An ANN-based approach as support for decision-making processes regarding design of ecodistricts Giovanni Virgilio Department of Architecture and Spatial Planning, University of Bologna, Italy
AIM OF THE STUDY• The aim of this study is to methodologically delineate a process for providing public decision-making support for analysis and assessment of the performance ratings of eco- sustainable settlements on all levels.• The results obtainable from this methodological development are significant from the training angle, also for designers of eco-sustainable settlements, enabling learning and the gathering of knowledge from previous experiences in regard to urban contexts.• Hermeneutic analysis of a significant sample of relevant case studies is required in order to extract all information necessary for decision-making.
AIM OF THE STUDY• Given the considerable quantity of information and the extremely complex interactions of the analysed variables, however, considerable problems may arise which can objectively impede the operation as a whole, at least in terms of costs/derivable benefits.• An artificial neural network was therefore tested. Neural networks can extract new information from raw data by constructing computer-aided decision-making models shown to be efficacious as support for planner decision-making. Policymakers’ and planners’ assessment capabilities are also enhanced.
THE ECODISTRICTS• It is not easy to establish with precision the specific characteristics that connote an eco-sustainable district.• Within the ambit of urban development processes, an “urban green catalyst” role may be ascribed to such districts.• “ […] A urban green catalyst reflects the principles of sustainability, and is able to stimulate new dynamics that guide urban transformations: ecological, economic, social, institutional, etc. But it is the engine of change too, able to stimulate a new culture of urban planning, based on an integrated approach that combines reduction of consumes and use of renewable energies, innovation and local participation, creativity and good governance”• (Cerreta and Salzano, 2009).
THE ECODISTRICTS• It should now be clear what ecodistricts are. However, it is less clear how one is to plan for, or assess, the efficacy of the strategies adopted in creating an eco-compatible district.• Selecting efficacious design strategies is an increasingly difficult task since public decision-makers must face up to highly complex problems with limited resources.• Lessons can be drawn from consolidated practices already considered successful. Of course, these experiences cannot be transposed without considering the specific conditions of the territorial context within which they developed.
THE ECODISTRICTS• However, by analysing a significant sample of cases we can go some way toward isolating shared structural conditions which, in all likelihood, provide the essential premises for successful projects.• Enormous quantities of non-homogeneous data and items of information must be processed and rendered comparable (through in-depth study of the aforesaid cases).• For analysis, we require expert knowledge to extrapolate quantitative and qualitative information considered suitable for the purposes of pinpointing the structural characteristics of the cases analysed.• Furthermore, various sophisticated analytic instruments are required which are capable of ‘perceiving’ not only the nature and extent of interactions within a local territorial context but also the context’s ability to self-organise. Such instruments are vital, since traditional techniques are limited in their capacity to pinpoint the “implicit nexuses” subtending the dense web of interactions that turns territorial systems into complex systems.
METHODOLOGICAL APPROACH• This study is therefore methodological in nature. Its aim is to demonstrate that we can extract informaƟon on complexity from complexity itself − via the proposed method of systemic analysis using Artificial Neural Networks. The general aim is to pinpoint underlying trends characterising behaviour patterns, ascribable to actions, reactions and repercussions within urban systems.• Observation of territorial dynamics tells us that the territorial components are related to each other in a variety of manners. They form a lattice of “implicit nexuses” which we must uncover if we are to grasp the principles upon which the urban project is based (a lattice that generates efficacious performance through interaction with the territorial context).• An artificial neural network (ANN) is used for this purpose. The idea is to reveal the relations between the design options pertaining to the various ecodistricts and respective contextual attributes. ANNs can in fact capture the non-linear behaviour patterns of planning processes in complex urban systems.
METHODOLOGICAL APPROACHTo achieve all this, a fully articulated methodological approach isrequired.as outlined below. The steps which characterize themethodological path are the following:• Selecting case studies;• Analysis of experts group;• The choice of the neural network model;• The test phase: recognition of the identification profiles for the various ecodistricts;• Interrogation of the network: identification of ideal-type profiles
SELECTING CASE STUDIESThe case studies analysed were selected on the basis of the following characteristics:• they must have been developed in medium-sized or large European towns/cities, to ensure, at least in theory, uniformity of the basic problem areas, as well as potential homogeneousness of social, political and cultural dynamics;• the interventions must regard processes of new urban transformation (expansion or integration), and must be executed on an urban district scale;• the case studies have been implemented and are therefore economically and socially consolidated and comparable;• completed interventions are preferred, involving as many aspects of urban sustainability as possible;• a characteristic appearance specifically connoting intervention;• various typologies of promoters;• validity of the intervention, and availability of reference material.
