Modelling a Simulation-Based Decision Support System for Effects-Based Planning

Models constitute an important component in decision support functions of C4I systems. They improve the military commander’s abilities to create situation awareness, analyse threats and make proper decisions. Computerized models can be exploited for the purpose of simulation, which enables us to cover a much wider range of options and go deeper in impact assessments. In this paper we describe decision support and simulation techniques to facilitate Effects-Based Planning (EBP). In our approach, by using a decision support tool, a decision maker is able to test a number of feasible plans against possible courses of events and decide which of those plans is capable of achieving the desired military end-state. The purpose is to evaluate plans and understand their consequences through simulating the events and producing outcomes which result from making alternative decisions. Plans are described in the EffectsBased Approach to Operations (EBAO) concept as a set of effects and activities that together will lead to a desired military end-state. For each activity we may have several different alternatives. Together they make up all alternative plans, as an activity tree that may be simulated. The simulation of plans is designed to deliver results, indicating the (so far) best sequence, at each point of time. Hence, we have chosen to use the A*-search algorithm for traversing through the activity tree and choosing the next activity to be simulated. This method helps us to decide at any point of time which sequence of activities has the best result so far, i.e., has resulted in a system state that is “closest” to our end-state. The paper also includes a description of our model, different objects, their relations, and the structure of our simulation kernel. The system is still under development hence there are no experimental results obtained so far. 1.0 INTRODUCTION Models constitute an important component in decision support functions of C4I systems. The work of military commanders was previously mainly based on using mental models and maps for creating situation awareness, analysing threats, investigating different actions and making decision based on those. The development of computers has meant opportunities for a radical increase in capacity for these activities. By leveraging databases with information about the enemy, environment and past experience, we are strengthening our models and thereby our understanding. By further formulating our models in computerized form we also increase our ability to exploit them for simulations, by which we can both cover a much wider range of options and go deeper in impact assessments [1]. With regard to information interpretation, i.e., creation of situation awareness and situation understanding and threat analysis, models play a central role when it comes to assimilate large amounts of complex information from various sensors (including humans). One application is Information Fusion, which can be seen as a real-time simulation of the real course of events. Information Fusion gives the commander Modelling a Simulation-Based Decision Support System for Effects-Based Planning 11 2 RTO-MP-MSG-069 and his staff better ability to quickly discern which reports are linked and what is going on. It also improves the commander’s ability to control its sensory resources to achieve desired situational awareness [2]. Another way to use models in information interpretation is the so-called indicator modelling [3]. Here we assume a case database where situations from previous missions are stored together with a set of events that follow that situation, and a final event, such as a riot. With an appropriate situation model, we can then compare a current situation with the database situations, and select those that are most similar. The decision-maker is thus made aware of the similar corollary events, which indicate that the corresponding end event can occur even now. As for evaluating the impact of alternative decisions, it is possible to use expert systems models which simulate experts’ way of drawing conclusions [4]. They allow for “faster” decision-making for the experienced, and “safer” decisions for the less experienced, by ensuring that “no” relevant aspects are forgotten. The disadvantage is that they can easily lead to a stereotyped and predictable behaviour, which might be partly handled by making the system self-learning. However, a more flexible decision support can be achieved with a tool that in addition to an expert system also includes a model with which we can simulate the events and produce outcomes resulting from making alternative decisions [5]. Such a tool should not only simulate a single thread of activities, but rather calculate statistical values of outcomes of different threads. It must also have a high credibility, and be fast to allow many simulations. Hence, the model may not be too detailed. Nevertheless, large plans with many alternative activities may take a long time to execute. A more practical simulation system should be able to, at any moment in time, suggest an alternative that best suits the commander’s criteria of a successful plan. Furthermore, it must have a flexible and easy to use interface, which enables us to quickly and easily define different alternatives and modify the model as our knowledge of other actors’ characteristics/capabilities grows. “Real-time