Soft computing methods for control and instrucmention

Institute of Intelligent Power Electronics Publications Espoo, May 1999 Publication SOFT COMPUTING METHODS FOR CONTROL AND INSTRUMENTATION Thesis for the degree of Doctor of Science in Technology Xiao-Zhi Gao Institute of Intelligent Power Electronics Publications Espoo, May 1999 Publication SOFT COMPUTING METHODS FOR CONTROL AND INSTRUMENTATION Xiao-Zhi Gao Dissertation for the degree of Doctor of Science in Technology to be presented with due permission for public examination and debate in Auditorium S4 at Helsinki University of Technology (Espoo, Finland) on the 21st of June, 1999, at 12 o’clock noon Helsinki University of Technology Department of Electrical and Communications Engineering Institute of Intelligent Power Electronics Distribution: Helsinki University of Technology Institute of Intelligent Power Electronics P O Box 3000 FIN-02015 HUT Tel +358-9-451 2434 Fax +358-9-460 224  Xiao-Zhi Gao and Helsinki University of Technology ISBN 951-22-4529-9 ISSN 1456-0445 Libella Painopalvelu Oy Espoo 1999 i Abstract The development of soft computing methods has attracted considerable research interest over the past decade They are applied to important fields such as control, signal processing, and system modeling Although soft computing methods have shown great potential in these areas, they share some common shortcomings that hinder them from being used more widely For example, neural networks, a component of soft computing, often suffer from a slow learning rate This drawback renders neural networks less than suitable for time critical applications Therefore, the objective of this thesis is to explore and investigate the soft computing theory so that new and enhanced methods can be put forward The applications of soft computing in control and instrumentation are also studied to solve demanding real-world problems In this work, the existing soft computing techniques have been enhanced, and applied to control and instrumentation areas First, new soft computing methods are proposed A Modified Elman Neural Network (MENN) is introduced to provide fast convergence speed Based on Muller’s method, we propose a new reinforcement learning method, which can converge faster than the original algorithm As a fusion of fuzzy logic and neural networks, a new fuzzy filter using the selforganizing map to fine tune the membership functions is studied The new soft computing schemes presented in this thesis improve the performance of those earlier methods Second, we study the MENN-based identification and control problems A dynamical system identification scheme as well as a trajectory tracking configuration using the MENNs are discussed, respectively Our MENN-based identification structure belongs to the ‘black box’ identification catalogue It has the advantageous feature of not knowing the exact order of the system The inverted pendulum is utilized here as a testbed for the MENN-based trajectory control scheme It is shown that neural networks are very efficient in dealing with nonlinear system identification and control In addition, they need little prior information of the plant to be identified or controlled However, the existence of local minima, under-fitting, and over-fitting may reduce the identification and control accuracy Third, the applications of soft computing methods in velocity and acceleration acquisition in motion control systems are discussed The aforementioned fuzzy filter is applied to filter out the velocity noise in the feedback loop without introducing any harmful delay This could lead to a better servo control performance Moreover, we construct a neural network-based acceleration acquisition scheme to obtain clean and delayless acceleration signals Our method has the advantage ii of implicit adaptation It can be used for any slowly altering velocity signal, which overcomes the drawback of polynomial predictor-based approaches Finally, the power prediction and regulation in mobile communications systems are studied with these soft computing methods An optimal neural predictor is selected by applying the Predictive Minimum Description Length (PMDL) principle A Temporal Difference (TD)-based multi-step ahead prediction scheme is also considered for the fading signals Simulations demonstrate that the neural predictors offer better results than conventional filters Their weakness is the accompanying high computational complexity We introduce an MENN-based power controller at the base station, which takes advantage of the inverse radio channel model On the other hand, the efficiency of the power controller employed is clearly limited by the single-bit command transmission mode Meanwhile, an Embedded Fuzzy Unit (EFU) is proposed to provide versatile alternatives The EFU can effectively tackle the bottle-neck of incremental command transmission mode, and thus achieve better power regulation Additionally, a comparison between the conventional and soft computingbased power control methods is made This research gives us useful guidance in determining appropriate power regulation configurations in terms of effectiveness and complexity In conclusion, the theory and applications of soft computing methods are studied in this thesis Nevertheless, our research is not aiming at implementation of the proposed schemes Therefore, the validation is verified only by numerical simulations All the simulations are based on simplified application models without consideration to practical details iii Preface It was a great pleasure for me to work on this thesis I hope, too, readers will find it useful and comfortable to read First and foremost, I am truly indebted to my advisor, Professor Seppo J Ovaska, who guided me throughout the whole of my research work Professor Ovaska has proposed countless suggestions during this thesis writing procedure as well as through my course study I have learnt quite a lot from his comments, which are always inspiring and fruitful I also thank him for providing me with the invaluable opportunity enabling me to come to Finland and study for my D.Sc (Tech.) degree at the Helsinki University of Technology Here, I can only extend my best wishes to him in his future career Thank you very much indeed, Seppo And, I want to thank all the personnel at the laboratory of Electric Drives and Power Electronics Professor Jorma Kyyr is especially thanked for his warm-hearted help Secretary Leena V is nen deserves my gratitude for helping me cope with practical matters including all the difficulties that a student may encounter in a foreign country