Method and device for setting at least one parameter of an actuator control system, actuator control system and data set

Learning system (40) for automatically setting at least one parameter (θ) of an actuator control system (45), which is set up to control a control variable (x) of an actuator (20) to a predefinable target value (xd), wherein said learning system (40) is set up to carry out a method, wherein the actuator control system (45) is set up, depending on the at least one parameter (θ), the target value (xd) and the control variable (x) to generate a manipulated variable (u) and depending on this manipulated variable (u) to control the actuator (20),

• wherein a new value (θ*) of the at least one parameter (θ) is selected depending on a long-term cost function (R), wherein this long-term cost function (R) is determined depending on a predicted temporal evolution (F) of a probability distribution (P) of the control variable (x) of the actuator (20) and the parameter (θ) is then set to this new value (θ*);
• wherein the expected temporal evolution (F) of the probability distribution (p) of the control variable (x) is determined by an approximation of an integration over possible values of the control variable (x), this approximation being done through numerical quadrature;
• wherein a density of the supporting points (ξ) depends on a determined temporal evolution (τ1 . . . τT), determined in particular by means of at least one of the model (g) and the actuator control system (45), of the control variable (x), starting from a randomly determined initial value (τ0) of the control variable (x) from an initial probability distribution (p(x0));
• and
• wherein the density of supporting points (ξ) also depends on a determined temporal evolution (τ′1 . . . τ′T) of the control variable (x), determined in particular by means of at least one of the model (g) and the actuator control system (45), starting from the target value (xd) as the initial value (τ′0) of the control variable (x).