| | SLO | ENG | Cookies and privacy

Bigger font | Smaller font

Show document Help

Title:ZGODNJE NAPOVEDI STOPNJE ALFA-KISLIN V HMELJNI RASTLINI (Humulus lupulus L.) Z MODELI STROJNEGA UČENJA
Authors:ID Pavlovič, Viljem (Author)
ID Rozman, Črtomir (Mentor) More about this mentor... New window
ID Bohanec, Marko (Comentor)
Files:.pdf DR_Pavlovic_Viljem_2011.pdf (2,71 MB)
MD5: 036ECDFEB6B704DEC1BAD92D90374E5C
PID: 20.500.12556/dkum/d41be720-8998-46a6-aa3a-f4c8df1956fd
 
Language:Slovenian
Work type:Dissertation
Organization:FKBV - Faculty of Agriculture and Life Sciences
Abstract:Hmelj (Humulus lupulus L.) je ključna surovina v pivovarstvu, saj značilno vpliva na organoleptične lastnosti piva, vključno z okusom in aromo. Grenčice hmelja (alfa-kisline) so že vrsto let eden najpomembnejših parametrov kakovosti in s tem tržne vrednosti hmeljskih proizvodov v vseh državah pridelovalkah. Zgodnje napovedi vsebnosti alfa-kislin v storžkih slovenskih kultivarjev hmelja, so zato ključnega pomena tako za hmeljarje, kot tudi za trgovce s hmeljem. V pomoč slovenskim ekspertom pri ocenah letnega pridelka alfa-kislin smo zgradili modele za zgodnje napovedi vsebnosti alfa-kislin treh pomembnejših kultivarjev, ki jih gojimo v Sloveniji. Vsi trije kultivarji (Aurora, Savinjski golding in Bobek) se precej enotno odzivajo na meteorološke vplive v istem časovnem obdobju. V raziskavi so s podatki in izkušnjami sodelovali tudi raziskovalci Inštituta za hmeljarstvo in pivovarstvo Slovenije v Žalcu. Analiza meteoroloških vplivov na območjih pridelave hmelja v Sloveniji je pokazala, da imajo močan vpliv na vsebnost alfa-kislin slovenskih kultivarjev predvsem nadpovprečno visoke temperature v obdobju nastajanja generativnih organov (r = –0,95, p < 0,001), padavine pa že v času intenzivne rasti hmeljne rastline (r = 0,94, p < 0,001), pri tem pa vpliv padavin ni linearen. Izračunali smo koeficient med logaritmirano vrednostjo skupne količine padavin od 21. maja do 22. julija in temperaturno vsoto od 18. junija do 22. julija. Soodvisnost med koeficientom (kTD) in vsebnostjo alfa-kislin modelnega kultivarja Virtual je močna (r = –0,94, p < 0,001) in statistično značilna. Precej enoten odziv kultivarjev Aurora, Savinjski golding in Bobek nam je omogočil gradnjo univerzalnega modela s pomočjo navideznega kultivarja Virtual in neodvisnih spremenljivk skupne količine padavin, temperaturnih vsot in teoretične potrebe po vodi za hmeljno rastlino. Za preliminarno napoved vsebnosti alfa-kislin v hmelju smo razvili univerzalni model G2G2DSMO, ki ga sestavljata dva modela z različnim vplivom. Oba modela sta zgrajena z atributom temperaturnih vsot v časovnem intervalu med 18. junijem in 21. julijem (T2529). Model G2SMO vsebuje še neodvisno spremenljivko skupne količine padavin za obdobje od 21. maja do 21. julija (D2129), model G2DSMO pa neodvisno spremenljivko teoretične potrebe po vodi v istem časovnem obdobju (Dd2129). Vpliv modela G2SMO na amalgamiran model G2G2DSMO je 60 %, s 40 % pa vpliva model G2DSMO. Povprečen model G2G2DSMO je dovolj natančen in robusten. Vrednost DW d-statistike = 1,91 dokazuje, da v modelu avtokorelacija ni prisotna. Model, ki smo ga v raziskavi razvili je v funkcijski obliki zapisan: G2G2DSMO = 14,025 – 13,46.10-3 T2529 + 2,92.10-3 Dd2129 + 3,72.10-3.D2129. S pomočjo značilnih konstant kultivarjev Aurora = 1,45, Savinjski_golding = 0,62 in Bobek = 0,92, ki smo jih izračunali, smo dobili napovedi vsebnosti alfa-kislin za posamezne kultivarje prve skupine. Vrednosti napovedanih odvisnih spremenljivk (Ki) so močno korelirane (rVirtual = 0,95, rAurora = 0,93, rSavinjski golding = 0,92, rBobek = 0,90) z njihovimi pravimi vrednostmi (Ci), ki so bile določene s kemijskimi analizami. Vse korelacije so statistično značilne pri stopnji tveganja manjši od 0,1 %. Model, ki smo ga v nalogi razvili, bo lahko služil kot osnova za nadaljevanje dela v smeri izgradnje ekspertnega sistema za zgodnje ocene vsebnosti alfa-kislin v hmelju. V prihod-nosti bo možno vključiti v model še nove kultivarje hmelja, pri čemer bo potrebno dodatno izračunati njihove modelne konstante.
Keywords:: napovedni model, strojno učenje, hmeljna rastlina, alfa-kisline, sušni management
Place of publishing:Maribor
Year of publishing:2011
PID:20.500.12556/DKUM-18507 New window
COBISS.SI-ID:3144492  New window
NUK URN:URN:SI:UM:DK:VRTMKAZP
Publication date in DKUM:31.05.2011
Views:5544
Downloads:421
Metadata:XML DC-XML DC-RDF
Categories:FKBV
:
Copy citation
  
