Method for optimizing at least one controller parameter of a pump control system and / or at least one physical variable that is used

Abstract

The invention relates to a method for optimizing at least one controller parameter of a pump control system and / or at least one physical variable that is used, which is used by the pump control system, the method being carried out independently by an integral control system of the pump, comprising the method steps of: selecting at least one target variable of the pump control system that is supposed to be minimized by the method, defining one or more search parameters that are varied to minimize the target variable, the one or more search parameters corresponding to the at least one controller parameter to be optimized and / or to the at least one physical variable to be optimized, and performing a learning cycle, the defined one or more search parameters being varied using the Kiefer-Wolfowitz approach during the learning cycle until the target variable has been minimized.

Description

[0001] Description

[0002] Method for optimizing at least one controller parameter of a pump control and / or at least one physical variable used

[0003] The invention relates to a method for optimizing at least one controller parameter of a pump control and / or at least one physical variable used by the pump control.

[0004] Modern pumps in the power range up to 3 kW (and in individual cases up to 50 kW) are often speed-controlled and therefore equipped with a frequency converter. To control the motor speed using a frequency converter, current controllers and speed controllers (typically designed as PI or, in exceptional cases, PID controllers) are commonly used. To control the motor efficiently and stably, several interdependent controller parameters of the controller used must be optimized. For example, the controller parameters are often determined with a certain degree of accuracy using common methods such as Ziegler-Nichols and then optimized through manual tuning. However, this optimization is time-consuming, and the result is usually far from optimal.

[0005] This approach is particularly disadvantageous when the converter and pump motor are separate units and sold separately. In this case, it is the customer's responsibility to configure and optimize the controller. However, the customer is not always qualified for this task. Modern pumps often come with PID controllers that the customer can and must configure for their system. The customer can then specify the desired controlled variable themselves. This can be done using signals that are either already available in the pump (e.g. pressure at the pump, flow rate, etc.) or in the form of external signals (e.g. temperature, pressure at any point in the system or a self-defined variable) that the customer provides via an interface to the pump control system. The customer can typically choose between speed, pressure and flow rate as the manipulated variable.However, due to the high flexibility in configuring the control system, the controller parameters cannot be preset and must be adjusted and optimized on-site, either by the customer or a technician.

[0006] Depending on the design of the pump control system, not only the controller parameters are important for the control, but other variables may also need to be defined, in particular physical variables of the pump or pump drive. Such variables can often only be estimated and therefore also require optimization. For example, information on the current rotor angular position is required for speed control. Since sensors are often not installed in the pump for this purpose for cost reasons, the angular position is calculated using an angular position estimator (PLL - Phase Locked Loop) during pump operation. The PLL also has PID parameters that need to be optimized. Furthermore, the PLL requires parameters that describe physical variables of the connected motor (such as flux linkage). Such physical variables are currently determined using complex measuring methods that do not always deliver optimal results.Furthermore, such variables may be subject to change with increasing operating time, so that optimization may also be useful here.

[0007] There is therefore a desire for a simplified method for optimizing the controller parameters and the physical variables, which can preferably be carried out by the pump itself and requires only minimal specialist knowledge from the operator. This object is achieved by a method according to claim 1. Advantageous embodiments of the method are the subject of the dependent claims.

[0008] According to the invention, a method for optimizing at least one controller parameter of the pump control used and / or a physical variable of the pump drive is proposed. A characteristic of the method is that it can be carried out independently by an integral control of the pump and is carried out automatically, for example, cyclically or upon request by the operator on the pump during ongoing pump operation. This makes it possible to optimize the required controller parameters for the integral pump control even after the pump has been put into operation, whereby these parameters can be continuously adapted to the actual conditions in the system. For the independent optimization of one or more controller parameters or a physical variable of the pump, a selection is first made via at least one target variable of the pump control.

