Скачать презентацию Intelligent decision support system for river floodplain management Скачать презентацию Intelligent decision support system for river floodplain management

1dd4f35b103687adf0fa2ac2bec0d049.ppt

  • Количество слайдов: 28

Intelligent decision support system for river floodplain management Peter Wriggers, Marina Kultsova, Alexander Kapysh, Intelligent decision support system for river floodplain management Peter Wriggers, Marina Kultsova, Alexander Kapysh, Anton Kultsov, Irina Zhukova Leibniz University Hannover, Germany, Volgograd State Technical University, Volgograd, Russia,

Biosphere Reserve Elbe River Landscape in Lower Saxony More then 20 stakeholders: • • Biosphere Reserve Elbe River Landscape in Lower Saxony More then 20 stakeholders: • • • The Elbe Riverland covers the area within the borders of the UNESCO Biosphere Reserve Elbe River Landscape in Lower Saxony reaching from Elbe-km 472. 5 near Schnackenburg to Elbe-km 569 near Lauenburg. Altitude varies from 5 to 109 m NN. The overall area of the biosphere reserve comprises some 56, 760 hectares. • • • Administration of the Elbe river biosphere reserve (ABRE); Environmental Ministry of Lower Saxony; District of Lueneburg (Water Department, Nature Conservation Department); District of Luechow-Dannenberg (Water Department, Nature Conservation Department); Municipalities with areas in the Elbe floodplains; Water Management and Flood Control Authority; Dike Associations; Disaster Control Authority; Water and Soil Associations; Agricultural Ministry, Chamber and Associations of Lower Saxony; Fishing Associations and Companies; Nature Conservation Associations; Ministry of Economy of Lower Saxony; Tourist Associations and Companies; Administrations for Waterways and Navigation; Navigation companies.

Identification of aims, subaims and attributes Global aim: Conservation and development of the Elbe Identification of aims, subaims and attributes Global aim: Conservation and development of the Elbe river floodplains in a state, that the claims of water management, flood protection, agriculture, fishery, nature conservation, tourism and navigation are answered at the best possible rate. Priorities: Aspects: 1 st priority Flood protection in case of extreme flood events if life Flood protection; or properties of people are acutely threatened Water management; Agriculture; 2 nd priority Fishery; Flood Reduction Conserva. Optimi Nature conservation protection of risks for tion and zation of / ecology; in case of farmers developoffers Tourism; regular concerning ment of for Navigation. flood contaminant (semi-) tourists events input natural situation

Table of subaims and attributes (fragment) Table of subaims and attributes (fragment)

Table of subaims and actions (fragment) Table of subaims and actions (fragment)

Domain knowledge features Description of a decision making problem has a complex and variable Domain knowledge features Description of a decision making problem has a complex and variable structure; Components of the problem and solution descriptions can significantly differ in their importance; There are multiple logical relations between components of problem description; Both qualitative and quantitative parameters, which should be interpreted in a context-dependent way (not by absolute value), are taken into account; Knowledge about previously solved problems which are similar to the current one is actively used while decision making.

Concept of CBR-based decision making (I) Appropriate AI technologies for the intelligent decision making Concept of CBR-based decision making (I) Appropriate AI technologies for the intelligent decision making in floodplain management: ontological knowledge representation model, case based reasoning and qualitative reasoning. While CBR is a leading reasoning mechanism, QR and ontology carry out the knowledge-based support of the main stages of CBR process is strongly supported with ontology and qualitative reasoning at all stages, such "knowledgeintensive"approach allows to improve the accuracy and correctness of the obtained CBR solutions.

Concept of CBR-based decision making (II) Ontology de fines case representation model and sets Concept of CBR-based decision making (II) Ontology de fines case representation model and sets of parameter description with their possible values. Ontology contains information about the aims of stakeholders, aim attributes and their possible values. Onto foundation for case description, and it can be used as a knowledge base for subsystem of automated case formulation; If the user has de ned case index incompletely, ontology can be ap the stage of case retrieval for rede nition of the case. Also the algor similarity measure calculation depends heavily on the case representation model, which is defi ned by ontology; Ontology is used at the stage of case adaptation to make decision about alternative values of parameters if adaptation algorithm has proposed a few equivalent choices, for example, a few possible values for one attribute; Ontology contains qualitative model of case and consistency rules which can be used at the stage of case revision. Consistency rules de ne infe combinations of parameter values for di erent aspects of case description.