Laboratory manager Ismo Vainiom ki and laboratory technician Ilkka Hanhivaara are thanked for their efficient efforts in creating such a fresh and pleasant research environment I really enjoy working in this laboratory I am very grateful to my colleagues, Jarno Tanskanen, Sami V liviita, Vlad Grigore, and Adrian Dumitrescu, with whom I had so many helpful discussions Special thanks go to Jarno Tanskanen and Sami V liviita for their cooperative work in our joint publications I am also very much obliged to the secretary of the Graduate School of Electronics, Telecommunications, and Automation (GETA), Marja Lepp harju, for her kind assistance in my graduate study at GETA Joe O’Reilly is thanked for checking the language used in this thesis All the errors possibly remaining in the text have been introduced by me alone at the final stages of revision Last but not the least, I want to say a heart-felt ‘thank you’ to my parents and brother They play an important role in my life, study and work, both morally and financially The author wishes to express his deep thanks to the Center for International Mobility (CIMO) and GETA for their financial support Otaniemi, May 1999 Xiao-Zhi Gao iv Table of Contents Introduction…………………………………………………………………………………… 1.1 Definition of Soft Computing……………………………………………………………… 1.2 Intelligent Control………………………………………………………………………… 1.3 Aim of This Dissertation…………………………………………………………………… Introduction to Soft Computing Methods……………………………………………………….12 2.1 Neural Networks…………………………………………………………………………… 12 2.1.1 Back-Propagation Neural Network………………………………………………… 13 2.1.2 Elman Neural Network……………………………………………………………… 16 2.1.3 Self-organizing Map………………………………………………………………… 19 2.2 Fuzzy Logic………………………………………………………………………………… 22 2.2.1 Basic Theory of Fuzzy Logic Systems……………………………………………… 22 2.2.2 Fuzzy Logic-based Control………………………………………………………… 26 2.2.3 Fuzzy Neural Network……………………………………………………………… 30 2.3 Reinforcement Learning…………………………………………………………………… 33 2.3.1 Single-step-ahead Predictor-based Critic-Actor Algorithm………………………… 35 2.3.2 Temporal Difference Method-based Prediction…………………………………… 38 Related Research………………………………………………………………………………… 42 3.1 Predictive Filtering Methods and Their Applications…………………………………… 42 3.1.1 Predictive Filtering Methods………………………………………………………… 42 3.1.2 Power Prediction in Mobile Communications Systems…………………………… 46 3.1.3 Acceleration Acquisition in Motor Control Systems……………………………… 49 3.2 Neural Network-based Dynamical System Identification………………………………… 53 3.2.1 Neural Network-based Forward Model Identification……………………………… 54 3.2.2 Neural Network-based Inverse Model Identification……………………………… 57 3.3 Neural Network-based Control Applications………………………………………………60 3.3.1 Neural Network-based Control Schemes…………………………………………… 60 3.3.2 Power Regulation in Mobile Communications Systems…………………………… 66 3.3.3 Inverted Pendulum Control………………………………………………………… 71 Summary of Publications……………………………………………………………………… 78 4.1 Neural Network-based System Identification Techniques………………………………… 78 4.1.1 Publication [P1]……………………………………………………………………… 78 v 4.2 Neural Network-based Control Methods………………………………………………… 83 4.2.1 Publication [P2]……………………………………………………………………….83 4.3 Soft Computing Methods in Instrumentation……………………………………………….87 4.3.1 Publication [P3]……………………………………………………………………….87 4.3.2 Publication [P4].…………………………………………………………………… 90 4.3.3 Publication [P5] …………………………………………………………………… 93 4.3.4 Conclusions of Publications [P3]—[P5]………………………………………………96 4.4 Power Prediction in Mobile Communications Systems…………………………………… 96 4.4.1 Publication [P6]……………………………………………………………………….96 4.4.2 Publication [P7]…………………………………………………………………… 101 4.4.3 Conclusions of Publications [P6]—[P7]…………………………………………….103 4.5 Power Control in Mobile Communications Systems…………………………………… 103 4.5.1 Publication [P8]…………………………………………………………………… 103 4.5.2 Publication [P9]…………………………………………………………………… 107 4.5.3 Publication [P10]…………………………………………………………………… 111 4.5.4 Conclusions of Publications [P8]—[P10]………………………………………… 115 4.6 Contribution of the Author……………………………………………………………… 116 Conclusions and Discussion…………………………………………………………………….119 5.1 Main Results……………………………………………………………………………….119 5.2 Scientific Importance of the Author’s Work……………………………………………… 120 5.3 Topics for Future Research………………………………………………………….….… 122 References………………………………………………………………………………………124 Appendix A: Publications [P1]—[P10] Appendix B: Errata vi List of Publications This thesis consists of an introduction and the following ten publications which are referred to by [P1], [P2], …, [P10] in the text: [P1] X Z Gao, X M Gao, and S J Ovaska, “A modified Elman neural network model with application to dynamical systems identification,” in Proceedings of the 1996 IEEE International Conference on Systems, Man, and Cybernetics, Beijing, P R China, October 1996, pp 1376-1381 [P2] X Z Gao, X M Gao, and S J Ovaska, “Trajectory control based on a modified Elman neural network,” in Proceedings of the 1997 IEEE International Conference on Systems, Man, and Cybernetics, Orlando, FL, October 1997, pp 2505-2510 [P3] X M Gao, X Z Gao, and S J Ovaska, “Power command enhancement in mobile communication systems using an embedded fuzzy unit,” in Proceedings of the 1997 IEEE International Conference on Systems, Man, and Cybernetics, Orlando, FL, October 1997, pp 4364-4369 [P4] “Neural networks-based approach for the acquisition of acceleration from noisy velocity signal,” in Proceedings of the 1998 IEEE Instrumentation and Measurement Technology Conference, St Paul, MN, May 1998, pp 935-940 [P5] X Z Gao and S J Ovaska, “A new fuzzy filter with application in motion control systems,” in Proceedings of the 1999 IEEE International Conference on Systems, Man, and Cybernetics, Tokyo, Japan, October 1999, IN PRESS [P6] X Z Gao, “A temporal difference method-based prediction scheme applied to fading power signals,” in Proceedings of the 1998 IEEE International Joint Conference on Neural Networks, Anchorage, AK, May 1998, pp 1954-1959 vii [P7] X M Gao, X Z Gao, J M A Tanskanen, and S J Ovaska, “Power prediction in mobile communication systems using an optimal neural-network structure,” IEEE Transactions on Neural Networks, vol 8, no 6, pp 1446-1455, November 1997 [P8] X M Gao, X Z Gao, J M A Tanskanen, and S J Ovaska, “Power control for mobile DS/CDMA systems using a modified Elman neural network controller,” in Proceedings of the 47th IEEE Vehicular Technology Conference, Phoenix, AZ, May 1997, pp 750-754 [P9] X Z Gao, X M Gao, and S J Ovaska, “Fast reinforcement learning algorithm for power control