SELECTING CASE STUDIES • On the basis of the above requisites, the following case studies were selected:- SCHEDA 01: Hammaeby Sjostad (Stockholm, Sweden);- SCHEDA 02: Bo01 (Malmö, Sweden);- SCHEDA 03: Viikki project (Helsinki, Finland);- SCHEDA 04: Greenwich Millenium Village (London, UK)- SCHEDA 05: Villaggio Olimpico ex MOI (Turin, Italy);- SCHEDA 06: Kronsberg (Hannover - Germany);- SCHEDA 07: Vauban (Freiburg - Germany);- SCHEDA 08: Südstadt (Tübingen- Germany);- SCHEDA 09: Scharnhauser Park (Ostfildern – Germany);- SCHEDA 10: Solarcity (Linz, Austria);- SCHEDA 11: GWL-Terrein (Amsterdam - Netherlands);- SCHEDA 12: EVA(Culemborg, Netherlands);
SELECTING CASE STUDIES• Structuring of information• This step provides an articulated classification of the information collected on various case studies, based on following criteria:
-Environmental sustainability: understood as the ability to4 SPHERES conserve the quality of natural resources and safeguard their reproducibility; -Economic and financial sustainability: understood as the ability to generate incomes and employment as means of support for the population; -Social sustainability: understood as the ability to ensure, in terms of class and gender, fairness in distribution of the conditions of human wellbeing (safety/security, health, education); -Political and institutional sustainability: understood as the ability to ensure conditions of stability, democracy, participation, justice.
4 SPHERES Amb.1. Biodiversity Amb.2. Soil Consumption Amb.3. Settlement Amb.4. Bioclimatic Amb.5. Energy23 GENERAL THEMES Amb.6. Water Amb.7. Materials Amb.8. Waste Amb.9. Transportation Amb.10. Health Eco.1. Governance Eco.2. Implementation Eco.3. Management Eco.4. Settlement Costs Eco.5. Economic Activities Soc.1. Population Soc.2. Quality Living Soc.3. Management Soc.4. Services And Public Facilities Pol.1. Implementation Pol.2. Participation Pol.3. Governance Pol.4. Management
Biodiversity: vegetation Health: Noise And Air Pollution4 SPHERES Biodiversity: Landscape Governance: Public Participation Soil consumption: pre-existing Governance: Private Participation Soil consumption: soil / subsoil Implementation: Acquisition Areas Soil consumption: land use Implementation: Programming Settlement: settlement location Implementation: Security Settlement: settlement morphology Manager: Maintenance Settlement: building typologies Management: Use23 GENERAL THEMES Bioclimatic: Building orientation Manager: Taxation Bioclimatic: natural lighting Settlement: Property Bioclimatic: Hygrothermal comfort Settlement: Social Housing Bioclimatic: ventilation Economic Activities: Job Offer Energy: reduction / control Economic Activities: Uses / Destinations Energy: optimizing resources Economic Activities: Tourism60 FEATURES Energy: use of renewable resources Population: Composition Water: outflow Population: Marginality Water: treatment Population: Aggregation / Inclusion Materials: construction Population: Social Security Materials: impact energy Quality of Living: Sense of Space Waste recovery Quality of Living: The Aesthetic Quality Waste management system Management: Sociability Waste treatment Services: Sociability Transportation: connectivity Services: Destination-scale Neighborhood Transportation: collective mobility Services Destinations Urban Scale Transportation: Individual Mobility Implementation: Ad Hoc Initiatives Transportation: slow mobility Participation: Involvement Transport: Road safety Governance: Institutional And Policy Health: Indoor Air Quality Governance: Political Orientation Management: Institutional And Communication
4 SPHERES23 GENERAL THEMES60 FEATURES some examples of targets .............131 TARGETS AMB A.1. - Integration of green space to built environment and all living spaces. AMB B.1. - Facilitate and promote action of plant vegetation. AMB C.1. - Environmental qualification of mobility infrastructures …………………………………………………………………………………………..
