best represents the effects of control actions (at) on the system's state variables (st) is taken into account. In particular, the several independent process models (PM) are implemented using different computer-aided neural networks. When controlling a system using the method according to the invention, the process model that currently best represents the effects of control actions (at) on the system's state variables (st) can thus be taken into account.

[0037] According to the method according to the invention, the process models (PM) consider several future-oriented assumed effects of control actions (at) on the state variables (st) of the system. The process models (PM) thus consider a prediction of assumed effects of control actions (at) on the state variables (st) of the system.

[0038] The process models (PM) can therefore take into account past effects of actuating actions (at) on the state variables (st) of the system, continuously take into account the effects of actuating actions on the state variables (st) of the system and furthermore make a prediction about assumed future effects of actuating actions (at) on the state variables (st) of the system and also take these into account.

[0039] The invention further relates to a device for carrying out one of the aforementioned methods, comprising a system to be controlled with sensors for detecting state variables (st), an actuating device for executing actuating actions (at), and a control device connected to the system, preferably a computing device. The method according to the invention can be implemented, for example, by software on the computing device.

[0040] The method according to the invention is described below using the Fig. 1 illustrated embodiment is explained in more detail.

[0041] It shows: Fig. 1The sequence of a method according to the invention for controlling a combustion process in a boiler or furnace.

[0042] Fig. 1 shows the sequence of a method according to the invention for controlling a combustion process in a boiler or furnace.

[0043] In a first step, the state variables (st) of the system are recorded 1. The recording 1 of the state variables (st) is carried out by means of sensors of the system or by means of manual and / or automatic sample evaluations.

[0044] When carrying out the method according to the invention, a plurality of independent disturbance models (SM) are created 2, which describe the effects of disturbance-related system changes (vt) on the state variables (st) of the system. The disturbance models (SM) are created, for example, by means of a test run of the system and / or through expert knowledge 2. When carrying out the method according to the invention, during the creation 2 of the disturbance models (SM), past disturbance-related system changes (vt) can be taken into account for integration in a temporal context. Furthermore, the disturbance models (SM) can be continuously adapted based on the effect of disturbance-related system changes (vt) on the state variables (st) of the system.

[0045] The disturbance models (SM) are created, for example, using a computer-based neural network 2 and, in particular, are continuously adapted.

[0046] According to the invention, several independent disturbance models (SM) are created 2 and when controlling 4 the process within the system, the disturbance model (SM) is taken into account which currently best represents the disturbance-related system changes (vt).

[0047] To improve the accuracy of the method according to the invention, the disturbance models (SM) can take into account several assumed future effects of disturbance-induced system changes (vt) on the state variables (st) of the system.

[0048] In a further step of the method according to the invention, several independent process models (PM) are created which describe the effects of actuating actions (at) on the state variables (st) of the system.

[0049] The process models (PM) are created, for example, by means of a test run of the system with at least exemplary execution of possible control actions (at) and / or through expert knowledge. 3 During the creation of the process models (PM), preferably past effects of control actions (at) on the system's state variables (st) are taken into account for integration in a temporal context. Furthermore, the process models (PM) are preferably continuously adapted based on the effects of control actions (at) on the system's state variables (st).

[0050] The process models (PM) are expediently created using a computer-based neural network 3 and, in particular, are continuously adapted.

[0051] According to the invention, several independent process models (PM) are created 3 and when controlling the process within the system, the process model (PM) is taken into account which currently best represents the effects of control actions (at) on the state variables (st) of the system.

[0052] To further improve the method according to the invention, the process models (PM) take into account several future-oriented assumed effects of control actions (at) on the state variables (st) of the system.

[0053] According to the method according to the invention, the process within the system is controlled by executing control actions (at) taking into account the process model (PM), the disturbance model (SM) and predetermined control objectives 4. List of reference symbols

[0054] 1Capturing state variables (st) 2Creating a disturbance model (SM) 3Creating a process model (PM) 4Controlling the process within the system

Claims

1. Method for controlling (4) a combustion process in a boiler or furnace or a grinding process in a grinding device, comprising the following steps: - capturing (1) of state variables (st) of the system; - creating (2) multiple, independently from one another, interference models (SM), which describe the effects of interference-based system changes (vt) on the state variables (st) of the system, wherein during the controlling (4) of the process within the system one interference model (SM) of the multiple, independently from one another, interference models (SM) is considered that momentarily best describes the interference-based system changes (vt); - creating (3) multiple, independently from one another, process models (PM), which describe the effects of setting actions (at) on the state variables (st) of the system, wherein during the controlling (4) of the process within the system one process model (PM) of the multiple, independently from one another, process models (PM) is considered that momentarily best describes effects of setting actions (at) on the state variables (st) of the system; and - controlling (4) the process within the system by performing setting actions (at) by considering the process model (PM), the interference model (SM) and predetermined controlling goals wherein during the creation of the multiple, independently from one another, interference models (SM) a distinction is made between effects of short-term interference-based system changes (vt) on the state variables (st) of the system and effects of long-term interference-based system changes (vt) on the state variables (vt) of the system, where the distinction between short-term interference-base changes (vt) and long-term interference-based changes (vt) depends on the process within the system and is done by considering the duration of the controlled process in the system; and wherein the multiple, independently from one another, process models (PM) consider multiple assumed future effects of setting actions (at) on the state variables (st) of the system.

2. Method according to claim 1, wherein the state variables (st) are captured (1) by sensors of the system or a manual and / or automatic sample evaluation.

3. Method according to claim 1 or claim 2, wherein during the creation (2) of the interference models (SM) past interference-based system changes (vt) are considered for integration in a temporal context.

4. Method according one of claims 1 to 3, wherein the interference models (SM) are adapted continuously by effects of interference-based system changes (vt) on the state variables (st) of the system.

5. Method according to one of claims 1 to 4, wherein the interference models (SM) are created (2) by a test run of the system and / or by expert knowledge.

6. Method according to one of claims 1 to 5, wherein the interference models (SM) are created (2, 3) by a computer-based neural network and particularly are continuously adapted.

7. Method according to one of claims 1 to 6, wherein the interference models (SM) consider multiple assumed future effects of interference-based system changes (vt) on the state variables (st) of the system.

8. Method according to one of claims 1 to 7, during the creation (3) of the process models (PM) past effects of setting actions (at) on the state variables (st) of the system are considered for integration in a temporal context.

9. Method according to one of claims 1 to 8, wherein the process models (PM) are adapted continuously by effects of setting actions (at) on the state variables (st) of the system.

10. Method according to one of claims 1 to 9, wherein the process models (PM) are created (3) by a test run of the system with at least an exemplary execution of possible setting actions (at) and / or by expert knowledge.

11. Method according to one of claims 1 to 10, the process models (PM) are created (2, 3) by a computer-based neural network and particularly are continuously adapted.

12. Apparatus for executing the method according to one of claims 1 to 11 comprising a system to be controlled with sensors for capturing of state variables (st), actuators for performing setting actions (at) and a control unit, preferably a computing device, connected to the system.