Análise de Sensibilidade na Plataforma + Precoce

Abstract

The Plataforma + Precoce (P+P) is an application that allows to model and simulate improved beef cattle production systems, assisting the producer in decision making and providing information about cow-calf, rearing and finishing systems for beef cattle production. Currently, P+P alows a user to simulate results from beef cattle production systems only “point to point”, i.e., it’s only possible to perform the simulation of a givem system by selecting specific values for each parameter, but without directly understanding how much the variation of these parameters can influence the final result of system’s indicators. This work aimed to develop a web application for sensitivity analysis through graphics in production systems used by P+P, so the user is able to identify relevant parameters for the result of an indicator in a given system. One of the main aspects in the tool development was to use methods that could be executed in the client's browser, with the minimum possible requests to the server. With that in mind, we opted for the use of analysis methods by screening and local analysis, mainly due to their low complexity in the method execution. The screening analysis methods make use of a "control scenario" where each parameter has a minimum and a maximum value and the parameters are varied individually and analyzed throughout this interval. This is very interesting because the parameters of the P+P have already been defined in advance with a default value, minimum and maximum, which facilitates the method implementation. On the other hand, the methods of local analysis work under the variation of the parameters individually only in the proximity of a standard value. In this way, it was verified that the values related to the parameters of a system rarely change, so that the sensitivity analysis for these parameters can be persisted in the server in the future, without the need to be calculated in real time by the client. A simulation, on the other hand, has the parameter values calibrated in real time by the user, which requires sensitivity analysis also in real time. For this reason, we opted for the analysis of a larger range of parameter values (analysis by screening) for systems analysis and the use of local analysis for analysis of simulations. To analyze all variations of a parameter in the established intervals, all other parameters are kept at their default values, while for each variable point of a parameter, the P + P Simulator is activated. The results are presented through graphics to facilitate the reading and understanding by P + P users. In the developed tool it’s possible to have four types of graphs, three related to the system and one related to a simulation. The graphs related to the system are tornado graphs, graph indicator/parameter and probability histogram. The graph related to a simulation is represented by a simulation tornado graph. To avoid recalculation and calls to the Simulator with the same parameter values, which would produce the same result for the indicators, a "memoize" technique was used, which consists of storing the values already calculated in a "research table", in order to always refer to it when a new calculation needs to be performed. In this way, if a calculation has already been carried out for a set of parameter and indicator values, it is only retrieved from the table, with the Simulator call made only in cases where the combination is not found in the research table. A limitation found during the development is that, at this moment, the P + P Simulator is undergoing an update and technology change that is still in progress, which made it impossible to integrate the developed tool with P + P. In order to solve this problem and carry out the necessary call tests to the Simulator, a "pseudo-simulator" was developed, which will be replaced as soon as the real P + P Simulator is updated. It’s expected that the development of this tool will assist beef cattle producers and technicians in making even more efficient decisions, thus increasing beef cattle quality and productivity.