4 SPHERES23 GENERAL THEMES60 FEATURES some examples of strategies .............131 TARGETS AMB A.1.1. - Environmental connotation of public space AMB A.1.2. - Inserting inverdite between built areas AMB A.1.3. - Creation of green spaces for sport activities AMB B.1.1. - Protection of biodiversity AMB B.1.2. - Improve the greening of plants vegetation AMB B.1.3. - Formation of an urban network of link with the biological system of green land.285 STRATEGIES AMB C.1.1. - Environmental connotation of mobility spaces ……………………………………………………………………………….
ANALYSIS OF EXPERTS GROUP4 SPHERESThe phase encoding of the data thus allows us to define the MATRIX OF URBAN SUSTAINABILITY23 GENERAL THEMES60 FEATURES131 TARGETS285 STRATEGIES
ANALYSIS OF EXPERTS GROUP• The adopted approach has been focused on an analysis conducted by a group of experts based on the method of Nominal Group Technique (Delbecq et al., 1975). More precisely, here it has evaluated the level of implementation of various strategies, that is been identified by assigning a structured score as follows:• 0 = no strategy adopted• 0.5 = partially adopted strategy• 1 = full implementation of the strategy.• The final step is to assign a weight to each "general theme". This value expresses the significance that theme assumes in the development of the project. This result is easily achieved for all the specific themes, by adding the scores attributed to the different implemented strategies and by dividing the obtained value for the maximum score theoretically achievable. By classifying the various strategies adopted in the case studies according to the criteria of interpretation given below, we can establish the evaluation matrix (the matrix being based on the judgments of analysts or experts).
ANALYSIS OF EXPERTS GROUP• Interpretation of the information produced is an arduous task, and the sheer volume of this information very greatly complicates decision-making tasks. Above all, considering the various strategies adopted, an account of the (positive and negative) interactions noted cannot be fully provided.• It was therefore decided to employ an instrument capable of efficaciously and rapidly performing this function. When compared to those obtained exclusively through the assessments of experts, the error margins are generally encouraging.• The opportunities provided by Artificial Neural Networks are therefore of considerable interest.
THE CHOICE OF THE NEURAL NETWORK MODEL• There are different types of networks, for the purposes of our study, we have chose a neural network algorithm that exploits the internal recirculation. Recirculation Neural Network (RNN) were introduced by Geoffrey Hinton and James McClelland (Hinton and McClelland,1988).• This type of networks belongs to the family of autoassociative neural networks, their main feature is take input and output the same vector more specifically, in a RNN, data is processed in one direction and learning takes place using only local knowledge. The weights matrix of a RNN consists of maximum gradient connections between the Input and the Output layer.• Also, each element in both the hidden and visible layers are connected to a bias element. These connections have variable weights which learn in the same manner as the other variable weights in the network.
THE CHOICE OF THE NEURAL NETWORK MODEL• Therefore, if there are N Input Nodes in a RNN, the weights matrix Wij will be made up of N2 connections. The scheme of used RNN is shown in the following scheme: Massimo Buscema and the Semeion Group have developed a technique called the Re-entry (see Buscema, 1994,1999), that is repurposed in this study. The algorithm, on which this technique is founded, is very simple, but at the same time effective, as it allows to create a cycle in which the output that is generated by the network is re-entered into the network as new input. The process ends when the output generated is not subject to additional mutations.
THE CHOICE OF THE NEURAL NETWORK MODEL• The training phase We can, at this stage, proceed with the RNN training phase by inserting the following vector as real input: where GThi indicates the value attributed in the evaluation matrix to the h-th general theme relative to the i-th district. The term aij= 0 if j ≠ i and aij= 1 if j = i; the term in question therefore assumes the value of 1 only for the i-th district, thus enabling identification of the said term.