Average score:(0 votes)
Your score:Voting is allowed only for logged in users.
Share:Bookmark and Share



Hover the mouse pointer over a document title to show the abstract or click on the title to get all document metadata.

Secondary language

Language:English
Title:MODELLING OF ALPHA-ACID CONTENT EARLY PREDICTION BY HOPS (Humulus lupulus L.) WITH MACHINE LEARNING MODELS
Abstract:Hops (Humulus lupulus L.) are vital for the brewing industry, as they contribute significantly to the organoleptic qualities of beer, including taste and flavor. Bitter substances in hops (alpha-acids) are already many years one of the most important quality parameters and the market value of the hop products in all producing countries. Early forecasts of alpha-acid contents in hop cones of the Slovenian varieties are therefore vital for both the hop growers, as well as for the merchants with hops. To help the Slovenian experts in estimates of annual yield of alpha-acids, we built models for early prediction of alpha-acid contents of three major cultivars grown in Slovenia. All three cultivars (Aurora, Savinjski golding and Bobek) are fairly uniform in response to meteorological influences in the same time period. The expertise and data from the staff of the Slovenian institute for hop research and brewing in Žalec are appreciated. Analysis of meteorological effects in hop production areas in Slovenia demonstrated that above-average high temperatures during a formation of generative organs (r = –0.95, p < 0.001), have mostly a strong influence on the alpha-acids of Slovenian varieties while rainfall at the time of intensive growth of the hop plant (r = 0.94, p < 0.001), with the effect of rainfall is not linear. We calculated the ratio between the logarithm value of total rainfall from 21st May to 22nd July and the temperature sum from 18th June to 22nd July. An interdependence between the coefficient (kTD) and the alpha-acid content of the model cultivar Virtual is strong (r = –0.94, p < 0.001) and statistically significant. Considerably uniform response of varieties Aurora, Savinjski golding and Bobek has allowed us to build a universal model through virtual cultivar Virtual and independent variables such as the total amount of precipitation, temperature sums and theoretical water demand for a hop plant. For a preliminary assessment of alpha-acids in hops we developed an universal model G2G2DSMO, which consists of two models with different influences. Both models are built with attribute temperature sums in the time interval between 18th June and 21st July (T2529). In addition, the model G2SMO contains the independent variable the total rainfall for the period from 21st May to 21st July (D2129), and the model G2DSMO an independent variable of theoretical water demand over the same time period (Dd2129). The impact of the model G2SMO on the model G2G2DSMO is 60%, while to 40% of the impact is related to the model G2DSMO. The average model G2G2DSMO is sufficiently precise and robust. The value of the DW d-statistics = 1.91 points out that the autocorrelation is not present in the model. The model we developed in this research is written in a functional form: G2G2DSMO = 14.025 – 13.46.10-3 T2529 + 2.92.10-3 Dd2129 + 3,72.10-3.D2129. With the help of characteristic hop variety constants that we calculated such as Aurora = 1.45, Savinjski_golding = 0.62 and Bobek = 0.92, we got the forecast levels of alpha-acids for the individual cultivars of the first group. Predicted values of dependent variables (Ki) are highly correlated (rVirtual = 0.95, rAurora = 0.93, rSavinjski golding = 0.92, rBobek = 0.90) with their true values (Ci), which were determined by chemical analysis. All correlations are statistically significant at the level of risk being less than 0.1%. The model that we developed in this research can serve as a basis for continued work toward building an expert system for early assessment of alpha-acids in hops. In the future, additional new hop varieties can be incorporated into the model, where only additional constants of those cultivars will need to be calculated.
Keywords:prediction model, machine learning, hops, alpha-acid content, draught management


Comments

Leave comment

You must log in to leave a comment.

Comments (0)
0 - 0 / 0
 
There are no comments!

Back
Logos of partners University of Maribor University of Ljubljana University of Primorska University of Nova Gorica