基于粒子群的多BP网络并行非线性预测控制

针对传统预测控制算法在解决非线性系统控制问题时，存在难以建立精确的数学模型、控制精度不高等缺点，提出一种新的非线性系统预测控制方案。以多BP神经网络作为并行预测模型，克服误差积累以及网络规模庞大的缺点；运用粒子群优化(PSO)算法完成非线性预测控制的滚动优化。仿真表明，该方案的控制效果比常规动态矩阵控制效果有所提高，该方案是可行和有效的。     

From shallow to deep waters: habitats used by larval lampreys (genus Petromyzon and Lampetra) over a western European basin

Abstract –  Habitat requirements of lamprey ammocoetes (Petromyzon marinus and Lampetra genus) were investigated, for the first time, from shallow to deep waters, at different spatial scales across the Gironde‐Dordogne continuum, thanks to a water suction dredge. Fish‐habitat relationships were assessed through two complementary statistical analyses: habitat‐use curves and habitat suitability models using the Boosted Regression Trees (BRT) technique. Analyses were performed on a small‐size data set that was characterised by the low prevalence of lamprey. The sea lamprey larvae occurred in deeper areas than their Lampetra genus counterparts. ‘Pools’ of 2 m’ depth and more were optimal habitats for the former species. Among the environmental variables retained to model lamprey occurrences, the mesohabitat (a categorical variable) was demonstrated to be highly influential, in terms of fine grain‐size substratum and vegetation cover. These preliminary results suggest that monitoring using the water suction dredge method may contribute to sea lamprey conservation.     

Evaluation of unconstrained and constrained mathematical functions to model girth growth of rubber trees (Hevea brasiliensis) using young age measurements

No attempt has been made to date to model growth in girth of rubber tree (Hevea brasiliansis). We evaluated the few widely used growth functions to identify the most parsimonious and biologically reasonable model for describing the girth growth of young rubber trees based on an incomplete set of young age measurements. Monthly data for girth of immature trees (age 2 to 12 years) from two locations were subjected to modelling. Reparameterized, unconstrained and constrained growth functions of Richards (RM), Gompertz (GM) and the monomolecular model (MM) were fitted to data. Duration of growth was the constraint introduced. In the first stage, we attempted a population average (PA) model to capture the trend in growth. The best PA model wasfitted as a subject specific (SS) model. We used appropriate error variance-covariance structure to account for correlation due to repeated measurements over time. Unconstrained functions underestimated the asymptotic maximum that did not reflect the carrying capacity of the locations. Underestimations were attributed to the partial set of measurements made during the early growth phase of the trees. MM proved superior to RM and GM. In the random coefficient models, both Gf and G0 appeared to be influenced by tree level effects. Inclusion of diagonal definite positive matrix removed the correlation between random effects.The results were similar at both locations. In the overall assessment MM appeared as the candidate model for studying the girth-age relationships in Hevea trees. Based on the fitted model we conclude that, in Hevea trees, growth rate is maintained at maximum value at t0, then decreases until the final state at dG/dt ≥ 0, resulting in yield curve with no period of accelerating growth. One physiological explanation is that photosynthetic activity in Hevea trees decreases as girth increases and constructive metabolism is larger than destructive metabolism.     

Evaluation of uncalibrated preferential flow models against data for isoproturon movement to drains through a heavy clay soil

The uncalibrated predictive ability of four preferential flow models (CRACK‐NP, MACRO/MACRO_DB, PLM, SWAT) has been evaluated against point rates of drainflow and associated concentrations of isoproturon from a highly structured and heterogeneous clay soil in the south of England. Data were available for four plots for a number of storm events in each of three successive growing seasons. The mechanistic models CRACK‐NP and MACRO generally gave reasonable estimates of drainflow over the three seasons, but under‐estimated concentrations of isoproturon over a prolonged period in the first season and over‐estimated them in the two remaining seasons. CRACK‐NP simulated maximum concentrations of isoproturon over the first two events of each of the three seasons of 156, 527 and 24.4 µg litre^?1, respectively, and matched the observed data (465, 65.1 and 0.65 µg litre^?1) slightly better than MACRO (69.1, 566 and 58.5 µg litre^?1). Automatic selection of parameters from soils information within MACRO_DB reduced the emphasis on preferential flow relative to the stand‐alone version of MACRO. This gave a poor simulation of isoproturon breakthrough and simulated maximum concentrations were 0, 50.1 and 35.1 µg litre^?1, respectively. The capacity model PLM gave the best overall simulation of total drainflow for the first two events in each season, but over‐estimated concentrations of isoproturon (967, 808 and 51.3 µg litre^?1). The simple model SWAT represented total drainflow reasonably well and gave the best simulation of maximum isoproturon concentrations (140, 80.2 and 8.2 µg litre^?1). There was no clear advantage here in using the mechanistic models rather than the simpler models. None of the models tested was able to simulate consistently the data set, and uncalibrated modelling cannot be recommended for such artificially drained heavy clay soils. © 2001 Society of Chemical Industry     

Postcards from the past: charting the landscape-scale conversion of tropical Australian savanna to closed forest during the 20th century

Repeated sequences of digitised and geo-referenced historical aerial photography provide a powerful means of understanding landscape change. We use this method to demonstrate a landscape wide expansion of closed forest (42% increase in total coverage) in the Australian monsoon tropics over the past five decades. Retrospective habitat suitability models (HSI) of closed forest derived using four landscape measures (drainage distance, slope angle, aspect and elevation) for imagery taken in 1947 correctly forecast the subsequent spatial distribution of the expansion, with topographic fire protection primarily determining the closed-forest distribution. The dynamics of the closed forest-savanna boundary were predicted accurately by generalised linear models, with closed-forest expansion in fire-protected sites along forest edges and regression in the more fire-prone areas. Two factors may plausibly explain the expansion of closed forests. First, eco-ethnographic records stress the skilful use of fire by Aboriginal people in protecting isolated and locally resource-rich closed-forest patches. Second, the recent global increase in atmospheric CO₂ may be changing the competitive balance between savanna and forest by enabling C₃ trees to grow fast enough to escape the fire trap presented by flammable C₄ grasses.     

