Feed resource allocation optimization in dairy systems.
In this thesis, the problem of feed resources allocation to a heterogeneous dairy herd was studied. We focused on how to allocate available feed resources by grouping cows based in their energy requirements and distribute them to the available feed resources like pasture and/or supplements. This pro...
Պահպանված է:
| Հիմնական հեղինակ: | |
|---|---|
| Ձևաչափ: | doctoralThesis |
| Լեզու: | անգլերեն |
| Հրապարակվել է: |
2023
|
| Խորագրեր: | |
| Առցանց հասանելիություն: | https://hdl.handle.net/20.500.12008/43053 |
| Ցուցիչներ: |
Չկան պիտակներ, Եղեք առաջինը, ով նշում է այս գրառումը!
|
| Ամփոփում: | In this thesis, the problem of feed resources allocation to a heterogeneous dairy herd was studied. We focused on how to allocate available feed resources by grouping cows based in their energy requirements and distribute them to the available feed resources like pasture and/or supplements. This problem was modeled as a combinatorial optimization problem and solved with exact methods and Evolutionary Algorithms (EA). Considering that exact methods may have limitations due to a great computational demand (which causes extremely high executions times), initially, our approach used a single objective mathematical model and a Genetic Algorithm (GA). An experimental evaluation was performed in order to analyze the quality solution of the GA and to study how the resource allocation should be performed by interpreting the solutions’ structure for both methods. The results showed that the values obtained by the GA were very close to the exact values (the maximum gap value was 1.09% and the average gap value was 0.50%), but generating different assignment structures, presenting a good diversity and a wider exploration of the solutions’ space. In particular, we found many solutions with a very low gap value and large structural difference, reaching a maximum of 49.9%. Then due to the complexity of dairy systems and the need to contemplate several objectives, a Pareto-based multi-objective optimization with the Differential Evolution (DE) algorithm was applied. To evaluate the DE algorithm, we performed experiments to compare the solutions quality of the DE with exact Linear Programming (LP) solutions. As part of this analysis, the influence of different stocking rates (number of cows/ha) on milk production, feed allocation and economic performance indicators was also evaluated as a source of variation. The DE solutions that minimize the feeding costs for different stocking rates closely approached the solutions derived with LP (with average values slightly higher, between 0.4% and 5.6%), confirming the quality of the DE algorithm. The multi-objective model scenarios demonstrated that increasing stocking density would enhance milk production and gross margin per unit of area at largely unchanged productivity per animal, by shifting the feed ration from roughage to a large proportion of supplementary concentrate feed. In particular, for stocking rates of 1.1, 1.6, 2.1 and 2.6 cows/ha, gross margins of 6.1, 8.9, 11.8 and 14.7 US dollars/ha/day were obtained, respectively. From the results, we concluded that the multi-objective optimization with a Pareto-based DE algorithm was highly effective to explore the interrelations among conflicting objectives and to find suitable solutions. Finally, and considering there are many variants of EA which have different performances depending on the problem being solved, we decided to evaluate some of the most successful algorithms presented in the literature to address the feed resource allocation problem. In particular, a performance evaluation of four methods (two GA: NSGA-II, SPEA-2; and two DE algorithms: GDE-3, and the Pareto-based DE) was done. The algorithms were evaluated taking into account execution times, objective functions values attained, Pareto front comparisons and performance metrics values. The results showed significant differences between the algorithms in their ability to approach solutions in the Pareto front and in their computational times. In particular, the SPEA-2 algorithm obtained optimal values for all objectives, its solutions represented a large part of the Pareto front approximation, and it presented the best results in terms of convergence, diversity and cardinality; but required higher execution times. Depending on the parametric settings of the algorithms, the execution times of NSGA-II and GDE-3 were between 5 and 23 seconds, while the times of SPEA-2 were between 105 and 28400 seconds. |
|---|