【正文】 s w hose parameters are identified using an online learning procedure. This procedure bines model identification w ith pre and post identification steps to provide reliable operation. T he controller monitors and evaluates the control performance of the closed loop system. The controller w as imple me nted o n a pro gra mmable lo gic co ntroller (P LC). T he perfor ma nce is illustrated on a field test application for control of pressure on a hydraul ic valve?s 2021 Elsevier Ltd. All rights reserved. Ke ywo rds: Co nt ro l e ng ine e ring。 Fuz z y mo de lling。 Mode lbase d control。 Programma ble logic controlle rs。 He nso n amp。 Murra yS mith amp。 Araki,1998。 Chai,1997。 Bruijn,2021), multiplemodel control (Dougherty amp。 Gundala, Hoo, amp。 Sebor g, 1994。 gla nd amp。 Jennings, 1995). The main purpose is to simplify controller configuration by partial automation of the missioning procedure, w hich is typically performed by the control engineer. ABS solve difficult problems by assigning tasks to w orked softw are age nts. T he so ftw are age nts are characterized b y properties suc h as autonomy (operation w ithout direct intervention of humans), social ability (interaction w ith other agents), reactivity (perception and response to the environme nt), proactiveness (goal directed behaviour,ta ki ng the initiative), etc. This w ork does not address issues of ABS theory, but rather the application of the basic concepts of ABS to the field of process systems engineering. In this context, a number of limits have to be considered. For example: initiative is restricted, a high degree of reliability and predictability is dema nded, i nsi ght i nto the proble m do mai n is limited to the sensor readings, specific hardw are platforms are used, etc. The ASPECT controller is an efficient and user friendly engineering tool for implementation of parameter scheduling control in the process industry. The missioning of the controller is simplified by automatic experimentation and tuning. A distinguishing feature of the controll er is that the algorithms are adapted for implementatio n on PLC or open controller Industrial hardw are platforms. The controller parameters are automatically tuned from a nonlinear process model. The model is obtained from operating process signals by experimental modelling,using a novel online learning procedure. This procedure is based on model identification using the local learning approach (MurraySmith amp。 gland amp。 Section 3 gives a brief description of the CT。 Johansen, 1997). The local learning approach is based on the assumption that the par ameters of all local models w ill not be esti mated i n a single re gression operation. Co mpared to the global approac h it is less prone to the problems of ill conditioning and local minima. This method is w ell suited to the needs of industrial operation (intu itiveness, grad ual b uilding o f the no nlinear model, modest co mp utatio nal de ma nds). It enables inve ntor y o f the local models tha t are no t esti mated properly d ue to insufficient excitation. It is efficient and reliable in early configuration stages, w hen all local models have not been estimated yet. On the other hand, the convergence in the vicinity of the optimum is slow . Therefore, it is likely to yield a w orse model fit than methods employing nonlinear follow ing briefly describes the proce dure. Model identification is performed for each selected local model (denoted by the index j) separately. The initial estimated parameter vector ^ ,jMIA? is copied fro m the ac tive MF M, a nd the covariance ma trix ^ ,jMIAp is initialized to 105 I (identity matrix). The FLS (fuzzy least squares) estimates, ^ ,jMIA? and ^ ,jMIAp are obtained usi ng w eighted leastsq uares identificatio n, w ith bj( k) used for w eighting. The calculation is per formed recursively to avoid matrix inversion. The FIV (fuzzy instrumental variables) estimates, ^ ,jMIA? and ^ ,jMIAp are calculated using w eighted instrumental variables identification. In order to prevent res ult de grad a tion b y noise, adead zo ne is used i n eac h s tep of FIV and FLS recursive Esti matio n. T he vector o f para meters and the covariance matri x are upda ted only if the absolute w eighted difference betw een the process output and its prediction is above the configured noise threshold. In case of lack o f e xcita tion i n the bra nc h fro m u to y or i n the model bra nc h fro m v to y (or w hen measured disturba nce is not present at all), variants of the method w ith reduced parameter estimate vectors are used. . Verification/validation This step is performed by paring the simulation output of a selected local model w ith the actual process output in the proximity of the local model position. The normalized sum of mean square errors (MSEj) is calculated. The proximity is defined by the membership functions bj. For each of the selected local models, this step is carried out w ith three sets of model parameters: ^ ,jMIA? ^ ,jFLS? and ^ ,jFIV? the set w ith the low est MSEj is selected. Then, global verification is performed by paring the simulation output of the fuzzy model including the selected set w ith the actual process o utp ut. T he nor malized s um of mea n sq uare errors (MSEG) is calculated. If the global verification result is improved pared to the initial fuzzy model, the selected set is sent to the MIA as the result of online learning, otherw ise the original set ^ ,jMIA? remains in use. For each processed local model, the MIA receives the MSEj, w hich serves as a confidence index, and a flag indicating w hether the model is new or not. . Model structure estimation Tw o model structure estimation units are also included in the OLA. The deadtime unit (DTU) estimates the process time delay. Th