Comparison of the Sensitivity of Landscape-fire-succession Models to Variation in Terrain, Fuel Pattern, Climate and Weather

The purpose of this study was to compare the sensitivity of modelled area burned to environmental factors across a range of independently-developed landscape-fire-succession models. The sensitivity of area burned to variation in four factors, namely terrain (flat, undulating and mountainous), fuel pattern (finely and coarsely clumped), climate (observed, warmer & wetter, and warmer & drier) and weather (year-to-year variability) was determined for four existing landscape-fire-succession models (EMBYR, FIRESCAPE, LANDSUM and SEM-LAND) and a new model implemented in the LAMOS modelling shell (LAMOS(DS)). Sensitivity was measured as the variance in area burned explained by each of the four factors, and all of the interactions amongst them, in a standard generalised linear modelling analysis. Modelled area burned was most sensitive to climate and variation in weather, with four models sensitive to each of these factors and three models sensitive to their interaction. Models generally exhibited a trend of increasing area burned from observed, through warmer and wetter, to warmer and drier climates with a 23-fold increase in area burned, on average, from the observed to the warmer, drier climate. Area burned was sensitive to terrain for FIRESCAPE and fuel pattern for EMBYR. These results demonstrate that the models are generally more sensitive to variation in climate and weather as compared with terrain complexity and fuel pattern, although the sensitivity to these latter factors in a small number of models demonstrates the importance of representing key processes. The models that represented fire ignition and spread in a relatively complex fashion were more sensitive to changes in all four factors because they explicitly simulate the processes that link these factors to area burned. The US Government's and the Canadian Government's right to retain a non-exclusive, royalty-free license is acknowledged     

Agent-based land-use models: a review of applications

Agent-based modelling is an approach that has been receiving attention by the land use modelling community in recent years, mainly because it offers a way of incorporating the influence of human decision-making on land use in a mechanistic, formal, and spatially explicit way, taking into account social interaction, adaptation, and decision-making at different levels. Specific advantages of agent-based models include their ability to model individual decision-making entities and their interactions, to incorporate social processes and non-monetary influences on decision-making, and to dynamically link social and environmental processes. A number of such models are now beginning to appear—it is timely, therefore, to review the uses to which agent-based land use models have been put so far, and to discuss some of the relevant lessons learnt, also drawing on those from other areas of simulation modelling, in relation to future applications. In this paper, we review applications of agent-based land use models under the headings of (a) policy analysis and planning, (b) participatory modelling, (c) explaining spatial patterns of land use or settlement, (d) testing social science concepts and (e) explaining land use functions. The greatest use of such models so far has been by the research community as tools for organising knowledge from empirical studies, and for exploring theoretical aspects of particular systems. However, there is a need to demonstrate that such models are able to solve problems in the real world better than traditional modelling approaches. It is concluded that in terms of decision support, agent-based land-use models are probably more useful as research tools to develop an underlying knowledge base which can then be developed together with end-users into simple rules-of-thumb, rather than as operational decision support tools. This paper arises from research conducted as part of the UK Research Councils’ RELU Programme (award number RES-224-25-0102). RELU is funded jointly by the Economic and Social Research Council, the Biotechnology and Biological Sciences Research Council and the Natural Environment Research Council, with additional funding from the Department for Environment, Food and Rural Affairs and the Scottish Executive Environment and Rural Affairs Department.     

Predicting non‐native fish dispersal under conditions of climate change: case study in England of dispersal and establishment of pumpkinseed Lepomis gibbosus in a floodplain pond

Predictions of future climate change include shifts in patterns of precipitation, evapotranspiration and water run‐off, resulting in increased periods of drought as well as variability and intensity of rainfall events. In the United Kingdom, the non‐native North American sunfish, pumpkinseed Lepomis gibbosus (L.), is expected to benefit from these changes. We examine how hydrological variability induced by predicted changes in climate will affect the dispersal and spread of pumpkinseed in England by: (i) determining the relationship between discharge regime and pumpkinseed propagule pressure; (ii) examining a newly‐established pumpkinseed population following a flood event in 2007; and (iii) comparing the growth and life‐history traits of this new population with fish collected from the source population to demonstrate how the pumpkinseed's life‐history plasticity contributes to its success as a coloniser. Using Bayesian modelling, we determined that the number of pumpkinseed escapees is likely to increase with increasing discharge. The newly‐established pumpkinseed population showed fast juvenile growth, early age at maturity and small size at maturity. These traits differed significantly from the source population, specifically total length (TL) means at ages 1 and 2 were significantly greater in the new population, whereas TL at age 4 was significantly greater in the source population, and a significantly higher proportion of mature females were found at smaller size classes in the newly established pumpkinseed population. This study demonstrates the potential link between hydrological variability (current and future) and the dispersal of non‐native pumpkinseed, leading to the establishment of new populations.     

The fallacy of the missing middle: physics →…→ fisheries

In this paper I discuss some of the shortcomings of today's marine science in response to a recently published paper by Ann Gargett (1997). Three problem areas have been identified in the field linking climate forcing and fish production: First, the yet-to-be established observational proof for a relationship between the two; Second, the strongly neglected biology of organisms at evolutionary and ecological time scales; and thrird, the disregard for spatial and temporal scales in the discussion of mechanisms and supporting data.