In this paper, we consider the application of the PESoT technology and a tool (namely, Personal Knowledge Base Designer) for prototyping rule-based knowledge bases by using the automated analysis and transformation of decision tables presented in the CSV format. Created prototypes of knowledge bases designed for intelligent decision-making support when analyzing and forecasting the risk (probability) of forest fires based on information about the forest fire hazard class, weather conditions, and other factors. A description of the main stages of the approach and an illustrative example are presented.
In this paper, we consider the application of the PESoT technology for the automated creation of knowledge bases including formalization and code generation tasks in the context of the second technique. So, the main contribution of the paper is a description of the application of this technology when solving the task of decision-making support for forest fire risk forecasting.
The paper is organized as follows. Section 2 presents a background, including the main principles of the PESoT technology, our motivation, and the brief state of the art in the field of forest fire forecasting. Section 3 contains the application including a detailed description of separate steps, and an illustrative example, while Section 4 presents some concluding remarks and future works.
Next, let’s consider the main principles of the used PESoT technology, as well as the subject domain (forecasting the risk of forest fires). 2.1.
The PESoT technology [4-7] implements the model-driven EUD principle and includes methods and tools for prototyping rule-based expert systems and decision-making software components for intelligent systems.
Formally, this technology can be described by the following [4]:
MDE ES = MOF , LES ,CIM ES , PIM ES , PSM ES , PDM ES , FCEIMS −to−PIM , FPEIMS −to−PSM , FPESSM −to−CODE where LES is a set of languages and formalisms used for modeling expert systems and components; in our case LES = {UML,CM , DT ,CT , RVML} where UML is a Unified Modelling Language; CM is a concept or mind maps formalism; DT is a formalism for the representation of decision tables; CT is a formalism for the representation of canonicalized tables; RVML is a Rule Visual Modeling Language [5];
CIMES is a computation-independent model for PESoT, in our case, it is a domain model represented with the aid of LES;
PIMES is a platform-independent model for PESoT, in our case this model represent logical rules in our notation RVML;
PSMES is a platform-specific model for PESoT, in our case this model takes into account the features of the programming language, we use RVML;
PDMES is a set of platform description models for PESoT, in our case PDM ES = {CLIPS, DROOLS, PHP, PKBD} .
FCEIMS −to−PIM , FPEIMS −to−PSM , FPESSM −to−CODE are the rules for model transformations.
The process of creating prototypes of knowledge bases and expert systems is represented by the sequence (a chain) of the following stages: building domain models, building platform-independent models, building platform-specific models, generating source codes and specifications, testing, and integration.
A detailed description of the technology is given in [4-6]. Below, a detailed description of the main stages of its application is considered. 2.2.
The motivation of this work is due to the grant No. 075-15-2020-787 of the Ministry of Science and Higher Education of the Russian Federation "Fundamentals, methods and technologies for digital monitoring and forecasting of the environmental situation on the Baikal natural territory" [8]. This project includes solving a set of tasks, one of which is the development of thematic WPS services for digital monitoring, analysis, modeling, and forecasting of the environmental situation, as well as the risk of natural and technogenic fires.
Researches are being conducted to solve this problem that is linked to: • Collecting, cleaning, and analyzing forest fire data and identifying specific heuristics. These heuristics allow one to identify the territorial risk of forest fire hazard that will be considered as fire risk for forest quarter (aka dacha or forestry); • Developing a knowledge base for analyzing and forecasting the risk (hazard) of fires based on information about the forest fire hazard class, weather conditions, and other factors. 2.3.
The topic of the automated formation of knowledge bases for intelligence decision-making support in forecasting the risk (danger) of fires in forest quarters is poorly represented in the scientific literature. There are some works related to a model transformation and the application of a modelbased approach (e.g., MDD), as well as some papers aimed at forecasting the risk (hazard) of forest fires.
The first group of works is considered in more detail in [4-6].
Several following works that illustrate the main areas of research that can be selected for the second group: • The identification of factors affecting the fire hazard of forests [9-10], including a height, a slope, a topographic humidity index, a distance from urban areas, an average annual temperature, land use, a distance from roads, average annual precipitation, a distance to rivers, air temperature (average daily and maximum), dates of the transition of average daily temperatures through threshold limits, dates of onset and descent of a stable snow cover, a relative humidity (average daily and minimum), an air humidity deficit, a count of days with relative humidity < 30% in one of the observation periods for a certain period, an annual precipitation regime, a count of days with rain, a dryness index, a wind regime, a count of days with thunderstorms, etc.; • The improvement of the scale for assessing forest fire hazard classes depending on weather conditions [11] to account new factors, in particular, humidity indicators [12], or taking into account regional features [13]; • The use of existing methods for assessing fire risk in different regions [14].