[0009] The selected target variable describes, in a broader sense, a desired behavior of the pump control. It is conceivable that by specifying at least one suitable target variable of the controller, a control strategy can be implemented whose focus is on energy-efficient operation of the pump, in particular on energy-efficient reaching and maintaining the control setpoint. It is also conceivable to select a suitable target variable that results in the lowest possible controller overshoot. Other control properties that can be implemented using suitable target variables are, for example, the highest possible control quality and / or the fastest possible achievement of the final value and / or a low variance of the manipulated variable, e.g. a setpoint speed of the pump. The target variable is usually defined by one or more operating values ​​that are to be minimized within the framework of the process. Minimizing the winding current consumed by the pump motor is conceivable here.Furthermore, the control deviation between the setpoint and actual value can serve as a concrete value for the target variable.

[0010] According to the method according to the invention, one or more search parameters are defined depending on the selected at least one target variable, which are varied to minimize the at least one target variable. The search parameters correspond to the controller parameters to be optimized and / or the at least one physical variable of the pump drive to be optimized. The definition of the search parameters is preferably carried out automatically by the pump control system executing the method. After the search parameters have been defined, a learning cycle is executed. During the execution of the learning cycle, the one or more search parameters are varied using the Kiefer-Wolfowitz approach, and after each adjustment, the resulting influence of the variation on the selected target variable is checked. The search parameters are then further adjusted depending on the updated target variable until the target variable reaches a minimum value.The minimum value can be a local minimum, ideally a global minimum. If such a minimum is reached, the learning cycle is aborted, and the resulting search parameter values ​​are used as the new values ​​for the controller parameters or the physical quantity in regular pump operation.

[0011] In their investigations, the inventors discovered that particularly good results for optimizing the controller parameters of a pump control system can be achieved using the Kiefer-Wolfowitz approach. In particular, this approach can be implemented with comparatively limited hardware resources of a pump controller. The approach also delivers good optimization results even when the operating parameters of the pump control system are subject to high measurement noise. This is particularly advantageous when the pump control system operates partially sensorless and some of the operating parameters required for the control system can only be estimated.

[0012] At the beginning of the learning cycle, it can be specified that one or more defined search parameters are assigned an initial start value. This is particularly useful when the pump control is put into operation for the first time. If the learning cycle is instead carried out at a later point in time when the pump was already in operation, it is advantageous if the previous controller parameter values ​​are used as the initial start values ​​instead. However, it can also be specified that fixed initial start values ​​are always used for the execution of a learning cycle. For the execution of a single learning cycle, i.e. the Kiefer-Wolfowitz approach for a target variable, it is also advantageous if one or more boundary conditions for the execution of the learning cycle are first specified.Such boundary conditions can, for example, limit the variation range of one or more search parameters or other pump operating parameters that can be changed during the execution of the learning cycle in order to specify minimum or maximum values ​​for the individual controller parameters or physical variables or other parameters. It is also conceivable to define a minimum or maximum rate of change for the variation of at least one search parameter or the controlled variable or the control value. For example, in the case of a PID controller for controlling the speed of a pump, this could be a maximum rate of change of the speed change specified by the controller.

[0013] When selecting at least one target variable, it is also possible to select multiple target variables. Multiple selected target variables can be considered in different ways. For example, it is conceivable that individual target variables are initially combined into a single overall target variable, whereby it may be useful to incorporate the individual target variables into the overall target variable with different weightings. For example, it is conceivable that numerical values ​​of the target variables are first added together, ideally applying the individual weighting factor to each.

[0014] Alternatively, it can be provided that several target variables are minimized separately from one another by sequentially executing separate learning cycles. This means that a separate learning cycle is initiated and executed for each selected target variable. It is also conceivable that here, too, prioritization can be carried out when selecting several target variables, so that the chronological order of the separate learning cycles can be determined. To prevent a deterioration of the first target variable when executing the learning cycle for the second target variable, it is conceivable to add the value of the first target variable as a boundary condition for the second learning cycle. It is also conceivable to define several search parameters to be varied for the execution of the learning cycle and the implementation of the optimization algorithm.In this case, too, it makes sense to weight the individual search parameters, so that when varying the individual search parameters, for example, the focus of the variation is on the search parameter with the highest weighting.