in cellular communication systems,” in Proceedings of the 1997 IEEE International Conference on Systems, Man, and Cybernetics, Orlando, FL, October 1997, pp 3883-3888 [P10] X Z Gao and S J Ovaska, “Comparison of conventional and soft computing-based control methods in a power regulation application,” in Proceedings of the 1998 IEEE International Conference on Systems, Man, and Cybernetics, San Diego, CA, October 1998, pp 20752082 account In conclusion, our research highlights the application of modern power control methods in mobile communications systems with respect to the severe command transmission constraint This paper is organized as follows In Section 2, we first briefly discuss the working principle of the DS/CDMA Return Channel Error Base Station Mobile Station KTp Loop Delay Power Controller Step Size P Transmitting Power Tp P(t ) Desired Level dB x(t) Integrator Fading Signal Fig Power control structure in DS/CDMA cellular systems system and the conventional ‘bang-bang’ power control method The soft computing-based power regulation schemes are introduced in Section Only the necessary outlines of these control methods are provided Detailed description of the algorithms can be found in the literature In Section 4, we compare the selected performance criteria of the combinations of different power control methods; conventional and advanced, as well as different transmission modes; full command, single bit, and fuzzy command enhancement Simulation results are also provided in this section Finally, in Section 5, we draw some conclusions and remarks to conclude the paper equivalent in-phase and quardrature components is approximately kHz, and the applied vehicle speed is km/h (a typical ‘slow speed’ in an urban environment) B Conventional Feedback Power Control As mentioned above, because of the complex nonlinear and stochastic characteristics of the fading channel, a power control method is needed to regulate the received power level at the base station Fig shows the conventional feedback power control scheme To put it into more details, the user’s transmitting signal power P( t) (dB) is updated by a fixed step ∆p (dB), typically 0.5 dB, every T p seconds at the mobile station T p is the power control sampling period At the sampling point t, the signal power received at the base station is P(t) + x(t ) (dB), where x(t ) (dB) is the fading component The received signal power is then compared to the desired power level, which is always assumed to be dB at the base station Based on the observed difference, a hard-quantized power command bit is constructed, and transmitted back to the mobile user over the return path More precisely, if the received signal power is lower than the desired power level, a positive power command, e.g., +1, will be sent to the mobile user to increase the current transmitting power level by ∆p ; if the received signal power is higher than the desired power level, a negative power command, e.g., -1, will therefore be sent to the mobile user to reduce the current transmitting power level by ∆p ; if the received signal power is just dB, the current transmitting power will be kept constant by a zero power command This constitutes, therefore, a typical POWER CONTROL IN CELLULAR COMMUNICATION SYSTEMS A Fading Channel Model A wealth description of modeling of the Rayleigh fading radio channel is given by Jakes in [1] In this paper, our channel simulator assumes the superposition of plane waves whose arrival angles are uniformly distributed Different plane waves are associated with different Doppler shifts ranging from the minimum to the maximum specified by the mobile speed The simulator consists of low frequency oscillators at these Doppler shift frequencies, and the frequency distribution results in a satisfactory approximation of the Rayleigh fading The in-phase and quadrature components are formed by summing the appropriately weighted oscillator outputs After multiplication with the corresponding carrier component, the signal is centered at the carrier frequency Our carrier frequency is 1.8 GHz, the sampling rate of the baseband 2076 ‘bang-bang’ control scheme As a matter of fact, full power command transmission could also be in principle used here, but it would reduce the effective channel capacity, and is therefore not practical ing fading signal In other words, at iteration t, the controller can only compensate for the power error caused by the fading signal x(t − 1) , which is known as feedback at that time instant That is one reason why such kind of ‘bang-bang’ controller always produces slow transient response and considerable static error Therefore, a natural idea arises to estimate one-step-ahead prediction of the fading signal in order to improve the system performance This principle forms the foundation of our prediction-based power control In brief, the essential idea of conventional power regulation is that when the received power is below the reference level, the mobile user’s current power is increased by the power command On the other hand, if the received power is above the reference level, this difference is employed to decrease the mobile user power Nevertheless, the main drawback of such a simple scheme is that the compensation commands are always one step delayed of the fading signal, which definitely causes over shoots and steady tracking error In addition, in practical cellular systems, due to the restraint of the whole channel transmitting bit rate, the power commands cannot be implemented in full mode In other words, the power command output must be in discrete values instead, for example, 0.5 dB at each sampling point This restriction degrades greatly the actual closed loop performance Neural network is considered to be an attractive candidate as a prediction tool because of its strengths of approximation, adaptation, and generalization Among numerous models available, the Elman neural network was first presented for the purpose of speech processing [9] In fact, it belongs to the class of locally recurrent and globally feedforward neural networks The distinguishing self-connections of the context nodes in the Elman neural network make it sensitive to the history of input data, which is very useful in dynamical system identification, as well as time series data modeling and prediction Moreover, a modified Elman neural network model (MENN) was proposed by the authors in a recent paper [10] to enhance the dynamical characteristics, and further improve the convergence speed It was shown that the MENN converges much faster than the original model, and