THE CHOICE OF THE NEURAL NETWORK MODEL • The data-input matrix• Slightly more than 14,000 ANN training cycles (epochs) took place. The error value noted is approximately 0,4% .• Following the training phase, the network should be capable of reconstructing its own matrix of weights (which can be attributed to the various strategies applied to the analysed districts) and of providing useful indications if appropriately interrogated.•
THE CHOICE OF THE NEURAL NETWORK MODEL• The test phase: recognition of the identification profiles for the various ecodistricts• At this stage, verification of the level of “self-awareness” reached by the neural network becomes important. In other words, we must verify whether the network is capable of reconstructing the profile of the 12 districts.• Hence, the aim is to test the capacity of the network to recognise a given district, to see what scores the network assigns to the set of eco-oriented strategies, and then to assess the deviation between the data of the original evaluation matrix (produced by expert group) and the data of the evaluation matrix produced by the network.• The Index of compliance obtained as total value of the ratio for the value proposed by the neural network (NN) and the real value (R) is meanly equal to 0.95. A lower mean deviation of 5% is noted between the profile proposed by the neural network and the real profile of the 12 ecodistricts. All in all, for the scopes of this study, this deviation may be considered satisfactory
THE CHOICE OF THE NEURAL NETWORK MODEL• Interrogation of the network: identification of ideal-type profiles• At this stage we are in a position to interrogate the network in order to obtain indications relative to the performance ratings of individual districts. In other words, the aim of this phase is that of assessing which of the analysed cases may constitute a model (and therefore an ideal-type profile) with reference to the strategies adopted in order to deal with the four spheres of sustainability.• Hence, the first type of interrogation aims to pinpoint 4 ideal-type district profiles, a summary description of which may be provided as follows:• Environmental Profile: this profile is characterised by a focus on the sphere of environmental sustainability, and, therefore, the strategies which are adopted aim, primarily, at ensuring environmental quality of the settlement and reduction of consumption of natural resources;• Social Profile: for this profile the strategies receiving most attention are those that aim at attaining conditions for the settlement such as are capable of fostering social cohesion and inclusion;• Economic Profile: this profile is characterised by the mainly economic nature of the strategies adopted, which aim to produce wealth and jobs for the inhabitants;• Political Profile: here, the focus is on forms of direct democracy and on the manners of participation through which direct democracy is implemented in collective decision-making processes.
INTERROGATION OF THE NETWORK: IDENTIFICATIONOF IDEAL-TYPE PROFILES• In the cases analysed, of course, these characteristics coexist (with varying intensities). Our investigation aims to identify which characteristic emerges significantly, to such an extent that it univocally connotes the case analysed.• Operationally speaking, it is a question of constructing an input vector relative to each district which, in regard to the sphere analysed, shall present maximum intensity relative to the themes characterising it. ENVIRONMENTAL PROFILE amb.1 1,00 amb.6 1,00 eco.1 0,00 soc.1 0,00 pol.2 0,00 INDICATORS: amb.2 1,00 amb.7 1,00 eco.2 0,00 soc.2 0,00 pol.3 0,00 ACTIVED amb.3 1,00 amb.8 1,00 eco.3 0,00 soc.3 0,00 pol.4 0,00 amb.4 1,00 amb.9 1,00 eco.4 0,00 soc.4 0,00 input amb.5 1,00 amb.10 1,00 eco.5 0,00 pol.1 0,00 The network result is the following (case of Hammarby Sjöstad) ENVIRONMENTAL PROFILE amb.1 0,74 amb.6 0,75 eco.1 0,45 soc.1 0,37 pol.2 0,10 STRATEGIES amb.2 0,73 amb.7 0,68 eco.2 0,38 soc.2 0,52 pol.3 0,23 IDENTIFIED amb.3 0,65 amb.8 0,65 eco.3 0,48 soc.3 0,28 pol.4 0,53 amb.4 0,68 amb.9 0,69 eco.4 0,18 soc.4 0,61 output amb.5 0,75 amb.10 0,54 eco.5 0,34 pol.1 0,88
INTERROGATION OF THE NETWORK: IDENTIFICATIONOF IDEAL-TYPE PROFILES• The network identifies Hammarby Sjöstad as the district, among those analysed, with the most marked pertaining characteristics .• However, the network also indicates the importance assumed, for the project’s success, both by community involvement during the implementation phase and by the array of services at hand.• This latter aspect reveals the network’s ability to pinpoint the implicit nexuses which come about among the various strategies, underscoring an ability to fully grasp the complexity of the system and provide valid support during the decision-making phase.
Interrogation of the network: identification of ideal-type profiles ENVIRONMENTAL PROFILE HAMMARBY SJÖSTAD (STOCKHOLM, SWEDEN)
INTERROGATION OF THE NETWORK: IDENTIFICATIONOF IDEAL-TYPE PROFILES • Likewise, the other ideal-type profiles were analysed. While, for all interrogations, the network is capable of detecting the presence of the analysed characteristic, in the case of the social and political profiles the intensity of the characteristic is not high. • This is probably due to the scarcity of the information gathered during the reconnoitring phase relative to these two aspects. Hence the input data were found to be fairly homogeneous. In any case, this fact underscores even more clearly the potentials provided by the network during exploratory investigation, in that the network manages (at least in part) to highlight districts in which certain characteristics are most marked.