[0015] It is conceivable that the weighting of the search parameters could be defined manually by the operator. However, it is preferred if the weighting of the individual search parameters is performed automatically by the control system, particularly depending on the one or more previously selected target variables.

[0016] As already mentioned above, the search parameters to be varied can be the individual controller parameters of a corresponding controller, for example, a PID controller, PI controller, or PITi controller. The search parameters to be varied then correspond to the P component and / or the I component and / or the D component of the controller used and / or the time constant Ti of an attached first-order low-pass filter.

[0017] As already indicated above, the method according to the invention can also be used for sensorless pump controls, since good optimization results can also be achieved when individual operating parameters are estimated rather than measured, and are therefore subject to higher measurement noise. An example of sensorless or sensorless control of a pump is the speed control of a drive motor, in particular a permanent magnet synchronous machine (PMSM), without an additional sensor / sensor to determine the current angular position of the rotor. When field-oriented control is used, the rotor angle can be estimated using a PLL structure. The methods used for angular position estimation usually require information about the anisotropy of the machine (physical quantity), which is usually determined in advance and stored in the control system at great expense.An example of such a physical quantity is the magnetic flux linkage of the motor, which is required to estimate the angular position of the rotor. Using the method according to the invention, such physical quantities can also be optimized. This is particularly advantageous because physical quantities are subject to changes with increasing operating time of the pump or motor, so that the implementation of the method always allows for adaptation of the physical quantities to the current machine state.

[0018] An acceleration of the Kiefer-Wolfowitz approach can be achieved, for example, through a well-known modification of the algorithm, also known as the simultaneous perturbation stochastic approximation (SPSA) approach. This modification requires fewer measurements for trend estimation and can therefore lead to an acceleration of the learning cycle.

[0019] The pump's drive motor is preferably a speed- and / or current-controlled permanent magnet machine, in particular a PMSM. The pump control system preferably includes a PID controller for both speed and current control. The control system preferably operates in a field-oriented manner. The pump preferably does not include a sensor for measuring the current angular position of the rotor, which is why the pump control system operates sensorless and estimates the angular position using a PLL (phased locked loop) structure. The PLL structure is also equipped with a PID controller, whose controller parameters can be optimized using the method.

[0020] In addition to the method according to the invention, the present invention also relates to a pump, in particular a circulating pump, with integral speed control and / or current control. The pump is characterized in that the integral control / regulation of the pump is configured to carry out the method according to the present invention. Consequently, the pump has the same advantages and properties as those already demonstrated above with reference to the method according to the invention. A repetitive description can therefore be omitted. Further advantages and properties of the invention are described in more detail below using an exemplary embodiment. In the drawings:

[0021] Fig. 1 : an overview of the overall structure of a generic PID controller,

[0022] Fig. 2: a list of the adjustable parameters of a PID controller,

[0023] Fig. 3: Diagrams of the value history of a search parameter and the target variable during the execution of the optimization algorithm,

[0024] Fig.4: a representation of an enlarged partial section of the diagram representations of Fig. 3, and

[0025] Fig. 5: Diagrams of the value history of three search parameters and the target variable during the execution of the optimization algorithm,

[0026] The method according to the invention will be described below using a centrifugal pump with a pump controller as an example. The pump controller is either an integral component of the pump unit or is installed as a separate component at the pump's point of use and communicates with the pump. The pump controller has a generic PID controller that can be used for various applications. For the sake of completeness, such a PID controller will be discussed in more detail below as an example.

