SU744447A1 - Extremum searching method - Google Patents

Description

Depending on the initial deviation of the output signal of the object from the extreme value, the change in the input signal when moving to the point He occurs at practically constant average speed, since the period of trial action during the search process varies slightly. This circumstance lengthens the search process in time, does not allow obtaining a substantially proportional approach to the He point, leads to unjustifiably large fluctuations at the entrance of the object upon reaching the extremum and maintaining it in a steady state.

The purpose of the invention is to speed up the search progress, obtain a proportional approximation to the extremum point and reduce oscillations applied to the inputs of the object in a steady state, which increases the search accuracy.

The goal is achieved by the fact that the amplitude of the probe signals at the time point corresponding to the beginning of the trial period change in proportion to the amplitude of the output signal when searching for the minimum and in proportion to the difference between the constant signal and the output signal when searching for the maximum more of the output signal, and the amplitude is kept constant during the whole test exposure period, the search time is shortened, the proportionality of the search process, proportionality under move to the extremum point and decrease the oscillation at the entrance of the object in the steady state operation. When searching for the extremum of a function of several variables, the amplitude of the test signals for each variable is changed at the time point corresponding to the beginning of the test period for this variable, and all variables are searched simultaneously.

The drawing shows a block diagram of an automatic search system, realizing, the proposed method for it.

The output signal Q of object 1 is a function of p input variables Xi, ..., Xn. After differentiation Q differentiator 2, the derivative dQIdt is fed to the inputs of all generators 3i, ..., Zn pulses, which control the triggers 4i, ..., 4n. Pulse-width modulated signals from the output of each trigger are fed to the corresponding output integrating links 5i, ..., 5p. The pulse generator Zg, the trigger 4 {and the output integrating link 5g form one channel of the multichannel optimizer. On the output integrator links 5i, ..., 5p of the optimizer channels, in addition to the signals U, from the outputs of the corresponding triggers, additional signals f / dopg Ki-Qsign-t / T are given; from blocks 61, ..., 6n, which except

the multiplication operations maintain the Unoai values unchanged during one trial period and are proportional to the value of the signal Q at the time corresponding to the beginning of the trial period in each channel, respectively. The f / sub value is changed discretely from the period to the period of the test action in accordance with a change in the signal Q. The need to maintain the value of Sdop constant; during the trial period, is caused by the fact that the continuously varying signal t / apt; would distort work

an automatic optimizer with pulse width modulation that would make it difficult to find an extremum.

If the coefficients / Ci are chosen close to unity, then it is possible to decrease the values of the i / Ti signals and then when searching for the minimum

Q on reaching (in the particular case it may turn out that) the amplitude of test signals applied to each input of the object will vary from a significant value at the beginning of the search process when Q is large to a small value in the steady state when Q is small and all the amplitude is determined by the t / T signal; If it is required to find the maximum value of the signal Q from the output of the object, then the additional signals from blocks 6i 6 and should have the form

, i (U, Q) signU ,,

where the constant auxiliary signal UQ must be chosen to be obviously larger than Q.

By introducing discretely variable from period to trial period, the amplitude of the probe signals, depending on the output signal of the object Q, the search process time is reduced, proportionality of approach to the extremum point is reached (the average movement speed at the beginning of the search process is significant, and then decreases) and the amplitude of the oscillations applied to the inputs of the object in a steady state is reduced, which distinguishes the proposed method from the known solution.

Invention Formula

The extremum search method based on the formation of periodic pulse-width modulated control probe signals, the period and duration of which change in proportion to the derivative of the output signal, characterized in that, in order to improve speed and search accuracy, the amplitude of the test control signals change proportionally to the amplitude of the output signal when searching minimum and proportional to the difference between the bias signal and the output signal when searching for the maximum, and the magnitude of the constant signal and choose more output.

Sources of information taken into account in the examination

1. USSR author's certificate number 122936, cl. G 05B 13/00, 1959.

2. Journal of Instruments and Control Systems, No. 7, 1970, p. 27-29 (prototype).

Claims (1)

Claim An extremum search method based on the formation of periodic pulse-width modulated trial control signals, the period and duration of which are proportional to the derivative of the output signal, characterized in that, in order to increase the speed and accuracy of the search, the amplitude of the probe control signals is changed proportionally to the amplitude of the output 744447 signal when searching for the minimum and in proportion to the difference between the constant signal and the output signal when searching for the maximum, and the value of the constant signal but they choose more output.