INTERROGATION OF THE NETWORK: IDENTIFICATION OF IDEAL-TYPEPROFILES ECONOMIC PROFILE SÜDSTADT (TÜBINGEN- GERMANY)
INTERROGATION OF THE NETWORK: IDENTIFICATION OF IDEAL-TYPEPROFILES SOCIAL PROFILE KRONSBERG (HANNOVER - GERMANY)
INTERROGATION OF THE NETWORK: IDENTIFICATION OF IDEAL-TYPEPROFILES POLITICAL PROFILE SÜDSTADT (TÜBINGEN- GERMANY) GREENWICH MILLENIUM VILLAGE (LONDON, UK)
INTERROGATION OF THE NETWORK: IDENTIFICATIONOF IDEAL-TYPE PROFILES• The next step consists in an analysis of the performance ratings of various districts in terms of the mix of strategies adopted. Here, too, the responses obtained on interrogation of the network are of particular interest.• As above, here too, from the operational viewpoint, no major problems arise, because it will be sufficient to activate with maximum intensity the themes that shall constitute the analysed combination of strategies. This can be readily achieved by assigning a value of 1 to the theme.• With the first theme that shall be illustrated, the idea was to carry out analysis in order to understand which mix will be capable of simultaneously ensuring encouraging performance ratings in the fields of environmental and economic options.
INTERROGATION OF THE NETWORK: IDENTIFICATIONOF IDEAL-TYPE PROFILES• The network identifies the winning strategies: – strengthening the transportation network and the coexistence of various forms of mobility; – the presence of public-private partnerships capable of activating efficacious governance processes; – accessibility of the settlement; – development and safeguarding of water as a resource; – development of economic activities, with particular attention paid to activities with low environmental impact; – activation of processes of implementation and political management.• Hammarby is the district which most satisfactorily brings together the aspects investigated in this profile. Hammarby not only confirms its environmental vocation; it is noted, also, that this vocation is even enhanced by the presence of strategies aiming at promotion of economic aspects
INTERROGATION OF THE NETWORK: IDENTIFICATIONOF IDEAL-TYPE PROFILES ENVIRONMENTAL-ECONOMIC PROFILE HAMMARBY SJÖSTAD (STOCKHOLM, SWEDEN)
INTERROGATION OF THE NETWORK: IDENTIFICATIONOF IDEAL-TYPE PROFILES• The last two profiles presented are socio-economic and political - economic.• In regard to the socio-economic profile, the network identifies as winning strategies implementation and development of economic activities together with collective forms of management. Thus we see a strengthening of the position of Kronsberg, which becomes the case study most capable of representing this profile.
INTERROGATION OF THE NETWORK: IDENTIFICATIONOF IDEAL-TYPE PROFILES• To conclude, the political and economic profile is the final profile recognised by the network.• This profile is characterised by special attention paid to development of strategies that foresee implementation of forms of political and economic governance; the participation of citizens; diversification of typologies of home ownership and, in particular, the presence of social housing; integration and integrated development of various forms of transportation and mobility; and forms of community management. The Südstadt district practically fully embodies the characteristics identified by the network, thereby confirming a satisfactory level of integration of the strategies adopted with reference to the problem areas analysed.
CONCLUSIONS• The methodological approach which has been developed has enabled a full appreciation of the structural complexity of the case studies analysed, while enabling identification of the particular characteristics which render exemplary a number of these cases.• The analytic instrument adopted has been found to be extremely powerful and efficacious in its ability to identify the “implicit nexuses” coming about among the various territorial components, and which are capable of conditioning the success of the various implemented strategies. Hence, from this point of view, the neural network is capable of pinpointing not only synergic dynamics but also potentially antagonistic dynamics, thereby most surely facilitating decision-making processes.
CONCLUSIONS• However, we should also note the manners in which the neural network’s assessment process is implemented. Since this process is based on a “black box model”, an account of the associative processes cannot be exhaustively provided. A potential danger, in this regard, may lie, for example, in inappropriate levelling of the proposals produced – as a consequence of homogenisation of information. This situation may come about where information in regard to the contexts of decision-making – where the choices are made – is insufficient. In other words, the phase of collection of information becomes vital.