[0027] PID controller

[0028] The PID controller can be used, for example, for constant temperature control or constant temperature difference control. The controlled variable, in this case the actual temperature or the difference between two actual temperatures, is recorded via the analog inputs and compared with the setpoint. Based on the current control deviation, a new manipulated variable is calculated, which is converted into a new speed in the drive. The overall structure of such a controller is shown in Figure 1. The hydraulic process 10 to be controlled, influenced by the speed of the frequency converter, represents the controlled system. The measured controlled variable is subtracted from the setpoint and thus forms the control deviation. The control deviation is fed to the actual process controller 20. The response of the controller 20 to a positive or negative control deviation is determined by the controller's action direction.With the normal controller action direction, the speed is reduced when the control difference is positive. With the inverted controller action direction, the speed is increased when the control difference is positive. The controller action direction can be set using parameter [3-3-6-6].

[0029] The individual controller parameters of the PID process controller 20 are set as follows:

[0030] The proportional component of the controller is defined via parameter [3-3-6-1]. The control deviation is amplified by the proportional component and applied to the control value.

[0031] To avoid a permanent control deviation, an integrating controller component is required in many hydraulic processes. For this purpose, the reset time of the integral component is set using parameter [3-3-6-2]. The control deviation is integrated, weighted according to the selected reset time, and added to the manipulated variable. Reducing the reset time leads to faster correction of the control deviation. If an reset time of 0 1 / s is selected, the integral component is deactivated. To prevent the integral component from integrating too far, for example because a specified setpoint is physically unattainable, there is an anti-windup protection that can be configured using parameter [3-3-6-7].

[0032] The derivative action allows controller 20 to respond to rapid changes in the control deviation. Whether a derivative action is necessary depends on the dynamics of the hydraulic process. It is not required in typical centrifugal pump applications. If a derivative action time of 0 s is selected, the derivative action of the process controller is deactivated. The derivative action time is set using parameter [3-3-6-3]. Increasing the derivative action time increases the response to rapid changes in the control deviation. The "D-action Limit" parameter [3-3-6-8] sets the maximum gain of the derivative action. This allows the effect of measurement noise on the control output to be limited. Reducing the limit value reduces the influence of the derivative action at high frequencies, which allows the influence of measurement noise to be suppressed.

[0033] The controller 20 shown operates internally with standardized values. This means that the setpoint and the actual value are standardized between 0 and 1 according to their value range. The controller's manipulated variable is then scaled at the controller's output between the minimum and maximum control signal. The table in Figure 2 lists all adjustable parameters of the controller 20.

[0034] Example: The temperature measuring range is 20 °C to 100 °C. Then, at an actual temperature of 20 °C, the value is normalized to the internal value 0. At an actual temperature of 100 °C, the value is normalized to the internal value 1. At an actual temperature of 60 °C, the value is normalized to the internal value 0.5.

[0035] The output is then scaled between the minimum and maximum manipulated variable. The manipulated variable here is the speed. This means that with a standardized manipulated variable at the controller output of 0.5 and a minimum speed of 1000 rpm and a maximum speed of 6000 rpm, the manipulated variable would be 3500 rpm.

[0036] The following will look at possible application examples for such a PID controller and the implementation of the method according to the invention for optimising the controller parameters of the PID controller. As already explained above, the controller is designed such that the manipulated variable and controlled variable are each normalised to 1. As a result, the controller parameters must also be set in a normalised range between 0 and 1. This has the advantage of finding stable controller parameters that function very well regardless of the control task and the size of the pump / system. The manufacturer can therefore set robust default parameters for the P, I and, if applicable, D components, regardless of the application area and pump size for which the controller is used. Nevertheless, it is often necessary for the end user to readjust, i.e. optimize, the parameters for their specific application and system.

[0037] One application of a PID controller can be, for example, pressure, flow, or temperature control. Hydraulic controls such as pressure and flow can usually be easily adjusted by the manufacturer and rarely require readjustment. The situation is different with temperature controls, as these can involve a wide variety of control tasks, such as controlling a hall temperature, controlling the medium temperature through the pump, or controlling the outlet temperature of a heat exchanger. Due to the multitude of possible applications and the significant differences between systems, it is not possible for manufacturers to offer a preset, optimal set of parameters for all scenarios. The method according to the invention is particularly valuable for precisely such a scenario.