General schema of integration of CBR, QR and ontology Query generation Ontology Case representation General schema of integration of CBR, QR and ontology Query generation Ontology Case representation model 1 Domain knowledge 2 Case base 3 4 Case qualitative model Case retrieval Case adaptation Case revision 5 Case retention 1 - case structure, 2 - formalized domain knowledge, 3 - stored case, 4 - new case, 5 - qualitative simulation results.

Modi ed CBR-cycle supported with and ontology 1 - problem description, 2 - CBR-query, Modi ed CBR-cycle supported with and ontology 1 - problem description, 2 - CBR-query, 3 - case index, 4 — retrieved cases, 5 - adapted solution, 6 - revised solution, 7 - new case, 8 - case structure, 9 - stored cases, 10 - expert rules, 11 adaptation rules, 12 - consistency rules, 13 - qualitative simulation results, 14 - domain conceptualization, 15 - case descriptions, 16 - general domain knowledge, 17 - qualitative dependencies between components of case description.

Ontological knowledge representation model ONT =< DO; CS; QM >; (1) where DO { Ontological knowledge representation model ONT =< DO; CS; QM >; (1) where DO { domain ontology, CS { case structure, QM { case qualitative model. DO = f. OB; ATT; DAT; ACT; CR; AR; QD; RELg; (2) where OB = f. OB 1; OB 2; : : : ; OBng { set of stakeholders' objectives; ATT = f. ATT 1 [ ATT 2 [ : : : [ ATTng { set of attributes which are used for problem description, each subset ATTi corresponds to objective OBi; DAT { set of ontology domains, which de ne the space of attribute values; CR { set of consistency rules; AR { set of adaptation rules; ACT = f. ACT 1 [ ACT 2 [ : : : [ ACTng { set of actions, each subset ACTi corresponds to objective OBi; QD { set of qualitative dependencies between attribute values, QD : ATT ! f. I+; I P where I+ { positive direct inuences, I ; P+; ; ? g, { negative direct inuences, P+ { positive indirect inuences, P { negative indirect inuences; REL = f. REL 1; REL 2; : : : ; RELkg { set of relations between ontology concepts.

Case representation model CS = f. P; S(P); CRELg; (3) where P { problem Case representation model CS = f. P; S(P); CRELg; (3) where P { problem description (also can be referred as a case index); S(P) { problem solution description as a set of actions which were applied; CREL = f. CREL 1; CREL 2; : : : ; CREL 8 g { set of case relations. Set CREL contains the following relations: CREL 1 { relation has. Initial. State, CREL 2 { relation has. Action. Set, CREL 3 { relation has. Result. State, CREL 4 { relation has. Causal. State, CREL 5 { relation has. Effective. State, CREL 6 { relation has. Assessment. Set, CREL 7 { relation has. Certain. Action, CREL 8 { relation has. Certain. Assessment.

Case qualitative model QM = f. BBC; ACg; (8) where BBC { set of Case qualitative model QM = f. BBC; ACg; (8) where BBC { set of QM model primitives; AC { set of compound components composed of the primitives. BBC = f. SB; BB; ABg; (9) where SB { set of structural primitives; BB { set of primitives describing behaviour of the model; AB { set of premises de ning the applicability of compound components of model. SB = f. SE; SCg; (10) where SE [ ATT [ ACT) { entity hierarchy reecting the concept hierarchy (OB of the problem domain; SC { set of relations between entities. BB = f. QU; QS; Dg; where QU { set of quantities; QS ATT DAT { set of quantity spaces; D { set of causal relations between quantities, D has the following range QD of values f. I+; I P ; P+; g. AC = f. SF; PF; SCg; (12) where SF { set of static model fragments; PF { set of process model fragments; SC { set of simulation scenarios.

Fragment of ontological representation of case structure Fragment of ontological representation of case structure

Fragment of ontological representation of case Fragment of ontological representation of case

Reasoning mechanism The following algorithms were developed to operate on the suggested ontological knowledge Reasoning mechanism The following algorithms were developed to operate on the suggested ontological knowledge representation model in the framework of the CBR mechanism implementation: CBR-query formulation support algorithm which allows reducing the amount of the routine work needed to input information and enforcing knowledge representation model integrity as well as rede ning the c with use of general domain knowledge in form of DL rules. ; Case retrieval algorithm which uses class (concept)-based similarity (CBS) computation algorithm and property (slot)-based similarity (SBS) computation algorithm; Case adaptation algorithm which uses domain knowledge in form of DL rules; Case revising algorithm which uses the results of qualitative simulation on the case qualitative model.