has better dynamical characteristics as well These advantages make it well suited to be employed on-line More information about the MENN with applications in multi-step ahead fading power signal prediction, nonlinear system identification and adaptive control can be found in references [11], [10], and [12], respectively To overcome these disadvantages, other efficient power control algorithms employed at the base station, such as adaptive fuzzy logic [7] and reinforcement learningbased control approaches [4], have been proposed as well It is worth pointing out that whatever the advanced control algorithm is used, its efficiency is clearly limited by the incremental command transmission mode In order to relieve this problem, we construct an incremental power command enhancement unit using fuzzy logic at the mobile station to generate smoother power output with no regard to the control structures applied at the base station [6] Specially, a soft computing-based power control method, neural network-based prediction control together with fuzzy command enhancement unit, will be explored in the next section Inspired by the above facts, we construct the MENNbased fading signal prediction training scheme as shown in Fig In Fig 2, x(t) represents the fading time series, and p(t) is the MENN’s prediction output Each x(t) , however, is associated with a p( t) , the prediction of the next step fading signal The training phase of our MENN-based predictor is divided into two stages: offline generalized training and on-line specialized training The off-line training is to give a rough model of a fading channel Therefore, training data is collected in advance In that situation, actual x(t + 1) is applied as the desired output for the MENN during the weight adaptation procedure When e(t ) , the error between the reference fading signal and the actual prediction, reaches a satisfactory level, the training phase is terminated, and the weights of the MENN are used as initial weights for next on-line training Due to the time varying characteristics of the radio channel as well as variable mobile velocities, on-line adaptation is necessary for the predictor Nevertheless, the true fading signal x(t + 1) is not available for the on-line case Therefore, we have to use the history data pair, x(t − 1) and x(t) , to further fine tune SOFT COMPUTING-BASED POWER REGULATION SCHEMES In recent years, soft computing methods including neural networks and fuzzy logic have played an emerging role in the field of power prediction [8] and control [5] in mobile communication systems Numerous design schemes, for example, adaptive fuzzy control and neural networksbased inverse dynamical system control [7], [5], have been proposed accordingly However, we only discuss two representative power regulation schemes using soft computing methods in this paper: neural network-based prediction control approach at the base station and fuzzy logic-based power command enhancement unit employed at the mobile station A Neural Network-based Prediction Power Control As mentioned above, the serious drawback of conventional mobile power control is that the compensation commands are normally one step delayed of the disturb- 2077 the predictor in order to track the changes of the fading channel different control methods at the base station The fuzzy logic-based embedded power command enhancement unit [6] is illustrated in Fig x(t 1) p(t ) x(t) One-Bit Power Command u(t 1) u(t) Integrator MENN Z u(t k) e(t) Fuzzy Logic Inference Unit Enhanced Power Command u˜(t) Transmitter Learning Method Mobile Station Fig The MENN-based fading power signal prediction scheme Fig The Embedded Fuzzy Unit (EFU) at the mobile station The prediction output of our MENN-based predictor can then be used at the base station to compensate for errors resulting from fading signals The structure of our MENN-based predictive controller is shown in Fig u(t ) is the power control output The simulation results are demonstrated in Section We point out, on the other hand, that this prediction control scheme must work cooperatively with the practical power command transmission mode In addition, other neural networks are also available for this prediction task For example, in [8], the authors have proposed a hybrid neural networks-based power prediction scheme Our thought of fuzzy command enhancement unit is unique because we switch the attention to the controller design at the mobile station rather than at the base station Nevertheless, the individual power increments received at the mobile station contain less information than the ideal full commands Only the discrete power step commands +1, 0, and -1 are available In addition, it is required that the control algorithms employed at the mobile station can be neither complex nor time-consuming Complex and time-consuming control schemes would lead to high implementaion cost and power consumption Since with a typical implementation of look-up table fuzzy logic has the advantage of simplicity and effectiveness, the embedded fuzzy unit (EFU) is set up to create enhanced power commands MENN-Based Predictive Controller p(t ) x(t) MENN-Based Predictor u(t) In Fig 4, u(t ),u(t −1),Lu(t − k) are the current and past power commands at the mobile station ˜u(t) is the enhanced power command output of our EFU We use a Tapped Delay Line (TDL) to store the current and past power commands for the input of the EFU The principle of our EFU scheme is making full use of the current and past power control commands received from the base station These power commands give us the only clue about the variation caused by the Rayleigh fading channel Thus, judged from this information, appropriate power commands can be generated by using the fuzzy inference unit Following the frame of the design of a fuzzy logic-based control system, our EFU is explained by the steps as below -1 Fig The MENN-based predictive controller B Fuzzy Logic-based Power Command Enhancement Unit Compared with the traditional methods, the neural networks-based prediction control approach can provide a better power response However, one weak point of the mentioned neural prediction control system is that in order to acquire the expected improved regulation performance, the power control commands must be sent out in full command mode This necessity would definitely either increase the whole transmitting bit rate or reduce the overall data transmission capacity To alleviate this problem, we build an incremental power command enhancement unit using fuzzy logic at the mobile station to generate smoother