[0038] Example: Hall temperature control

[0039] The following scenario presents a hall temperature control scenario. The heating system includes a circulation pump whose speed is to be controlled depending on the hall temperature to pump the heating medium. In particular, the hall temperature is to be kept at a constant value (e.g., 20 °C) (setpoint - must be set in the pump). The actual temperature in the hall is recorded by a temperature sensor and transmitted to the pump via an interface (e.g., 0 - 10 V). The measured temperature is then normalized to the value range 0 --> 1 and fed to the PID controller as a controlled variable. The controller calculates the manipulated variable (usually the speed) based on the set P, I, and D parameter values. This means that if the setpoint temperature is higher than the actual temperature, the speed increases. As a result, the pump pumps the heating medium more quickly, and the room heats up.In refrigeration applications or in certain applications with a heat exchanger, the direction of action of the controller can also be reversed.

[0040] If the end user now needs to set the P, I, and D parameter values ​​for their application, they can obtain support from the pump that executes the method according to the invention. Communication between the end user and the pump can take place, for example, via an application running on an external device or via a display and control element installed on the pump or pump controller.

[0041] The user interface prompts the user to configure the method according to the invention. The following settings must be made before executing the method:

[0042] - Configuration of the boundary conditions

[0043] - Configuration of one or more target variables

[0044] - Setting the optimization algorithm to be executed

[0045] Boundary conditions:

[0046] First, the user must define boundary conditions, i.e. limits within which the algorithm may move, such as:

[0047] - Speed ​​limits between which the algorithm may move

[0048] - Maximum increase in speed (ramp)

[0049] - Limits for flow or pressure

[0050] Target values:

[0051] The user then has to specify the optimization goal the algorithm is aiming for. Selectable target variables or optimization goals can be:

[0052] - Low overshoot

[0053] - High control quality (low variance in the temperature signal)

[0054] - Fast reaching of the final value Low variance in the speed signal

[0055] Energy-efficient achievement and maintenance of the setpoint (the pump's performance estimate can be used for this)

[0056] The optimization algorithm used in the method according to the invention requires exactly one target variable. However, it may happen that more than one target variable is important to the user. The method according to the invention therefore allows the selection of multiple target variables. The target variables can be processed according to two different approaches:

[0057] Approach 1 : Summarizing and weighting target variables

[0058] For example, if the target variables "low overshoot" and "low variance in the speed signal" are to be combined, the amplitude of the overshoot and the variance in the speed signal can be added to form a single overall target variable. Both individual signals can also be weighted to prioritize their influence on the overall target variable. The weighting is also determined by the user.

[0059] Approach 2: Perform optimizations (learning cycles) one after the other

[0060] A second option is to perform the optimizations (learning cycles) sequentially. For example, an optimization is first performed with only the target variable "low overshoot." The P, I, and D parameters found in this first optimization serve as the starting values ​​for the second optimization, which only specifies the target variable "low variance in the speed signal." To prevent the first target variable from deteriorating again during the second optimization, the first target variable "low overshoot" can be added as a boundary condition to the second optimization.

[0061] Setting the optimization algorithm:

[0062] The functionality and configuration options of the optimization algorithm used, i.e., the Kiefer-Wolfowitz approach, are discussed in more detail below. Many of the parameters can already be preset by the manufacturer.

[0063] However, the user should be prompted to complete the task for the following parameters:

[0064] - Scaling factor:

[0065] To do this, the process could query the user interface for the maximum noise level to be expected in the temperature signal. The user can obtain this value, for example, from the sensor's data sheet. The pump then calculates a reasonable scaling factor from this.

[0066] - Step size:

[0067] At this point, the process could query the temperature range within which the actual temperature of the building is likely to be. The pump would then calculate a suitable value for the step size.