In the context of this study, the above works were used to analyze the subject domain and identify factors affecting the assessment of forest fire risk.
The PESoT technology provides the automated formation of knowledge bases by using various information sources, including conceptual models and tables of different types. One of the tabular forms supported by this technology is a specialized form of decision tables described below. 3.1.
The specialized form of decision tables is an extension of the standard one [15] and consists of columns and rows. Columns represent names of independent and dependent properties (components or parts of rules), and rows represent specific rules. At the same time, table cells contain values of their properties. The tabular form used in this work has some features. In particular, these features were determined by further automated processing of tables in the context of knowledge engineering.
The main features of our tabular form: • A table may contain a column with rule names; it must be the first and have the "Rule Name" name; • Headers of dependent columns are marked with the "#" symbol; • A column header name can be compound, indicating an entity name (or a class name) and its property name separated by "::" string. • There are no restrictions on values in cells, i.e. they can not only consist of a set of values {yes, no} as, for example, in [16]. So, values in cells can contain specific arbitrary values and not only values that indicate the presence or absence of certain property (component) in a rule structure.
A specialized decision table fragment is presented in Figure 1.
This form of decision table provides the generation of logical rules of the "IF-THEN" type. In particular, an example of a decision table fragment presented in Figure 1 is interpreted as follows: IF there is a "Risk" of a certain "grade" and "kind", and a «Flood hazard» of a certain "level" and "probability", THEN some "Conclusion" with a certain "text" and "cf" (a certainty factor) is made. Accordingly, values for properties "grade", "kind" etc. are taken from the cells of a certain row.
The advantages of this form of representation of source data are the following: • It is the most popular way to represent logical rules for non-programming end-users; • It is used to represent the results of data mining, for example, when using the Deductor Studio or Loginom system (Base Group company); • It provides the ability to use publicly available and widespread software, such as Microsoft Excel for generating data, and then saving them in the CSV format; • There are examples of using this form when solving various practical tasks [17-19]. 3.2.
The development process with the aid of PESoT and the PKBD (Personal Knowledge Base Designer) tool is similar to other PESoT cases [17] and can be presented in the form of the following scheme (Figure 2).
In the current case, stage 2 has the greatest computational complexity. This stage is associated with data analysis and the formation of decision tables. Let’s consider the stages of this approach using an illustrative example in more detail. 3.3.
The task of prototyping knowledge bases for determining the risk of forest fires is considered an illustrative example. Information on forest fires in the Baikal natural territory for 2017-2020 years, weather data, as well as information on infrastructure (roads, settlements, etc.) and the type of vegetation were used as initial data. The database on fires includes more than 45 000 records describing information about heat points identified as a result of the analysis of satellite images.
Important and computationally complex tasks associated with the preparation of this data for the development of knowledge bases are the following: • Grouping (aggregating) information about fires with the definition of duration, minimum and maximum area of a certain fire; • Determining fires located within the boundaries of industrial zones, settlements, and mining zones that are not natural fires; • Determining fire statistics for certain classes of fire hazard based on forest plans of districts of the Irkutsk region (taking into account the structure of the forestry, aka plots (dachas) or quarters); • Determining a set of independent factors influencing the risk of fire hazard in a forest district; • Calculating factor values affecting the risk of the fire hazard of a forest district and their transformation to interval or fuzzy form; • Determining the risk (hazard) of a forest fire based on the current values of a complex of factors through a certain class of fire hazard and its statistics.
These tasks will be considered in more detail in other works. In this paper, the process of building knowledge bases based on the analysis of decision tables is considered, and we assume that the data preprocessing has already been completed.
In this case, the knowledge base consists of two segments and solving the following subtasks: 1. Forming a conclusion on the forest fire risk of a certain forest area based on the average monthly weather data, current weather conditions, information about the time of year, the proximity of rivers, lakes, roads, settlements, terrain, and vegetation type; 2. Forming a conclusion on the risk of a forest fire according to the fire hazard class of a certain forest area using fire statistics and information about the time of year (season).
