Design of an Automatic Control Algorithm for Energy-Efficient ...

More documents

Recommendations

Info

6 The optimiser: an evolutionary algorithm approach 56 • Select the control values with respect to the last one applied to avoid jumps. • Prevent control outputs which exceed limits by replacing them with panic handlers, i.e. special outputs. A flowchart of the algorithm is found in Figure 6.3. The methods applied are described in detail in the next sections. �� Figure 6.3: Flowchart of the incremental multi-objective evolutionary algorithm. At the start of each time-step the system states (temperatures, humidity and other sensor values) are updated. They are then used in the system prediction. The population is built up by the elite from the last run and a constant base population. The remaining individuals for the population are randomly chosen from the design space. Using the system model, the future states are predicted for each individual. Based on these the objectives are evaluated and the results stored. In the following step the population is compared using the limits and priorities ranking scheme. The ranking is decisive for the selection as parent. The parent couples are recom- bined and the resulting offspring mutated. The objectives are evaluated for the children and a ranking together with the parent generation is conducted. Finally this ranked list ��
6 The optimiser: an evolutionary algorithm approach 57 is truncated to the original population size. To obtain a better result, more generations can be realised with the algorithm. If this is desired, parents are chosen again from the population and a new offspring is created. After the set number of generations the most suitable individual is selected as control output. This output as well as the best individuals that are not too close to the base population are selected as elite and stored for the next run. 6.3.2 Design variable coding The algorithm is designed to work with continuous design variables. They are “coded” as real numbers, which is sufficient for the application as it is the most comprehensive and the simplest way. Discrete design variables, however, are possible with some extensions to the algorithm as it is done for GAME by H. Langer [34]. 6.3.3 Generation of the initial population The creation of the initial population has to fulfil two objectives: On the one hand it should be as close as possible to the optimum in order to find the best possible solution. On the other the non-stationary environment requires it to be widespread to be able to cope with changes. The presumptive closest individuals to the optimum that are known, are the fittest ones from the last run. Those have been stored (see also 6.3.6) and can be used now. If the boundaries changed these elite individuals have to be adapted. This is done by simply setting these values onto the limit that are out of bounds. In order to be able to track environment changes two mechanisms are used. Rea- sonable starting values for the most common cases are implemented. If these are spread over the design space and adapted to changing boundaries, they can assure good cover- age and give faster reaction of the algorithm. The implementation of fixed individuals called base population was already done by Lennon and Passino for a brake system control application [32]. The population is completed with random individuals, called random immigrants by Grefenstette [35]. These give another means of diversity and make it possible to find
Page 1:
ISSN 0280-5316 ISRN LUTFD2/TFRT--58
Page 5:
Diploma thesis subject Problem stat
Page 8 and 9:
Acknowledgements ii Acknowledgement
Page 10 and 11:
Acknowledgements iv 3.1.4 Windscree
Page 12 and 13:
Acknowledgements vi 8.5 System simu
Page 14 and 15:
Nomenclature viii Nomenclature Symb
Page 16 and 17:
Nomenclature x Other symbols Symbol
Page 18 and 19:
Nomenclature xii Subscript Descript
Page 21 and 22:
1 Introduction Energy consumption t
Page 23 and 24:
1 Introduction 3 adjustment of the
Page 25 and 26: 2 Control objectives 5 2.1 Windscre
Page 27 and 28: 2 Control objectives 7 2.2 Comfort
Page 29 and 30: 2 Control objectives 9 � Metaboli
Page 31 and 32: 2 Control objectives 11 ��
Page 33 and 34: 2 Control objectives 13 Figure 2.6:
Page 35 and 36: 2 Control objectives 15 2.4 Influen
Page 37 and 38: 3 System description 17 or heat com
Page 39 and 40: 3 System description 19 calculated
Page 41 and 42: 3 System description 21 ��
Page 43 and 44: 3 System description 23 � � �
Page 45 and 46: 3 System description 25 defrost tem
Page 47 and 48: 3 System description 27 Figure 3.4:
Page 49 and 50: 3 System description 29 width modul
Page 51 and 52: 3 System description 31 ��
Page 53 and 54: 3 System description 33 shown in Fi
Page 55 and 56: 3 System description 35 of performa
Page 57 and 58: 4 Control strategy The main goal of
Page 59 and 60: 4 Control strategy 39 limits are di
Page 61 and 62: 4 Control strategy 41 ��
Page 63 and 64: 4 Control strategy 43 cabin. This s
Page 65 and 66: 4 Control strategy 45 controlled. T
Page 67 and 68: 5 The disturbance estimator 47 in e
Page 69 and 70: 6 The optimiser: an evolutionary al
Page 75: 6 The optimiser: an evolutionary al
Page 89 and 90: 7 The HVAC-system controller 69 �
Page 91 and 92: 7 The HVAC-system controller 71 7.5
Page 93 and 94: 8 System and functionality integrat
Page 115 and 116: 9 The MUTE project 95 order to fit
Page 117 and 118: 10 Hardware and characteristics of
Page 127 and 128:
10 Hardware and characteristics of
Page 129 and 130:
11 Matlab implementation The climat
Page 131 and 132:
11 Matlab implementation 111 Only r
Page 133 and 134:
11 Matlab implementation 113 left i
Page 135 and 136:
11 Matlab implementation 115 tion a
Page 137 and 138:
12 Controller evaluation 117 ��
Page 139 and 140:
12 Controller evaluation 119 be not
Page 141 and 142:
12 Controller evaluation 121 Number
Page 143 and 144:
12 Controller evaluation 123 Table
Page 145 and 146:
12 Controller evaluation 125 In the
Page 147 and 148:
12 Controller evaluation 127 ��
Page 149 and 150:
13 Conclusion 129 13.2 Possible imp
Page 151 and 152:
Bibliography [1] Holger Großmann.
Page 153 and 154:
Bibliography 133 [22] M. Wang, T.M.
Page 155 and 156:
Appendices
Page 157 and 158:
B In- and outputs The interface bet
Page 159 and 160:
Signal Unit Range Type Comment Stat
Page 161 and 162:
In- and outputs 141 B.3 Error codes
Page 163 and 164:
and system parameters 2 concentrati
Page 165 and 166:
Simulation parameters 145 C.2 Contr
Page 167 and 168:
Simulation parameters 147 Panic han
Page 169 and 170:
Simulation parameters 149 C.5 Test
Page 171 and 172:
Additional simulation results 151 D
Page 173 and 174:
Optimiser code 153 real_T ∗panic_
Page 175 and 176:
Optimiser code 155 52 return 10; 54
Page 177 and 178:
Optimiser code 157 138 /∗ return
Page 179 and 180:
Optimiser code 159 230 ∗/ ∗ ===
Page 181 and 182:
Optimiser code 161 ∗ IMEA_breedin
Page 183 and 184:
Optimiser code 163 414 } 416 } } st
Page 185 and 186:
Optimiser code 165 if (sortedPop [
Page 187 and 188:
Optimiser code 167 596 ∗ Printout
Page 189:
Optimiser code 169 690 ∗ Latest R
show all

Design of an Automatic Control Algorithm for Energy-Efficient ...

Create successful ePaper yourself

Delete template?

Save as template?