Simulation for Supporting Effects-Based Planning” is an ongoing research project at the Swedish Defence Research Agency, which was initiated in 2008 with the goal of designing and developing a simulation-based decision support system for supporting the planning process of the EffectsBased Approach to Operations (EBAO) [6, 7]. In this paper we present the approach employed in the project to achieve the above goal. The decision support system enables a decision maker to test a number of feasible plans against possible courses of events and decide which of these plans is capable of achieving the desired military end-state. The purpose is to evaluate plans and understand their consequence through simulating the events and producing outcomes which result from making alternative decisions. The simulation of plans with a sequence of alternative activities is designed to deliver results, indicating the (so far) best sequence, at each point of time. Hence, we have chosen to use the A*-search algorithm [8, 9] for traversing through the activity tree and choosing the next activity to be simulated. This method helps us to decide at each point of time which sequence of activities has the best result so far, i.e., has resulted in a system state that is “closest” to our end-state. The remainder of the paper is structured as follows. In Sec. 2 we describe the overall Decision Support System giving support on operational planning by testing and evaluating alternative operational plans. In Sec. 3 we present our model, different concepts, and their relations. Sec. 4 gives an introduction to our simulation and its structure. Finally, Sec. 5 concludes the paper with current status of the project, general conclusions and future work. Modelling a Simulation-Based Decision Support System for Effects-Based Planning RTO-MP-MSG-069 11 3 2.0 DECISION SUPPORT In EBAO as in any other planning process, there is a need to assess possible plans before execution (Effects-Based Execution (EBE) in the context of EBAO) and to perform re-planning when necessary. The aim is to evaluate the plans, including discovering its weaknesses and understanding its implications. An important prerequisite for good planning is to find and use appropriate indicators so that intelligence questions can be asked. 2.1. Analyzing the operational plan Analyzing and simulating the operation plan can be made at any time. When an operation begins to take shape one should be able to analyze and simulate several alternative plans that are in the main direction of interest. This task uses large numbers of simulations with different alternative plans against various possible scenarios. The goal is to find robust groups of plans that have similar implications. We assume the plans and let them control the evaluation process. All activities of the plans are to be simulated against all events in the chain of events. The operational plan is simulated by providing a wide range of simulation tasks to the simulator. Any such assignment is made up of a particular alternative for a particular activity. This simulation task has a specific location in a tree of plans, where each level represents a new activity, and each branch one of the different options available for this activity. The decision maker can prioritize between different plans by letting the simulator know his current area of interest, times of interest, and through an activity clustering choose groups of prioritized activities, see figure 1. Figure 1: Through an input interface the user may select which part of the plan the simulator should focus on such as, an area in a map, start and end time in a Gantt schema, or clusters where activities that strongly influence each other are grouped together. These inputs are fused in the lower right area using the function μ (see below). Modelling a Simulation-Based Decision Support System for Effects-Based Planning 11 4 RTO-MP-MSG-069 This guidance will allow the simulator to focus on alternatives that are of interest to the decision maker. The function μ is showing decision-maker current interest in a particular activity. We have, where {μj} is drawn from all the views that the decision-maker uses for his prioritization of simulation tasks to the simulator. An event’s overall significance from an effects-based approach is obtained by the a priori information given by the function ω. This information is retrieved from a cross-impact matrix (CIM) [10]. We have, An assessment is made of how well each activity is managed by its mission. All such estimates based on various simulation tasks are stored in order to rapidly be re-used by future simulation and is transferred to decision support system so that a consolidated assessment can be made. The compilation of partial results from all simulation activities can be made by the statistical method developed for subjective Effects-Based Assessment (EBA) [11], where each assessment activity takes place separately, and are then fused statically to an overall assessment of entire operation plan. Plans are judged by their robustness. This is measured, not by the score the plan receives itself, but rather by the minimum score of all other plans that are structurally close, as well close in their consequences. We use an information measure to measure the structural distance between two plans Pi and Pj, or between two events Hi and Hj. We choose the Hamming distance [12] Hamming_distance Pi Pj , ( ) 0 Pi.Ak Pj.Ak = , 1 Pi.Ak Pj.Ak ≠ , ⎩ ⎨ ⎪ ⎧