Seite - 136 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

Figure3. ComparisonbetweenthepredictionoftheLSTMandtheactual temperatureofthefirst200instances of test3 (fold3)of the3-foldcrossvalidationexperiments. Figure4. Comparisonbetween theLSTMpredictionand theactual temperatureof the144 temperaturemea- surementsobtainedon17February2019, this corresponds toa full 24-hourday. 5. ConclusionsandFutureWork Thepredictionof lowtemperatures isa latentproblemin theagrarianworld.Knowingin advancewhetherornot lowtemperatureswilloccurhelps thefarmer toforeseeresources and apply anti-frost control techniques early enough to ensuremaximumeffectiveness. In this work an preliminary predictive model for temperature has been analyzed and constructedusingaLongshort-termmemoryneuralnetwork.TheLSTMhasbeentested usingactual temperaturedata providedbyan IoTsystemdeployedonvariousplots and currently in operation. The preliminarymodel obtained by the LSTM for temperature predictionwillbepartof theintelligentcomponentoftheIoTsysteminordertowarnand alert the farmerassoonaspossible.The twoexperimentscarriedout to test theproposed modelhaveobtainedsimilarresultsandwithquitesatisfactorygoodnessobtaininga99% modelfitandanRMSEof less than0.8degreesCelsius. Inaddition, type2errors (where M.Á.Guillén-Navarroetal. /AnLSTMDeepLearningScheme forPredictionofLowTemperatures136