power output while cooperating with the Step Select the number of input and output variables In the classic fuzzy regulator, there are usually no more than three input variables [13], i.e., the feedback control error e(t) , as well as the first and second order derivaÝ (t) , respectively, tives of the control error, eÝ(t) and eÝ 2078 since many inputs would increase the fuzzy inference computation burden, and result in even harder rule combination in the consequent part Therefore, we naturally choose u(t) , u(t − 1) and u(t − 2) as the inputs of the EFU The only output of our EFU is the enhanced power command The EFU in this way realizes the nonlinear mapping: [ ] ˜u(t ) = f u(t ), u(t − 1), u(t − 2) different roles in the final decision output However, in our fuzzy unit, we set all the weights to be Step Select the defuzzification method There are several defuzzification methods available to choose from, e.g., the Mean of Maximum Method (MOM) and the Center of Area Method (COA) We use the COA method here, because it is argued that the COA strategy yields superior results over other methods [15] (1) Step Determine the types and parameters of the membership functions of the input and output linguistic variables Step Fine tune the EFU Since the principle of fuzzy control is solely based on expert’s experience, there is no analytical way to determine all its parameters, although some kind of selforganized scheme has been proposed [16] Consequently, the parameters of our final EFU could be fine tuned during the running of the systems in order to get the optimal performance (by optimal here we mean that the received power level should be as flat as possible) This step is achieved by trial and error That is adjusting the parameters of the membership functions and adding new or deleting relevant fuzzy rules by observing the received power response A detailed description of the proposed fuzzy power enhancement is given in [6] Normally, we define the following seven fuzzy partition terms for both the received and enhanced power commands: NB, negative big; NM, negative medium; NS, negative small; ZE, zero; PS, positive small; PM, positive medium; PB, positive big More membership functions are helpful in improving the resolution, but they turn the rule combination more complex as well The type of the membership functions can be selected among bell-shaped, triangle-shaped, Gaussian-shaped, etc The parameters of the corresponding membership functions, however, are always application tailored Step Define the fuzzy control rules In this section, we have concisely introduced two soft computing-based schemes for power regulation Their efficiency has been demonstrated individually in the authors’ previous papers However, no careful examination is made on how well they can perform when applied cooperatively, and no comprehensive comparison between these advanced control methods and the conventional solution is available Thus, we will present illustrative computer simulation results on these aspects in the next section In our EFU, the fuzzy control rules have the general prototype [14] as follows: IF u(t) is A0 AND u(t − 1) is A1 AND u(t − 2) is A2 THEN u˜(t ) is B, (2) where A0, A1, A2 and B are the fuzzy partition terms defined above Although the design of fuzzy rules is human-dependent, there exist some heuristic ways To overcome the effect of the deep Rayleigh fading, our fuzzy rules in the EFU are based on the following ideas: Principle 1: if the mobile station received consecutive large power commands, the enhancement power command should also be large Principle 2: if the mobile station received consecutive small power commands, the enhancement power command should also be small Principle 3: if the mobile station received consecutive increasing power commands, the enhancement power command should be more increased Principle 4: if the mobile station received consecutive decreasing power commands, the enhancement power command should be more decreased One typical example of these fuzzy rules is: SIMULATION The structure and parameters of the Rayleigh fading radio channel has been given in Section II The fading signal sequence with an applied vehicle speed of km/h is shown in Fig We first investigate the performance of the conventional power regulation scheme under the circumstances of ideal full command and practical single-bit transmission modes Figs and show the received power at the base station when using the ‘bang-bang’ control method with full command and single-bit transmission modes, respectively As we discussed above, due to the intrinsic delay characteristics of the ‘bang-bang’ control, there are still some variations in the received power even in Fig However, it is clearly visible that compared with the full command control, the single-bit mode gives us much worse received power with considerable oscillations Here, we apply the standard deviation of the received power as the performance criterion The standard devia- IF u(t) is PB AND u(t − 1) is PB AND u(t − 2) is PB (3) THEN u˜(t ) is PB It is also desirable that relative weights [13] can be assigned to the output of each fuzzy rule to make them play 2079 tions of the received power for the two aforementioned cases are 0.52 and 1.34, respectively the main obstacle for applying advanced power control techniques at the base station Our EFU is very effective in attacking this problem In practice, the optimal power regulation configuration is the combination of the MENN-based prediction control at the base station together with the fuzzy command enhancement unit at the mobile station However, if we consider the computational complexity of on-line neural network training, the configuration of ‘bang-bang’ control with the fuzzy command enhancement unit is certainly an attractive choice Next, the received power by using ‘bang-bang’ controller together with our fuzzy power command enhancement unit is illustrated in Fig The high peaks in the received power curve in Fig are moderately reduced by the fuzzy unit The standard deviation of the received power is now 1.05 Actually, the performance criterion is improved by 22% by using the fuzzy unit To examine the soft computing-based control approaches, the effectiveness of the MENN-base prediction power control method is validated We first take 1000 sampling points from Fig This time series is divided into two sequences, each with 500 samples The first sequence is used as the training data for the MENNbased predictor, and the second sequence, namely validation data, is applied to examine the generalization of