����3^rd IEEE International Conference�on�Advanced Information and Communication Technologies (AICT) ‑ 2019�Lviv, Ukraine�July 4 2019���Knowledge Representation in Smart Rules Engine�

Professor, Anatolii Kargin (kargin@kart.edu.ua)

Associate Professor, Tetyana Petrenko (petrenko_tg@kart.edu.ua)

Department of Information Technology

Ukrainian State University of Railway Transport (UkrSURT)

Kharkiv, Ukraine

Data processing� for IoT applications

Features

​

• А large number of data sources
• Data from various types of sensors
• Real-time on-line processing of large-scale heterogeneous data from sensors
• Reactions to events in complicated situations

​

Two approaches

​

• The data from sensors are directly transferred to the cloud for further off-line intellectual analysis
• The data from sensors are on-line processed in real time on the Rules Based systems technology

​

2

Rules Based systems technology in IoT�Rules Engine

3

Rule based systems models

​

• Reasoning based on the search on the trees (Production rules, Propositional Logic, Proof Theory, Resolution)
• Reasoning based on the processing of the rules (Bayesian reasoning, Certainty factors theory and evidential reasoning, Fuzzy logic)

​

4

Rules Engine Frameworks in IoT - Flow diagrams, Complex Event Processing, Apache Spark, Business Process Management

5

Waylay IoT rules engine

The Waylay IoT rules engine belongs to the group of inference engines, and it is based on the Bayesian Networks.

Every node in the graph infers its state to all other nodes, and that each node “feels” what other node is experiencing at any moment in time.

Waylay IoT can model uncertainties, but only in probabilistic way.

6

Rules Engine problems that do not allow to fully take into account the features of the IoT application domain

7

Smart Rules Engine model

8

Abstraction engine based on L-R fuzzy numbers calculations

​

Two stages RE. Engine of 1-st stage is abstraction from data

Multilevel structured fact is computational meaning of data from sensors obtained by abstraction engine

Domain knowledge are represented by semantic structured prototypes

​

Lotfi A. Zadeh conception FROM NUMBERS TO WORDS

Objective of the paper is Smart Rules Engine knowledge representation model

​

• Knowledge representation model serves as a bridge between measurement-based information (data from sensors) and perception-based information (fact).
• According to conception FROM NUMBERS TO WORDS it is required semantic knowledge representation in the prototypes form.
• It is required to takes into account dynamic properties of the domain, incompleteness, uncertainty and relevance associated with the aging of the data from sensors.

​

9

Abstraction from sensory data in nature

10

Abstraction processing model.�Abstraction of input data from sensors� reduces the number of rules

11

​

What is a multilevel fact representing data from sensors at different levels of abstraction� and� how to get it

?

12

What is Multilevel fact ?

​

• Fact is computational meaning of data from sensors.
• Internally, the meaning of a fact is the Fact Fuzzy Characteristic which expresses the degree of agreement of computational meaning with data from sensors.
• Externally, the meaning of a fact is expressed in natural language through its definition.
• The definition of a fact is given through its connections with other facts of the lower levels of abstraction.

​

13

How to get Multilevel fact ?

14

15

Fact Fuzzy Characteristic (FFC)

​

16

Fact Fuzzy Characteristic (FFC)

17

Fact Fuzzy Characteristic (FFC)

​

​

• А set of facts determines a set of possible situations to which SRE can react.
• Inference engine actualizes a certain set of facts for given set of input data.
• Сrisp certainty factor cf is a degree of relevance of the fact.

​

18

Knowledge representation model.�Knowledge portion about concept

<N, know, {<M_i, (a_i, b_i, v_i, g_i)>, M_i 𝜖 Ω_N},

N – concept ID;

know is symbolic description of a knowledge portion about concept;

Ω_N={M_i}_i=1,2,...,k is set of lower levels concepts, which are used in the definition of know concept;

M_i is the lower level concept, the knowledge portion of which is used in determining the concept N;

(a_i, b_i, v_i, g_i) are knowledge portion characteristics of concept M_i.

19

Knowledge representation model.�Knowledge portion about concept (continue)

​

• (−1≤ a_i ≤+1) is expert's confidence that the concept M_i is necessary to determine the concept N;
• 0≤b<∞ is dynamic parameter and used in the definition of time events;
• 0.0≤v≤1.0 is the speed of information aging shows how quickly the relevance of the knowledge portion is lost;
• 0.0 ≤ g_i ≤ 1.0 is The information completeness parameter shows how sufficient one concept M_i (knowledge portions associated with it) is for understanding the meaning of the concept N.

