Forest Planning Model Based On Heuristics Algorithm Of Mengjiagang Farm

Forest management planning is one of important means for the implementation of sustainable forest management. It usually applies optimization methods for developing optimal management strategies. However, forest planning has some special characteristics such as multi-purposes, multi-levels, long terms, dynamics, spatial autocorrelations, and so on. In recent years, through scientific and rational forest management measures uses of forest resources have become predominating purposes and vital goals of forest management planning in order to continually and comprehensively utilize the benefits of forest resources. However, any silvicultural treatment and/or management operations cannot avoid the complicated spatial relationships between trees, stands, and forests. Aggregated cutting or selective cutting may be able to reduce the costs of forest management, but it is difficult to find the best solution by the traditional forest planning methods with spatial constraints. Therefore, modern heuristics algorithms have become popular to obtain the optimal strategies for sustainable forest management. Thus, heuristic-based forest management planning becomes one of burning issues in forestry.This study established stand growth and yield models and biomass models for major tree species in Wanda mountain of Heilongjiang, China. The data were collected from the fixed plots, stem/branch analyses and biomass studies in Wanda mountain of Heilongjiang Province from1990to2010. Forest spatial and non-spatial planning models were built with the goals of timber harvest, planned terminal volume and aggregated harvest degree. Using the thirty-year forest management planning with a ten-year planning interval as an example in Mengjiagang farm, random ascent algorithm, Hero algorithm, tabu search algorithm, simulated annealing algorithm and generic algorithmwere studied and compared for solving the sensitivity of major parameters in spatial (1-neighborhood) and non-spatial planning problems using the method of Hooke and Jeeves, and the basic rules of setting up key parameters of all heuristic algorithms were proposed for forest management planning. The changing processes of objective function values, timber harvest, planned terminal volume and adjacent cutting boundaries for different neighborhoods (1-neighborhood and2-neighborhood) were systematically analyzed based on optimal parameter combinations of various heuristic algorithms (1-neighborhood). Finally, the dynamics of forest resources characters (e.g., volume, biomass, carbon storage and age distribution) were analyzed based on the solution of2-neighborhood simulated annealing algorithm.The contributions and conclusions of this study include:1) Site class index, stand density index, individual tree growth models, mortality models, recruitment models, height-diameter equation, taper equation and biomass models of major species groups in Wanda mountain of Heilongjiang province were established. In general, all models were fit to the data reasonably well and could be applied for forest management planning in Wanda mountain of Heilongjiang province.2) Spatial and non-spatial planning models of forest resources were established for planned terminal volume, timber harvests of periods and aggregated cutting based on utility function theory, constraints of stand growth and spatial data of sub-compartment, and management measures of clear and selective cutting of stands.3) The key parameters of heuristic algorithms were not sensitive to planned terminal volume in non-spatial planning models, but had significant influences on solution time, which was increased by80%for solving each optimized parameter of every heuristic algorithm. The key parameters of heuristic algorithms were not sensitive to planned terminal volume in spatial planning problems, either, but had significant influences on function values, solution time and adjacent cutting boundaries. The solution time of1-neighborhood searching method in spatial planning problems increased about1000times. The average of objective function values increased about7%and the adjacent cutting boundaries were raised twice to those in the non-spatial planning problems.4) The key parameters of all heuristic algorithms were repeatedly optimized for20times in both spatial and non-spatial planning models and the basic rules were summarized as following:the random number of generating initial solutions was no less than100except generic algorithm and that was200in this study. For simple non-spatial planning models, the influence of setting up parameters of heuristic algorithms on objective function, that is, planned terminal volume, was not sensitive. Usually, the longer was iteration time, the better was the results. For spatial planning models, the iteration of random ascent searching algorithm was no less than500000times, simulated annealing algorithm was0.1-0.01, cooling rate was about0.9and cooling temperature was less than0.000001. The iteration of tabu search algorithm should not be less than5000times with tabu table length of50. For the forests with large selective cutting constraints using generic algorithm, chromosome number should be large (no less than50000), offspring number should be greater than50and mutation number should be5.5) Searching time of the2-neighborhood algorithms increased about2.5times than that of the1-neighborhood algorithms, while planned terminal volume of the2-neighborhood algorithms increased only0.22%, and the object function and aggregation cutting rate were enhanced about2%and15%, respectively, than the1-neighborhood algorithm. It indicated that the2-neighborhood searching techniques could improve the solution quality since2-neighborhood searching method could better arrange the location of sub-compartments for cutting and increase the common boundaries of adjacent cutting sub-compartments without changing other objectives.6) For spatial and non-spatial planning models, the Hero algorithm had the shortest searching time for solution, and followed by RA. SA could find the most optimized objective function values than others in a relatively reasonable time, and followed by Hero. GA had undesirable performance in solving spatial planning problems because neighborhood constraints and division and recombination of parents generation generated amount of infeasible solutions, only few parents generation had reasonable DNA, which limited the number of valid off-springs and influenced the quality of solutions. According to the performances of five heuristic algorithms, simulated annealing algorithm performed well in solving forest planning problems.7) All heuristic algorithms could achieve predetermined timber harvests (i.e.,3.0×105m3per planning period) as well as the continuous increase of planned terminal volume and requirements of stand aggregated cutting. However, the solution techniques of different algorithms had diverse trends of objective function values, adjacent cutting boundaries, planned terminal volume and timber harvest of periods. For non-spatial planning problems, the objective function values of RA, Hero, TS and SA rapidly reached maxima. For spatial planning problems, the objective function values and adjacent cutting boundaries of RA, Hero, TS and SA algorithms gradually increased by searching time, and then became stable; all indices of GA algorithm gradually increased and then became stable by searching time if it had enough chromosomes; the final volume of all algorithms reached about2.5×106m3, although planned terminal volume of diverse algorithms changed differently.This research studied procedures of solving forest planning models using five common heuristic algorithms and provided basic rules of setting up key parameters of the algorithms for different planning problems. The results would provide evidence of setting up schemes of forest management planning based on heuristic algorithms, and then offer decision-making and theoretical support for sustainable management of forest resources in this area.