[0068] The pump can determine the following parameters itself or at least suggest them:

[0069] - Weighting function:

[0070] This is where the weighting between the three controller parameters (P, I, D) is determined. The pump can independently suggest a sensible weighting based on the target variables. For example, with a target variable of "low variance," the D component plays a greater role. If the target variable is "high control quality," the focus should be somewhat more on the I component.

[0071] - Minimum parameter, Maximum parameter:

[0072] Since the dynamics of the pump are known, the manufacturer can set reasonable limits for its minimum and maximum P, I, and D parameters

[0073] Description of the optimization algorithm “Kiefer-Wolfowitz”

[0074] What does the algorithm do?

[0075] The goal of the algorithm is to minimize a defined value. This value is called the "target variable." To achieve this goal, optimal parameters (hereinafter referred to as "search parameters") are sought. The number of search parameters can be arbitrary. The target variable and the search parameters to be optimized must be determined before optimization. Furthermore, suitable starting values ​​for the search parameters must be specified.

[0076] How does the algorithm work

[0077] The function of the algorithm is first explained using a simple example with only one search parameter:

[0078] Step 1: Check whether increasing or decreasing the search parameters minimizes the target size.

[0079] To do this, an offset is added to the default search parameter and the target variable is determined using simulation. The same offset is then subtracted from the search parameter and the new target variable is determined using simulation or by measurement during pump operation.

[0080] Step 2: Specify a new value for the search parameter

[0081] The gradient between the results of the two simulations (with negative and positive offset) is calculated. Based on this, a new value for the search parameter is calculated.

[0082] Example

[0083] A system can be described by the following equation: y = (1.5 - a) 2

[0084] Here, y is the target variable to be minimized. The search parameter to be varied is a.

[0085] The starting value a = 1. The image in Figure 3 shows the courses of a and y during the algorithm runs. The results show that a value for the search parameter a can be found that significantly minimizes the target variable y. To understand how the algorithm works in more detail, the first two iterations are shown enlarged in Figure 4. The algorithm starts at time 0 (point 1 in Fig. 4) with the default value for the parameter a = 1 . An offset is subtracted from the search parameter a and the corresponding target variable y is determined (point 1 in Fig. 4). The same offset is then added to the search parameter a and the target variable is determined again (point 3).

[0086] The illustration in Figure 4 shows that in this example, the negative offset (point 2) increased the target variable, while the positive offset (point 3) decreased it. Therefore, the search parameter for a is increased in the next iteration (point 4), and the offset value is again alternately added or subtracted. The process is iterated until the target variable reaches a minimum value.

[0087] Parameterization of the algorithm

[0088] Scaling factor (for points 2 and 3):

[0089] The scaling factor determines the offset (points 2 and 3). The more noisy the raw signal is, the higher the offset must be to obtain valid results. If the raw signal is not noisy, a smaller value can be selected.

[0090] Step size (for point 4):

[0091] The step size determines by how much the search parameter a is increased or decreased between two iterations (point 4). If the initial values ​​are already good and optimization is only desired in the short range, a small value can be set here. If no good initial values ​​are known and the entire value range is to be searched, a large step size should be selected here.

[0092] Weighting function A, B, C (For point 4):

[0093] If there is more than one search parameter, it may be necessary to select different step sizes. However, there is only one parameter for the step size. A weighting can then be defined for each search parameter. For example, if there are three search parameters (A, B, C), there are also three weighting parameters. If one of the search parameters should have a smaller step size than the others, the corresponding weighting parameter is selected to be smaller than the other weighting parameters.

[0094] Minimum parameter, Maximum parameter (For point 4):

[0095] It is often necessary to limit the variation range of the search parameters. For example, if controller parameters are to be optimized in a running system, instabilities must be avoided.

[0096] Settling time:

[0097] If a search parameter has been changed, it may take some time for the target variable to settle. You should wait for this settling time before evaluating the results.

[0098] Measurement duration:

[0099] When optimization is run in a real system, the target signal is often overlaid with noise. This noise is averaged for evaluation by the optimizer. The averaging time is determined by this parameter.