20

Example Knowledge Portion Representation

​

Prototype Structure

​

​

​

​

Formal concept of 1.0 definition

<1.0, THERE IS AN OBJECT AT CLOSE RANGE,

{<0.1, (1.0, 0, 0.0, 1.0)>,

<0.2, (1.0, 0, 0.0, 1.0)>,

<0.3, (0.6, 0, 0.0, 1.0)>}>

​

21

Example Knowledge Portion Representation

know = THERE IS AN OBJECT AT CLOSE RANGE;

Ω_1.0 = {0.1, 0.2, 0.3} is identifiers of concepts representing prototype concepts such as SECTION 1 (some object is on the 1st section), SECTION 2, SECTION 3, respectively. g₁=1.0, reflects the expert's opinion that the concept SECTION 1 is distinguishing to determine concept THERE IS AN OBJECT AT CLOSE RANGE.

g₁=1.0 indicates that the information about the location of an object on the 1st section (concept SECTION 1) is sufficient to describe the situation with the concept THERE IS AN OBJECT AT CLOSE RANGE

The dynamic parameters (b = 0, v = 0.0) indicate that according to the prototype, it is not important to know when the object appeared on the 1st section, but the only significant thing is that it is now on the 1st section.

22

SRE adaptability: the replenishment of new knowledge and the inclusion of new sources of data from sensors

23

SRE adaptability: the replenishment of new knowledge and the inclusion of new sources of data from sensors

24

SRE adaptability: inclusion of new sources of data from sensors

1.<0.1, some object is on the 1st section, {<>}>;

..

7.<0.7, some object is on the 2th section, {<>}>;

12.<1.2, object has moved slowly from section 2 to section 1, {<0.1,(1.0, 0, 0.15, 0.5)>, <0.2, (1.0, 2, 0.15, 0.5)>}>;

…

20.<1.7, object has moved slowly from section 7 to section 6, {<0.6, (1.0, 0, 0.15, 0.5)>, <0.7, (1.0, 2, 0.15, 0.5)>}>;

27. <1.14, there is no object near, {<0.1, (−1.0, 0, 0.0, 0.33)>, <0.2, (−1.0, 0, 0.0, 0.33)>, <0.3, (−1.0, 0, 0.0, 0.33)>}>;

…

33. <5.0, at close range the situation is safe, {<4.0, (1.0,0, 0.0, 0.5)>, <1.10, (1.0, 0, 0.0, 0.5)>}>.

34.<3.2, the object is approaching with a slowdown», {<2.2, (1.0, 1, 0.0, 0.5)>, <2.1, (1.0, 0, 0.0, 0.5)>}>;

35.<4.1, object close, approaching slowly, slowing, {<1.0, (1.0,0,0.0,0.33)>,<2.1, (1.0,0,0.0,0.33)>,<3.2, (1.0,0,0.0,0.33)>}>;

…

38.<5.2, before the object was close and approached slowly and slowed down, now the object is close stopped, {<4.1, (1.0, 1, 0.0, 0.5)>,<3.3, (1.0, 0, 0.0, 0.5)>}>;

39.<6.0, safely,{<5.1, (1.0, t, 0.0, 1.0)>,<5.2, (1.0,t,0.0,1.0)>}>;

40.<7.0, near safe, <5.0, (1.0,0,0.0,1.0)>,<6.0, (1.0,0,0.0,1.0)>}>.

25

Reasoning engine

• Granulation and fuzzification of data
• Comparison is the operation of comparing two fuzzy L-R numbers: prototype and fact fuzzy characteristic
• Aggregation of fuzzy similarity is performed as an addition operation for k weighted fuzzy L-R numbers
• Actualization of facts consists in calculating the value of the certainty factor cf.

26

Research and Training polygon IoT&SM

• Testing of the model was performed on the resources of the Research and Training polygon IoT&SM
• YouTube. Department of IT UkrSURT

Available: https://www.youtube.com/channel/UCxCCUJvQwUBn9WEpaUh_fGg

(accessed Jun. 27, 2019).  

27

Summary

Abstraction approach to processing data from sensors in rules based system overcomes problems:

• dynamic changes in the environment,
• combining data from different sensors,
• uncertainty and incompleteness information

due to the use of the knowledge representation model in the form of parameterized semantic prototypes and multilevel facts.

28

Future

​

• Expand the capabilities of the SRE by implementing the function of automatically obtaining the missing knowledge from external repositories and knowledge bases in Internet.
• This will open up the SRE capability in real-time to access KBs and use this knowledge when processing data from sensors in the IoT.

29

Thanks

​

Questions ?

​

​

30