Adaptive strategy of error anomaly processing in human simulated intelligent control*

Run MA，Qian WU，，Chongqing 400， School of Computer Science and Engineering，，，

1.Introduction

The human has accumulated rich experience in the long process of evolution，in which，it also includes the experience of control behavior adaptability on the nature and human society，such as the ability of learning，observation，understanding and knowledge etc，and therefore，intelligence and wisdom of human beings can be said to be endless.Based on online feature identification and process characteristics memory it can abstract the error characteristic model of process.In the bottom layer control，because the process error and its change rate of control process are physically detectable，it can carry on the online feature recognition and characteristics memory for process control，and thereupon then，it summarizes and constructs up the control model and control algorithm based on human simulated intelligence［1-5］.

2.Model of human simulated intelligence

The basis thinking of human simulated intelligence based model is that it adopts different control strategies for different pattern of system error characteristics.In the generalized control model as shown in Figure 1，r(t)andy(t)are respectively the input and output of control process，process errore(t)=r(t)-y(t)，P(t)is the output of generalized controller.

Figure 1.Generalized control model

The basic control pattern of intelligence model is as follows.

①Ife·≥0 ande+≠0，then it takes the proportion and half derivation pattern(P-HD pat-tern).

②Ife·＜0 ore==0，then it takes the half open-loop pattern(HO pattern).

The above two patterns alternately appear in the actual application，and the corresponding relationship of process error signal and control pattern is shown in Figure 2.

Figure 2.Corresponding relationship of error signal and control pattern

The basic control algorithm of human simulated intelligent controller(HISC)is formed by two control patterns，one is P-HD pattern，and the other is HO pattern.The static characteristic is shown in Figure 3，and in which，the 0-asegment corresponds to PHD pattern and thea-b-c-dsegment corresponds to HO pattern.

Figure 3.Static characteristic and its corresponding input signal

With the continuous piecewise analysis method，it can be summarized as the following mathematical form of expression(1)and expression(2).

In the expressions，Pis the output of the generalized controller HISC(to controlled object)，eis the input(error signal process)of generalized controller HISC，，is respectively the first and second order derivative of input signal errore，em，iis theithextreme value of input signale，kpis the proportional gain，kis the inhibition coefficient，andis a constant value needed keeping atnthtime for the generalized controller output signal0(The extreme memory error)，PHDis half-differential component of output signalpin P-HD pattern，，iis theithextremum ofe˙in P-HD pattern，is the constant value needed keeping atLthtime for half-differential output componentPHDin P-HD pattern.0(The error extreme memory);i，l，nis respectively the natural number.δ is the sensitivity limit of generalized controller HISC.

3.Basic action characteristics of HISC

The expression(1)and(2)in the above clearly expounded the basic action characteristics of HISC.

①Wheneandis the same sign and not equal to zero at the same time，the HISC output is regulated according to the proportion half-differential adjustment regularity based on the original retention value(initial value)¯Pn-1，namely

In which，the characteristic of half-differential control actionPHDchange is shown in Figure 4.

Figure 4.Action process of“half-differential”control strategy

Figure 4 shows that when the derivative of error tends to increase(·≥0)，PHDchanges according to the differential regulation rule，namelyPHD=+kkp.When the derivative of error tends to decrease(·＜0)，PHDkeeps the original value unchanged(but it is proportional toLextreme algebra sum in the pattern)，namely

The half differential strategy simulated the differential predictive ability of human being.The training experience shows that it is necessary to introduce half-differential control strategy.Because whateand˙eis the same sign represents that the system error is in the increased stage，if the error derivative is more and more big，then the proportion control action is too weak，and it is not enough to suppress the error to increase.At this time，if the pure proportional pattern control is adopted，then it is bound to have a dynamic error.Because the proportional control pattern needs that after the error must be increased to a certain extent，it can play an effective role in control action.It can be seen that the introduction of halfdifferential strategy improved the control ability.

②After entering opposite sign pattern betweeneand˙e，the generalized controller HSIC begins to keep the new memory value，namely

It would be adjusted according to small proportion action，and the controller output is as follow.

When the error obtains a certain initial speed and reduces to the half of peak valueem，n，the passageway of input signaleis cut off，it adopts the open-loop control，and the output of controller keeps a certain value(the size is that it is proportional to the n error extremum algebraic sum before cutting off)，namely

The control pattern uses the open-loop control，but it is not fully open-loop HSIC controller，so it is known as the“Half open-loop pattern”，and simplified as“HO pattern”.

4.Adaptive strategy of error anomaly processing

The servo control system based on HISC is shown in Figure 5.