Next, we will consider the stages in more detail.
Stage 1. As a model of the domain, a conceptual model was created that describes the factors affecting the class of the fire hazard of a forest area and the risk (probability) of a fire. A fragment of this model is shown in Figure 3.
Stage 2. Next, decision tables were developed that describe the structural aspect of the domain. These tables contain information about combinations of features describing the fire hazard class of a forest area and the risk (hazard) of a forest fire.
In particular, the following table structure (headers) is used to define the hazard class: Road::distance_to_car_road, Road::distance_to_railway, River::distance_to_river, Lake::distance_to_lake, Meteodata::rrr, Meteodata::ff, Meteodata::u, Meteodata::t, Settlement::distance_to_settlement, Settlement::population, Region::population, Region::average_annual_temperature, Season::name, Forestry::staff_number, Square::landform, Square::forest_type, Square::underlying_surface_type, #Square::name, #Square::fire_hazard_class.
To determine the risk (hazard) of a forest fire, the following table structure (headers) is used: Square::name, Square::fire_hazard_class, Season::name, #Fire::risk[probability].
Further, the obtained rules are analyzed to define the indicators of the frequency of their appearance in the analyzed data. In particular, the support and confidence of the rules are determined. The confidence is used as a certainty factor of a rule.
The decision table with intermediate data is shown in Figure 4.
Stage 3. Next, with the aid of PKBD [6] (it is a tool of PESoT), the decision tables were imported and presented in the form of logical rules. The imported decision tables were refined in the RVML (Rule Visual Modeling Language) form (Figure 5).
Stage 4. For two segments of the knowledge base, a code was generated on CLIPS, which was used to debug the obtained knowledge bases, later presented in the form of PHP codes (Figure 6).
Stage 5. Testing and integration will be done in the future.
In this paper, we consider the use of the PESoT technology and tools for prototyping rule-based knowledge bases by using automated analysis and transformation of decision tables. The formed knowledge bases can be used to create an intelligent decision-making support software module in the form of a WPS service for analyzing and forecasting the risk (hazard) of forest fires based on information about the forest fire hazard class, weather conditions, and other factors. An illustrative example demonstrating the fundamental applicability of this approach is presented.
The technology is designed for end-users and reduces the time for creating prototypes of AI modules and expert systems by automating the codification stage and using existing domain models.
Our approach has a certain level of universality and after its improvement can be used in various domains, for example, for solving tasks in the field of industrial safety inspection [20].
In the future, we plan to make a quantitative evaluation of the proposed technology by conducting computational experiments.
The present study was supported by the Ministry of Education and Science of the Russian Federation (Project no. 121030500071-2 "Methods and technologies of a cloud-based service-oriented platform for collecting, storing, and processing large volumes of multi-format interdisciplinary data and knowledge based upon the use of artificial intelligence, model-driven approach, and machine learning"). Results are achieved using the Centre of collective usage «Integrated information network of Irkutsk scientific educational complex».
[4] A.Yu. Yurin, Technology for Prototyping Expert Systems Based on Transformations (PESoT):
A Method, CEUR Workshop Proceedings 2677 (2020) 36-50.
[5] A.Yu. Yurin, N.O. Dorodnykh, O.A. Nikolaychuk, M.A. Grishenko, Prototyping Rule-Based
Expert Systems with the Aid of Model Transformations, Journal of Computer Science 14 (5) (2018)
680-698. doi:10.3844/jcssp.2018.680.698.
[6] A.Yu. Yurin, N.O. Dorodnykh, Personal knowledge base designer: Software for expert systems
prototyping, SoftwareX 11 (2020) 100411. doi: 10.1016/j.softx.2020.100411.
[7] I.V. Bychkov, A.Yu. Yurin, A method and tools for prototyping components of intelligent
systems based on transformations, Journal of Physics: Conference Series. 13th Multiconference on
Control Problems (MCCP 2020) 6-8 October 2020, Saint Petersburg, Russia. 1864 (2021) 012042.
doi:10.1088/1742-6596/1864/1/012042.
[8] I.V. Bychkov, G.M. Ruzhnikov, R.K. Fedorov, A.E. Khmelnov, A.K. Popova, Organization of
digital monitoring of the Baikal natural territory, IOP Conference Series: Earth and Environmental
Science 629(