ПОИСК ПРЕЦЕДЕНТОВ Мера близости Общая мера близости: Мера близости дескрипторов интервального типа: Мера близости ПОИСК ПРЕЦЕДЕНТОВ Мера близости Общая мера близости: Мера близости дескрипторов интервального типа: Мера близости дескрипторов типа «дата» : Мера близости дескрипторов вещественного типа: Мера близости дескрипторов типа «перечисление» : 17

МЕРА БЛИЗОСТИ прецедентов CBS, SBS Для оценки меры близости дескрипторов, представляющих собой качественную переменную МЕРА БЛИЗОСТИ прецедентов CBS, SBS Для оценки меры близости дескрипторов, представляющих собой качественную переменную или таксономию применяется алгоритм оценки близости экземпляров онтологий на основе анализа иерархии классов онтологии: Алгоритм SBS (slot based similarity) предназначен для поиска меры близости между двумя прецедентами с учетом объектных отношений между компонентами описания прецедента в виде отношений онтологии и представляет собой рекурсивную функцию обхода графа, в котором узлами являются концепты онтологии, а дугами – объектные отношения онтологии: Предлагаемый алгоритм SBS отличается от известного тем, что позволяет учесть разницу в описании между прецедентом из базы прецедентов и запросом. 18

ПОКАЗАТЕЛЬ ЭФФЕКТИВНОСТИ для алгоритма поиска прецедентов Для каждого прецедента в базе прецедентов доступно описание ПОКАЗАТЕЛЬ ЭФФЕКТИВНОСТИ для алгоритма поиска прецедентов Для каждого прецедента в базе прецедентов доступно описание конечной ситуации, которая сложилась после применения указанных в прецеденте мер воздействия к начальной ситуации. Разницу в оценках начальной и конечной ситуаций прецедента предлагается использовать как показатель эффективности E, который позволяет оценить степень полезности решения рассматриваемого прецедента к РПП-запросу Q где: E – показатель эффективности, A – оценка начальной ситуации по i-му аспекту, – оценка конечной ситуации по i-му аспекту, diff – разница между оценками, W – весовой коэффициент i-го аспекта, N – количество аспектов. 19

Case retrieval algorithm 1. Defi ne preferences of decision PREFi and weighting factors WAi Case retrieval algorithm 1. Defi ne preferences of decision PREFi and weighting factors WAi ; maker 2. Compute Sim and E for each case Cj from case base CB; 3. Form empty result vector SC: SCj = (Cj ; vj ), where vj - con dence factor for Cj ; 4. Form set of cases Cp for which 8 Cj 2 CP : Sim (Cj ; Q) > Sim. MIN E (Cj ; Q) > 0, where Sim. MIN = min Cj 2 CB Sim (Cj ; Q); 5. Add to list SC not more then M pairs (Cj ; vj ), for which Cj : Cj 2 CP E (Cj) = EMAX, state vj = E (Cj ); 6. Add to list SC not more then M pairs (Cj ; vj ), for which Cj : Cj 2 CP Sim (Cj ; Q) = Sim. MAX, state vj = SIM (Cj ; Q). Vector SC is the output of retrieve algorithm.

Case adaptation algorithm 1. Form empty list Ac. SC of pairs f. Ac; vag Case adaptation algorithm 1. Form empty list Ac. SC of pairs f. Ac; vag where Ac - action, va - con dence factor; 2. For each pair f. Cj ; vjg from SC do: (a) For each action Aci from Cj add pair f. Aci; vjg to Ac. SC list (i. e. add actions from retrieved cases with con dence v); factor 3. Perform reasoning on ontology; 4. For each action Ac. R inferred in accordance with adaptation rules, add pair f. Ac. R; 1 g to Ac. SC list; 5. If list Ac. SC is empty, when case base and ontology is not applicable to current DSS-query, adaptation is unsuccessful; otherwise 6. Sort list Ac. SC by descendant of con dence factor va; 7. Remove incongruous actions from Ac. SC list: (a) Set i = 0; (b) For each j. Ac. SCj > i remove pair f. Acj ; vag from Ac. SC if MIij = 1; (c) Set i = i + 1; (d) If j. Ac. SCj > i then go to step (b) 8. Form case solution list Ac. S that consists of actions Aci from list Ac. SC. The case solution list Ac. S is the output of adaptation algorithm.

