In this chapter, the authors propose an approach to using a heterogeneous team of unmanned aerial vehicles and remote sensing techniques to perform tactical forest firefighting operations. The authors present the three-level architecture of the multi-UAV-based forest firefighting monitoring system; features of patrolling, confirming, and monitoring missions; as well as functions of UAV in such missions. The authors consider an infrastructure for the UAV ground support and equipment used for the UAVs control. The method of the data integration into a fire-spreading model in a real-time DSS for the forest fire response is proposed. The proposed approach has been tested with the multi-UAV team that included three drones for the patrol missions, one helicopter for the confirmation mission, and one octocopter for the monitoring mission. The performance of such multi-UAV team has been studied in the laboratory conditions. The result of the experiment has shown that the proposed approach provides required credibility and efficiency of fire prediction and response.
TopBackground
The number and impact of disasters seem to be increasing in the last few decades as a result of the growth of population, urbanization, environmental degradation, global warming etc. Forest fires are natural disasters that cause deaths, injuries, and huge damages. A number and intensity of forest fires are significantly increasing year by year due to a growth of human activity and climate changes. Thus, they become bigger, more dangerous, and more costly to put out.
Forest fire response operations are becoming increasingly important but challenging because they are traditionally based on visual observations and decision-makers estimations. Due to exposure to smoke, flame, and high temperature, visual observation is usually ambiguous, imprecise, incomplete, and inconsistent, while the instrumental measurement is expensive, difficult, and dangerous. Besides that, decision-makers often operate under the high responsibility conditions in a lack of time. This stipulates the development of real-time decision-support systems (DSS) for the forest fire response.
However, developing such DSS is a complex and non-trivial task because the forest fire is a process, which arises unexpectedly, proceeds fleetingly, and evolves in space and time transiently, non-linearly. Consequently, the forest fire is poorly modeled and unpredictable process. This completely excludes the use of well-studied classical decision support approaches based on models, rules, etc. In such conditions, the efficiency of forest fire operations strongly depends on the availability of online monitoring and measurement tools. To build such tools, we can use a suite of the most modern methods and techniques, such as remote sensing, geolocation, geographical information systems, unmanned vehicles, etc. Their efficiency primarily depends on the synergy of their joint use.
An indispensable part of such DSSs is online monitoring aimed at the computation of the most critical parameters related to the forest fire spreading. A full range of data for the forest fire monitoring is generated by remote sensing techniques and have a form of data streams of great volumes that come from sensors on a continuous basis at a high rate and should be analyzed in a real-time.
The important task for decision-making is to diagnose the situation based on the online monitoring data, namely, to assign a given data to some category and then respond appropriately based on such classification (Zharikova & Sherstjuk, 2017). The problem of intelligent diagnosis in DSSs for disaster response has been dealt with in investigations of a number of authors. In (Balakrishnan & Honavar, 1998) the authors define a diagnosis system as one that is capable of identifying the problem by examining the observable symptoms (observations or assessments provided by monitoring). Such system is called intelligent if its diagnoses closely match those produced by a human.
It is necessary to give a short review of approaches to the development of intelligent diagnostic systems. One of the approaches is model-based used in the case when a domain model representing a structure of the system is available (Lucas, 2001; Kurien & R-Moreno, 2008).
In most practical tasks, precise models of the domain are unavailable. The most commonly used technique in such cases is the development of knowledge-based diagnosis systems [Bhagwat, 2015; Verbert, De Schutter & Babuska, 2015), which are based on heuristic knowledge about the domain and model the diagnostic reasoning of human experts. Knowledge-based systems have some limitations such as the inability to make creative responses as a human expert would in unusual circumstances and to adapt to changing environments.
Expert systems are a specified domain of such systems and are more specific than knowledge-based systems (Tan, Wahidin, Khalil, Tamaldin, Hu & Rauterberg, 2016; Amarosicz, Psiuk, Rogala & S. Rzydzik, 2016). They are typically built through a careful, tedious, and often expensive process of knowledge engineering (Martinez, 2017). There are some difficulties related to knowledge engineering because on the one hand the domain experts not always being able to explain their logic and reasoning sufficiently precisely to be encoded in a machine form, and on the other hand different experts often have a different understanding of the problem.