Figure 5.Control system of basic algorithm

Owing to the output of regulating action of two control patterns is based on the keeping memory value，the inappropriate initial retention valueis bound to affect the control effect.

When the task variables such asr，fandG(s)etc takes place to change，it will cause the system process error to increase abruptly in the transient switching process.The keeping valuein the last process may be the unreasonable initial retention value in the new process.The generalized controller has to be able to identify and correct the unreasonable retention valueautomatically.

The keeping memory valueof HISC will gradually converge to an ideal valuefor the system error attenuation.The ideal final value is its steady state output of HISC，and it is only determined by task variables such asG(s)，rand interferencefetc，namely

In which，nis the switching times from P-HD pattern to HO pattern.

Whethercan quickly converge to the ideal valuereflected the tracking quality.The convergence ofis related to initial valueFor example，if

Then the final value of keeping memory valueis

When=0，the system process error and convergent process ofare shown in Figure 6(a).But if≠0 and it is opposite to the polarity of ideal final valueof，namely

Then the dynamic error increase(is corresponding to the unreasonable initial valve position)，and the system process response gets bad is shown in Figure 6(b).

Obviously the effective control action is weakened by not reasonable initial value

From here we can see that when the task variables(r，f，G(s))suddenly change，the system switches to a new dynamic process，it should make initialization processing for the original keeping mem-ory valuen-1，and clean the initial valuen-1.The adaptive strategy is divided into two steps.

Figure 6.Typical response process under different

The first step is to make the recognition for task change or transition process switching.Because the change in any case of system is reflected as the increase of system error，if the system error is always normal attenuation in the process of tracking，the sudden occurrence of abnormal error increased sharply(see Figure 7)，after switching at thenthtimes from HO pattern to P-HD pattern，if it appears one of three situations，then it shows that the task variable has been changed.

Figure 7.Sketch map of error anomaly situation

①Error exceeds previous the first opposite sign signal peak valueem，n-1，and it is in the accelerated rise.

③Error exceeds two times of previous the second same sign signal peak valueem，n-2.

②Error exceeds two times of previous the first opposite sign signal peak valueem，n-1.

The second step is to make the initialization process.There are only two sorts of situations as shown in Figure 7.The one is that the sign ofandis opposite(see Figure 8).

Figure 8.Opposite sign of Pand P* n -1 after task sudden change

This situation is similar to the case shown as in Figure 6(b).Due to being reverse polarity of-1ande(k)，it weakened the effective regulation action(kpe+PHD)of P-HD pattern so as to result in the error caused by sharp increase.In which，there are two possibilities in the relationship betweenP¯n-1andP，·P≤0 and·P＞0.For the former，the original valueshould be cleaned，and for the latter，there must be，and therefore it should merge to next situation to process(Concluding，to reduce not reasonable initial value).

The other is that it is the same sign betweenn-1and，but(see Figure 9).

Figure 9.More difference in same sign of¯Pn-1and

In this situation，due to bias effect being too much，it makesPof P-HD pattern be not able to restrain the dynamical error with different signn-1，and therefore it owns the characteristicn-1·e(k)＜0.The relation between-1andPhas also two sorts of possible situations.

The one is that if-1owns the same sign withP，then it must be，and let-1=P，to make-1andto reduce the gap.

The other is that if-1owns different sign withP，then it can handle according to the situation of-1andPbeing the same sign，and set-1to be as zero initial value.

In addition，if-1owns the same sign withe(k)，then it shows that-1strengthens the P-HD action.This moment，it should modify the effect of memory semi-differentialto strengthen the control action.

5.Conclusion

Finally，we can make a summary for adaptive strategy of error anomaly processing in human simulated intelligent control.In the P-HD pattern，if there is

Then it owns the following expressions.

In which，-1 shows modification of-1，the following is similar to the above.

［1］LI Shiyong.Fuzzy Control Neurocontrol and Intelligent Cybernetics［M］.Harbin:Harbin Industry University Press，2002.

［2］YI Jikai.Intelligent control Technology［M］.Beijing:Beijing Industry University Press，2004.

［3］LI Zushu，TU Yaqing.Human Simulated Intelligent Controller［M］.Beijing:National Defense Industry Press，2003.

［4］XIONG Renquan，QIAO Zhenghong.Control strategy of supply system based on HSIC［J］.Journal of Sichuan ordnance，2012，33(1):76-78.

［5］ZHANG Cuiying，TIAN Jianyan.Multi-modal Control of Temperature in the Reheating Furnace Based on Humansimulated Intelligent Control Theory［J］.JOURNAL OF TAIYUAN UN IVERSITY OF SCIENCE AND TECHNOLOGY，2008，29(1):12-15.