Case revising algorithm 1. Generate new case C as an individual of Case revising algorithm 1. Generate new case C as an individual of "Case" class which has index Ind coincident with CBR-query Q and solution Acs; 2. Perform the procedure "classify taxonomy" on ontology; 3. If case C was classi ed as individual of "inconsistency" class (in accordance with consistency rules), then revision is unsuccessful, remove case C; 4. Simulate the qualitative model for case C and de ne assessment A i for obtained result state; 5. Compute E(C) using (23); 6. If E(C) < 0, then revision is unsuccessful, remove case C; 7. If 0 6 E (C) < E , where E average E-value for stored cases, then is revision is quasi-successful, use solution Acs for current problem, but don't retain case C in case base; 8. If E (C) > E , then revision is successful, retain case C in case base.

j. Ramwass system architecture 1 - case structure, 2 - stored cases, 3 - j. Ramwass system architecture 1 - case structure, 2 - stored cases, 3 - domain knowledge in form of DL-rules, 4 - asserted and inferred knowledge, 5 - case qualitative model in OWL-format, 6 - case qualitative model in GARP 3 format, 7 - rede nition rules, 8 - adaptation rules, 9 - consistency rules, 10 qualitative simulation results, 11 - problem description, 12 - CBR-query (case index), 13 - values of similarity measures, 14 - set of similar cases, 15 - adapted solution, 16 - new case.

Decision making process using IDSS j. Ramwass 1. Formalization of problem description { identifying Decision making process using IDSS j. Ramwass 1. Formalization of problem description { identifying the aim hierarchy, aim attributes and values, priorities of decision maker. 2. Formulation of CBR-query in framework of case representation model. 3. Rede nition of CBR-query using inference on expert rules and generation of case index (description of problem situation). 4. Retrieve of the similar cases in case base using retrieve algorithm based on CBS and SBS similarity metrics. 5. Selection of the relevant cases for reuse. 6. Adaptation of the selected cases' solutions to the new problem using adaptation algorithm based on inference on adaptation rules. 7. Revision of the new adapted solution using simulation on case qualitative model and inference on consistency rules. 8. Generation of a new case for the solved problem in framework of case representation model. 9. Retention of the new case in the case base for further use.

Пример решения задачи управления с помощью системы. Описание запроса к системе. Создание рекреационной зоны Пример решения задачи управления с помощью системы. Описание запроса к системе. Создание рекреационной зоны • Предпочтения хозяйствующих субъектов: – – – • Расположение территории: – – • 559 -й километр реки Эльба Левый берег Параметры ситуации: – – – – – • Туристические компании: высокий приоритет Экологи: средний приоритет Остальные: низкий приоритет Площадь участка: 67 км. кв. Строения: нет Способ землепользования: лиг, низкотравье Обзорность участка с дамбы: хорошая Доступность воды: плохая Количество обзорных точек: хорошо Высота дамб: хорошо Состояние дамб: хорошо Колебание уровня воды: хорошо Оценка начальной ситуации по аспектам: – – – – Сельское хозяйство: хорошо Рыболовство: удовлетворительно Защита от наводнений: удовлетворительно Судоходство: удовлетворительно Затраты на обслуживание: удовлетворительно Сохранение природы: хорошо Туризм: удовлетворительно 25

Пример решения задачи управления с помощью системы. Получение решения. Результаты поиска Прецедентов Наиболее близкие Пример решения задачи управления с помощью системы. Получение решения. Результаты поиска Прецедентов Наиболее близкие прецеденты: • № 16 – преобразование участка в рекреационную зону • № 21 – сохранение луга • № 13 – перемещение дамб Трассировка процесса адаптации Результаты адаптации 26

ТЕСТИРОВАНИЕ СИСТЕМЫ Тестовая база составляет 42 прецедента; Совпадение результатов стандартного алгоритма с экспертным решением ТЕСТИРОВАНИЕ СИСТЕМЫ Тестовая база составляет 42 прецедента; Совпадение результатов стандартного алгоритма с экспертным решением составляет в среднем 67%; Совпадение результатов предлагаемого алгоритма с экспертным решением составляет в среднем 88% , где NSOL – количество элементов решения прецедента С, N*SOL – количество элементов совпадающих элементов решения С и Сi 27

Thank you Thank you