[0100] The example outlined here can be extended to multiple search parameters. An example with three search parameters is outlined below:

[0101] The target variable y is to be minimized. Here, y depends on the search parameters a, b, and c, which are to be varied, according to the following exemplary equation: y = a + 2 * b - 3 * c - 40 2 ;

[0102] Figure 5 shows the value curves for the search parameters a, b, and c, as well as for the target variable y. The left-hand diagrams show the curves without noise. The diagrams on the right show the value curves of the search parameters and the target variable for the case where the raw signal y, which is fed into the optimizer, is overlaid with noise.

[0103] In this example, a minimum is also found for the target variable y. It is noticeable that the target variable is minimized both with and without noise, but with different values ​​for the three search parameters. This is not unusual because there are multiple solutions to the described problem and the solution path is changed by the noise. It is also noticeable that in the approach without noise, the search parameters and the target variable move evenly towards the optimum. In the approach with noise, on the other hand, the signals sometimes change direction abruptly. Nevertheless, the algorithm is well suited to finding an optimum even in the presence of noise. This algorithm is therefore particularly suitable for applications in operating plants.

Claims

Patent claims Method for optimizing at least one controller parameter of a pump control and / or at least one physical variable used 1 . Method for optimizing at least one controller parameter of a pump control and / or at least one physical variable used by the pump control, wherein the method is carried out independently by an integral controller of the pump, with the method steps: a. selecting at least one target variable of the pump control that is to be minimized by the method, b. defining one or more search parameters that are varied to minimize the target variable, wherein the one or more search parameters correspond to the at least one controller parameter to be optimized and / or the at least one physical variable to be optimized, c. executing a learning cycle, wherein during the learning cycle the defined one or more search parameters are varied using the Kiefer-Wolfowitz approach until the target variable has been minimized.

2. Method according to claim 1, characterized in that by means of the target variable a small overshoot of the controller and / or a high control quality and / or a rapid reaching of the final value and / or a small variance in the control signal and / or an energy-efficient reaching and holding of the setpoint of the control can be specified.

3. Method according to claim 1, characterized in that initial starting values ​​for the one or more search parameters are determined before the learning cycle is carried out.

4. Method according to one of the preceding claims, characterized in that before the learning cycle is carried out, one or more boundary conditions are additionally defined in order to limit the variation space of the one or more search parameters and / or the one or more physical quantities and / or one or more operating parameters of the pump that can be changed during the execution of the learning cycle and / or to define a minimum or maximum rate of change of the search parameters.

5. Method according to one of the preceding claims, characterized in that when several target variables are selected, these are combined, preferably by applying a weighting of the individual target variables.

6. Method according to one of the preceding claims 1 to 4, characterized in that when several target variables are selected, separate learning cycles are carried out sequentially for the individual target variables.

7. Method according to one of the preceding claims, characterized in that when defining several search parameters to be varied during a learning cycle, the search parameters are weighted such that the focus of the variation is on the search parameter with the highest weighting.

8. Method according to claim 5, characterized in that the weighting of the search parameters can be defined manually or is determined by the pump control as a function of the selected target variable.

9. Method according to one of the preceding claims, characterized in that the search parameters to be varied are the P component and / or the I component and / or the D component of a corresponding controller and / or the time constant T of a connected low-pass filter.

10. Method according to one of the preceding claims, characterized in that the pump control operates sensorless and the physical variable to be optimized is a necessary parameter for calculating / estimating an operating parameter of the pump.

11. Method according to one of the preceding claims, characterized in that the physical variable to be optimized is the magnetic flux linkage of the pump's drive motor.

12. Method according to one of the preceding claims, characterized in that an acceleration of the Kiefer-Wolfowitz approach is achieved by performing the simultaneous perturbation stochastic approximation (SPSA).

13. Pump, in particular a circulation pump, with integral speed control, characterized in that the integral speed control of the pump is configured to carry out the method according to one of the preceding claims.

Images