Electrical machinery control system main velocity component control and position control two big kinds, former multipurpose to electrical transmission, latter multipurpose to servo-control, also calls the movement control. May summarize quickly for to a control system performance requirement, is steady, the three characters, namely fast response, not ultraharmonic not static difference. In these three between often has the contradiction, in the nearly all control system between the stability and the rapidity has the contradiction. Until now still widely used the PID control method, often in these three between performed in the actual design process compromised, for instance said a sacrifice rapidity enhanced stably and so on.

As a result of the electrical machinery manufacture technology and the electronic accounting machine development, in the servo-control, the movement control domain, the high performance permanent magnetism synchronous motor (PMSM) has become most has the future the electric motor. Unceasingly expands along with the actuation capacity, it has saves the energy, the volume young, the weight is light and so on the merit. But permanent magnetism synchronous motor itself has the certain non-linearity, 强��?the natural prompt denaturation, in addition the servo object has a stronger not determinism and the non-linearity, as well as in the system movement receives question and so on disturbance, therefore based on is controlled the object precise teaching model the conventional control strategy to satisfy with difficulty the high performance permanent magnetism synchronous motor servosystem the control request. Along with the artificial intelligence technology development, the intelligent control has become to the complex object carries on the active control the important method. This article on fuzzily controls, the nerve network control realization intelligence control basic principle does to the outline introduced that, is for the purpose of in the servo-control promoting its application.

First, based on fuzzy logic intelligence servo-control strategy

In the fuzzy logic control essence uses the computer simulation person's fuzzy logic thought function realization one kind of numeral reaction control. Simulates person's intelligence is in fact simulates person's thought, the thought form is the concept, the judgement and the inference. Human's thought has the fuzzy logic the characteristic, therefore with the computer simulation person's fuzzy thought form - fuzzy concept, the fuzzy judgement and the fuzzy inference, is the fuzzy control thought science foundation, again unifies with the reaction control theory may realize the fuzzy control.

In the traditional PID control system design needs to produce is controlled the object the precise model, because the model inaccuracy and the determinism cannot affect the PID control performance. On the contrary, the fuzzy control does not need to know is controlled the object the precise model, it is based on the control system input/output data causal relation fuzzy inference control. These causal relation is the people operates, the control system experience summary, with "if (condition) then (conclusion)" or wrote "IF (A) THEN (B)" control rule form. A, B separately express the system input and the output variable. The usual fuzzy controller input variable selects erroneous E and the erroneous change, but outputs for controls U, such fuzzy controller is called the two-dimensional fuzzy controller.

The fuzzy controller input output variable is different with the PID control use value variable, but uses the language variable, for example: Negative big (NB), negative center (NM), negative small (NS), zero (ZE), small (PS), center (PM) and is honorable (PB). The language variable is good at describing the fuzzy concept, and through fuzzy set expression. Control rule is composed which by the language variable union, if "IF E=NB or NM and  E=NB or NM THEN U=PB", namely expressed "if error for negative big either negative center also erroneous change for negative big or negative center, then control quantity for is honorable" control rule. The all controls experience all summary is the rule, then constitutes the fuzzy control ruleset (storehouse). It may use a fuzzy matrix representation, this fuzzy matrix may be called is controlled the object the fuzzy model.

In the implementation controlled process, the computer unceasing sampling, obtains erroneous, the erroneous change precise quantity after the computation, becomes through fuzzy quantification processing it erroneous and the erroneous change input fuzzy quantity, then obtains the control quantity through the fuzzy logic inference the fuzzy quantity. This fuzzy control quantity needs again to transform into the precise quantity, in order to is controlled the object to exert the control.

The fuzzy control is not based on is controlled the object precise model the control mode, therefore has a stronger robustness, its stable state precision may accumulate the classification method through the introduction intelligence to achieve requests precision. In addition, but also may unifies the fuzzy logic inference and the PID control, carries on the auto-adapted adjustment to the PID controlled variable, realizes the non- static track servo-control.

Second, based on nerve network intelligence servo-control strategy

The artificial nerve network is uses the computer simulation humanity cerebrum nervous system the joint mechanism, but designs one kind of information processing network architecture, generally is called the nerve network (NN). In the nerve network the most basic unit is the nerve cell, is called the neuron. It is more than one kind of inputs lists output information processing unit, including input processing, the activation processes and outputs processes three parts. From the control viewpoint, the neuron model by the weighting accumulator, the single input list output linearity dynamic system and the static nonlinear function is composed, they simulate the nerve cell synthesis process information 突变�? 饱和�?non-linear characteristic.

The massive neurons through layered, the netted joint constituted the nerve network. Layered network is a kind of most basic network architecture form, like front to (�? the network is one kind of layered network, divides the input level, the concealment level (may have multilayer) and the output level. In each all includes certain neurons, the input level, the output level neuron integer decided by the question, but conceals the level the layer and each nerve number goal selection general basis remains unresolved the question complex degree through to experience and to test determined. Theoretically already proved that, in front of three can approach the free non-linear continuous function to the network by the free precision.

Why can the nerve network have the such outstanding characteristic? Is because constitutes by the massive neurons the network can act according to some kind of study rule, through adjusts between the neuron the joint intensity (weight) unceasingly to change the network to approach the performance, namely the nerve network has the extremely strong non-linear mapping ability. Because of this, the nerve network in the intelligent control, the pattern recognition, the breakdown diagnosis, the system recognized and so on the domain has obtained the widespread application.

A nerve network may act according to an unknown complex object the input output data to carry on to its model recognizes, the process which recognizes is in fact the nerve network through some kind of study algorithm unceasingly adjusts between the neuron the joint power value, finally achieved expected the precision to this unknown object dynamic characteristic approaching, the model which it recognizes is not, but is the concealment which the demonstration produces in nerve network power matrix W which the study obtains.

How through nerve network realization servo-control? Because the permanent magnetism synchronous motor model precisely describes with difficulty, might as well supposes its input output relations to be possible to use nonlinear function f () to express. Uses this electrical machinery three in front of to the nerve network the input output data to carry on the training, the training process is through the continual readjustment network in joint power value between the neuron, let the network remember these input output sample data through the power matrix. After appears the new input output data, the nerve network can through the study algorithm, the adjustment joint power value and the guarantee original approaches the precision. Like this trains nerve network dynamic characteristic g which and unceasingly studies () in the essence is object counter model f-1 (). When uses such nerve network directly makes the controller, as a result of g () f () =f-1 () f () =1, may realize lacks the precise model the object to the input follow-up control function. The literature [ 6> has studied through the wavelet nerve network to the asynchronous servo electrical machinery position intelligent control question, and pointed out this method is easy to promote to other forms servos.

Third, other forms intelligence servo-control strategies

Except above fuzzy servo-control, outside nerve network servo-control strategy, but also has based on the expert controls the thought the expert servo-control, based on the rule servo-control, the intelligence servo-control which compensates based on the forecast and so on. In brief, many forms intelligence control methods all may use in the servo-control, and has the better control performance.