our predictor The prediction results for the validation fading signal are given in Fig Solid line represents the actual signal while dotted line indicates the one-step-ahead prediction output We conclude from Fig that the MENN provides a reliable prediction of the fading power signal, which is not included in the training phase It is well known that the power control performance is determined by the prediction accuracy As we discussed above, for time varying mobile speeds, e.g., from km/h to 50 km/h, the MENN-based predictor must be adapted on line This technique is still under simulation investigation CONCLUSIONS In this paper, we made an interesting comparison between the conventional and soft computing-based power regulation schemes by demonstrative simulations With the consideration of the practical single-bit transmission mode, our work provides a useful guide of choosing appropriate power regulation configurations It is illustrated that the fuzzy enhancement unit is promising in obtaining high quality power regulation The cellular system performance can be improved by utilizing not only the advanced controllers at the base station but also some alike power command enhancement schemes at the mobile stations Our future research includes the adaptive MENN power predictor, behavior of the fuzzy enhancement with the reinforcement learning controller of [4], and the effects of different mobile velocities on our power regulation configurations REFERENCES The received power levels by employing this MENNbased prediction method with full command, single-bit transmission modes, as well as the fuzzy logic-based command enhancement unit are shown in Figs 10, 11, and 12, respectively Under the full command mode, the results of Fig 10, which have the standard derivation of 0.44, are clearly better than those of Fig However, if we compare Fig 11 with Fig 7, it is easily revealed that the shapes of the received power signals are almost equal, although the standard deviation of 1.19 is somewhat improved Furthermore, the performance of the MENN-based prediction control with the fuzzy unit is the best under the practical transmission mode restraint, as shown in Fig 12 The corresponding standard deviation is 0.91 Note that although there is still a transient over shoot at the end of the sequence which obviously results from overly deep fading, the average received power response is smooth It is also interesting to observe that with the fuzzy enhancement unit, the standard deviation of ‘bang-bang’ control is even better than that of the MENN-based prediction control scheme in single bit transmission mode, 1.05 vs 1.19 [1] W C Jakes (Ed.), Microwave Mobile Communications New York, NY: John Wiley & Sons, 1974 [2] K S Gilhousen, I M Jacobs, R Padovani, A J Viterbi, L A Weaver, Jr., and C E Wheatley III, “On the capacity of a cellular CDMA system,” IEEE Trans Vehic Technol., vol 40, pp 303-312, May 1991 [3] S Ariyavisitakul and L F Fang, “Signal and interference statistics of a CDMA systems with feedback power control,” IEEE Trans Commun., vol 41, pp 1626-1643, Nov 1993 [4] X Z Gao, X M Gao, and S J Ovaska, “Fast reinforcement learning algorithm for power control in cellular communication systems,” in Proc IEEE International Conference on Systems, Man, and Cybernetics, Orlando, FL, Oct 1997, pp 3883-3888 [5] X M Gao, X Z Gao, J M A Tanskanen, and S J Ovaska, “Intelligent power control for mobile DS/CDMA systems using a modified Elman neural network controller,” in Proc IEEE 47th Vehicular Technology Conference, Phoenix, AZ, May 1997, pp 750-754 [6] X M Gao, X Z Gao, and S J Ovaska, “Power command enhancement in mobile communication systems using an embedded fuzzy unit,” in Proc IEEE In- The standard deviations of the above power control configurations are summarized in table Finally, we reach the conclusion that the single-bit transmission mode is 2080 ternational Conference on Systems, Man, and CybernetProc IEEE International Joint Conference on Neural ics, Orlando, FL, Oct 1997, pp 4364-4369 Networks, Anchorage, AK, May 1998, pp 1954-1959 [7] P R Chang and B C Wang, “Adaptive fuzzy power [12] X Z Gao, X M Gao, and S J Ovaska, “Trajectory control for mobile radio system,” in Proc IEEE 45th control based on a modified Elman neural network,” in Vehicular Technology Conference, Chicago, IL, July Proc IEEE International Conference on Systems, Man, 1995, pp 927-931 and Cybernetics, Orlando, FL, Oct 1997, pp 25052510 [8] X M Gao, X Z Gao, J M A Tanskanen, and S J Ovaska, “Power prediction in mobile communication [13] C C Lee, “Fuzzy logic in control systems: Fuzzy systems using an optimal neural network structure,” logic controller, Parts I and II,” IEEE Trans Systems, IEEE Trans on Neural Networks, vol 8, pp 1446-1455, Man, and Cybernetics, vol 20, pp 404-435, Apr 1990 Nov 1997 [14] E H Mamdani and B R Gaines, Fuzzy Reasoning [9] J Elman, “Finding structure in time,” Cognitive Sciand Its Applications London, UK: Academic, 1981 ence, vol 14, pp 179-211, 1990 [15] M Sugeno and M Nishida, “Fuzzy control of model [10] X Z Gao, X M Gao, and S J Ovaska, “A modicar,” Fuzzy Sets and Systems, vol 16, pp 103-113, 1985 fied Elman neural network model with application to [16] L.-X Wang and J M Mendel, “Generating fuzzy dynamical systems identification,” in Proc IEEE Interrules by learning from examples,” IEEE Trans Systems, national Conference on Systems, Man, and Cybernetics, Man, and Cybernetics, vol 22, pp 1414-1427, Beijing, P R China, Oct 1996, pp 1376-1381 Nov./Dec 1992 [11] X Z Gao, “A temporal difference method-based prediction scheme applied to fading power signals,” in Table Standard deviations (dB) of various power control configurations Transmission Modes Control Algorithms ‘Bang-Bang’ Control Full Command Fuzzy Enhancement Unit Single Bit Command 0.52 1.05 1.34 MENN-based Prediction Control 0.44 0.91 1.19 10 -10 Received Power (dB) Fading Signal (dB) -5 -15 -20 -2 -25 -4 -30 -35 -40 -6 500 1000 1500 Time in Samples 2000 2500 -8 Fig The fading power signal caused by the transmitting channel 50 100 150 200 250 300 Time in Samples 350 400 450 500 Fig The received power level using ‘bang-bang’ control under full command transmission mode 2081 6 4 2 Received Power (dB) Received Power (dB) -2 -2 -4 -4 -6 -6 -8 50 100 150 200 250 300 Time in Samples 350 400 450 -8 500 Fig The received power level using ‘bang-bang’ control under single bit command transmission mode 50 100 150 200 250 300 Time in Samples 350 400 450 500 Fig 10 Received power level using prediction control method with full power command mode 8 6 Received Power (dB) Received Power (dB) -2 -2 -4 -4 -6 -6 -8 50 100 150 200 250 300 Time in Samples 350 400 450 500 -8 Fig Received power level using ‘bang-bang’ control and fuzzy power command enhancement unit 50 100 150 200 250 300 Time in Samples 350 400 450 500 Fig 11 Received power level using prediction control method using single bit power command 10 Received Power (dB) Fading Signal (dB) -5 -10 -15 -2 -4 -6 -20 50 100 150 200 250 300 Time in Samples 350 400 450 500 -8 Fig One-step-ahead prediction of the fading signal using the MENN-based predictor 50 100 150 200 250 300 Time in Samples 350 400 450 500 Fig 12 Received power level using prediction control method using fuzzy power command enhancement unit 2082 Appendix B Errata [P1] • On page 1379, Equations (17) and (18) should read [ OContext ( k ) = a W 1I MENN ( k ) + W 3OContext ( k − 1) ] and O MENN ( k ) = W 4OContext ( k ) + W 2OContext ( k + 1) , where I MENN and O MENN are the input and output of the MENN, respectively OContext is the output of the context nodes • On page 1379, Figure should be replaced with Figure 4.2 in this thesis [P2] • On pages 2508 and 2509, both the initial states x1 ( ) = pendulum in Examples II and IV should be replaced with π x1 ( ) = , x2 ( ) = 18 [P3] • On page 4368, Figure should be replaced with Figure 4.9 in this thesis [P9] • On page 3887, Figure should be replaced with Figure 4.29 in this thesis • On page 3888, Figure should be replaced with Figure 4.31 in this thesis π , x2 ( ) = of the inverted 60 RESUME Xiao-Zhi Gao Institute of Intelligent Power Electronics Department of Electrical and Communications Engineering Helsinki University of Technology Otakaari A, FIN-02150, Espoo Finland Tel: +358 451 2434 Fax: +358 460 224 E-mail: gao@csc.fi, gao@cc.hut.fi URL: http://www.hut.fi/Units/PowerElectronics WORKING EXPERIENCE December 1999  present, Senior Researcher in Institute of Intelligent Power Electronics, Department of Electrical and Communications Engineering, Helsinki University of Technology, Finland Project: Fault diagnosis and detection in electrical drives using soft computing methods; financed by Academy of Finland October 1997  November 1999, Research Scientist in Institute of Intelligent Power Electronics, Department of Electrical and Communications Engineering, Helsinki University of Technology, Finland Project: Soft computing techniques for control and instrumentation; financed by GETA September 1996  September 1997, Researcher in Institute of Intelligent Power Electronics, Department of Electrical and Communications Engineering, Helsinki University of Technology, Finland Projects: Fuzzy-Neuro technique in electrical drives, financed by CIMO Fuzzy-Neuro power prediction and control in next generation mobile communication systems (1996  1997), financed by Nokia and Tekes September 1995  August 1996, Teaching Assistant in Department of Control Engineering, Harbin Institute of Technology, China Taught undergraduate course "Design of Modern Control Systems: Simulation and CAD" July 1993  August 1995, Research Assistant in Department of Control Engineering, Harbin Institute of Technology, China -1- Projects: Study on neural networks theory with application in high precision servo control systems Research on neural networks-based identification of polluted water treatment procedure Development of Control Systems Computer Aided Design (CSCAD) software EDUCATION 1999, Ph.D., Department of Electrical and Communications Engineering, Helsinki University of Technology, Finland Thesis: Soft Computing Methods for Control and Instrumentation 1996, M.Sc., Department of Control Engineering, Harbin Institute of Technology, China Thesis: Neural Network Models with Applications in Adaptive Control 1993, B S., Department of Control Engineering, Harbin Institute of Technology, China Thesis: Design and Simulation of Computer-based Servo Control Systems EXPERTISE Neural networks model theory, learning algorithms, and stability analysis Neural networks in time series prediction, dynamical system identification, and adaptive control Fuzzy logic control theory Power prediction & control using neural networks and fuzzy logic in mobile communication systems Intelligent control of electrical drives in servo systems Reinforcement learning method Genetic algorithms and evolutionary programming Control system simulation and CAD software design -2- ACADEMIC ACTIVITIES Reviewer of X European Signal Processing Conference (EUSIPCO 2000), Tampere, Finland, 2000 Presenter of IEEE International Conference on Systems, Man, and Cybernetics, Tokyo, Japan, October 1999 Member of Best Paper Award Committee of 4th On-line World Conference on Soft Computing in Industrial Applications (WSC4), September 1999 Presenter of IEEE Mid-night Sun Workshop on Soft Computing Methods in Industrial Applications, Kuusamo, Finland, June 1999 Member of Technical and Organizing Committees of IEEE Mid-night Sun Workshop on Soft Computing Methods in Industrial Applications, Kuusamo, Finland, June 1999 Member of Student Activities Committee of IEEE Systems, Man, and Cybernetics Society, 1998  1999 Reviewer of IEEE Transactions on Industrial Electronics, 1998  1999 Presenter of IEEE Nordic Workshop on Power and Industrial Electronics, Helsinki, Finland, August 1998 Presenter of IEEE International Joint Conference on Neural Networks, Anchorage, Alaska, May 1998 Reviewer of IEEE Transactions on Instrumentation and Measurement, 1996  1997 Presenter of IEEE Instrumentation and Measurement Technology Conference, Ottawa, Canada, May 1997 Presenter of Heilongjiang Control Theory, Technology, and Conference, Weihai, China, September 1995 Application Presenter of China Intelligent Automation Conference, Tianjin, China, August 1995 PUBLICATION LIST Journal Papers: [1] X M Gao, X Z Gao, J M A Tanskanen, and S J Ovaska, "Power prediction in mobile communication systems using an optimal neural network structure," IEEE Transactions on Neural Networks, vol 8, no 6, November 1997 -3- [2] X Z Gao, C Wang, L Xu, and X Zhuang, "A new combined neural network model," Journal of Systems Simulation, vol 15, no 3, 1997 [3] C Wang, L Xu, X Z Gao, and X Zhuang, "Study on adaptive control in servo systems using recurrent neural networks," Journal of Systems Simulation, vol 15, no 4, 1997 [4] X Z Gao, C Wang, L Xu, and X Zhuang, "Adaptive algorithm for improving resolution in CMAC," Journal of Harbin Institute of Technology, vol 6, no 3, 1996 [5] L Xu, C Wang, X Z Gao, and X Zhuang, "Training recurrent neural networks with stochastic learning algorithm," Journal of Harbin Institute of Technology, vol 6, no 3, 1996 [6] X Z Gao, C Wang, L Xu, and X Zhuang, "A genetic algorithm-based solution to the inverted pendulum control," Journal of Control Theory and Technology, vol 12, no 4, 1995 Conference Papers: 1999 [7] X Z Gao and S J Ovaska, "A new fuzzy filter with application in motion control systems," IEEE International Conference on Systems, Man, and Cybernetics, Tokyo, Japan, October 1999 [8] X Z Gao and S J Ovaska, "Neural networks-based friction compensation with application in servo motor systems," 4th On-line World Conference on Soft Computing in Industrial Applications (WSC4), September 1999 [9] X Z Gao and S J Ovaska, "Friction compensation in servo motor systems using neural networks," IEEE Mid-night Sun Workshop on Soft Computing Methods in Industrial Applications, Kuusamo, Finland, June 1999 [10] L He, K Wang, H Jin, G Li, and X Z Gao, "The combination and prospects of neural networks, fuzzy logic, and genetic algorithms," IEEE Mid-night Sun Workshop on Soft Computing Methods in Industrial Applications, Kuusamo, Finland, June 1999 [11] J Q Zhang, S J Ovaska, and X Z Gao, "A novel MIMO fuzzy model," IEEE Southeastcon’99, Lexington, Kentucky, 1999 [12] J Q Zhang, S J Ovaska, and X Z Gao, "An eigenvalue residuum-based criterion for detection of the number of sinusoids in white Gaussian noise," IEEE Southeastcon’99, Lexington, Kentucky, 1999 -4- 1998 [13] X Z Gao and S J Ovaska, "Comparison of conventional and soft computing-based control methods in a power regulation application," IEEE International Conference on Systems, Man, and Cybernetics, San Diego, California, October 1998 [14] S Väliviita, X Z Gao, and S J Ovaska, "Polynomial predictive filters: complementing technique to fuzzy filtering," IEEE International Conference on Systems, Man, and Cybernetics, San Diego, California, October 1998 [15] X Z Gao, A Dumitrescu, V Burtea, and S J Ovaska, "A new fuzzy adaptive filter with application in DC motor control system," IEEE Nordic Workshop on Power and Industrial Electronics, Helsinki, Finland, August 1998 [16] X Z Gao, "A Temporal Difference method-based prediction scheme applied to fading power signals," IEEE International Joint Conference on Neural Networks, Anchorage, Alaska, May 1998 [17] X Z Gao, S Väliviita, S J Ovaska, and J Q Zhang, "A novel neural networks-based approach to the acquisition of acceleration from noisy velocity signal," IEEE Instrumentation and Measurement Technology Conference, St Paul, Minnesota, May 1998 1997 [18] X Z Gao, X M Gao, and S J Ovaska, "Fast reinforcement learning algorithm for power control in cellular communication systems," IEEE International Conference on Systems, Man, and Cybernetics, Orlando, Florida, October 1997 [19] X Z Gao, X M Gao, and S J Ovaska, "Trajectory control based on a modified Elman neural network," IEEE International Conference on Systems, Man, and Cybernetics, Orlando, Florida, October 1997 [20] X M Gao, X Z Gao, and S J Ovaska, "Power command enhancement in mobile communication systems using an embedded fuzzy unit," IEEE International Conference on Systems, Man, and Cybernetics, Orlando, Florida, October 1997 [21] X Z Gao, X M Gao, and S J Ovaska, "A/D converter resolution enhancement using neural networks," IEEE Instrumentation and Measurement Technology Conference, Ottawa, Canada, May 1997 -5- [22] X M Gao, X Z Gao, and S J Ovaska, "Power prediction using an adaptive neuro-fuzzy predictor," IEEE Instrumentation and Measurement Technology Conference, Ottawa, Canada, May 1997 [23] X M Gao, X Z Gao, J M A Tanskanen, and S J Ovaska, "Intelligent power control for mobile DS/CDMA systems using a modified Elman neural network controller," IEEE 47th Vehicular Technology Conference, Phoenix, Arizona, May 1997 1996 [24] X Z Gao, X M Gao, and S J Ovaska, "A modified Elman neural network model with applications to dynamical systems identification," IEEE International Conference on Systems, Man, and Cybernetics, Beijing, China, October 1996 [25] X Z Gao, C Wang, X M Gao, and S J Ovaska, "A new CMAC neural network model with adaptive quantization input layer," The Third International Conference on Signal Processing, Beijing, China, October 1996 [26] C Wang, L Xu, X Z Gao, and X Zhuang, "The training stability of fully recurrent neural networks," China Youth Automation Conference (CYAC’96), Wuyishan, China, September 1996 1995 [27] X Z Gao, C Wang, L Xu, and X Zhuang, "Control of the inverted pendulum using genetic algorithm," Heilongjiang Control Theory, Technology, and Application Conference, Weihai, China, September 1995 [28] X Z Gao, C Wang, L Xu, and X Zhuang, "CMAC-based reinforcement learning control," China Intelligent Automation Conference (CIAC’95), Tianjin, China, August 1995 THESIS X Z Gao, "Soft computing methods for control and instrumentation," Technical Report (Ph.D Thesis) of Institute of Intelligent Power Electronics, Department of Electrical and Communications Engineering, Helsinki University of Technology, Finland, June 1999 SKILLS Computers: IBM PC, Power Macintosh, HP 9000 workstation, Sun Ultra, Single-chip-based computer Operating Systems: Windows NT, Windows 95/98, Windows 3.1, MSDOS, Macintosh OS, Unix -6- Computer Languages: C/C++, HTML, Visual Basic, Fortran, Assembly languages (Intel 80x86) Software Packages: Matlab & Simulink, Matlab Application Toolboxes (Neural Networks, Control Systems, System Identification, Communications, Fuzzy Logic, Signal Processing, Optimization), Vissim, Mathematica Languages: Chinese: Mother tongue English: Fluent AWARDS Research Scientist position of Graduate School Telecommunications, and Automation (GETA), 1997  1999 of Electronics, Scholarship of Center for International Mobility (CIMO), 1996  1997 One of the top outstanding graduates (5%) in Harbin Institute of Technology, 1996 CAST scholarship of Aerospace Ministry of China, 1995 Scholarship of excellent student in Harbin Institute of Technology, 1989  1993 First prize for English contest in Harbin Institute of Technology, 1992 -7- ... Therefore, the objective of this thesis is to explore and investigate the soft computing theory so that new and enhanced methods can be put forward The applications of soft computing in control and. .. SOFT COMPUTING METHODS FOR CONTROL AND INSTRUMENTATION Xiao-Zhi Gao Dissertation for the degree of Doctor of Science in Technology to be presented with due permission for public examination and. .. solve demanding real-world problems In this work, the existing soft computing techniques have been enhanced, and applied to control and instrumentation areas First, new soft computing methods