Modeling the policy making process is a very challenging task. To the best of our knowledge the most widely used technique in this setting is agent-based simulation. Each agent represents an individual entity (e.g., citizen, stakeholder, company, public association, public body). The agent interaction enables emerging behaviours to be observed and taken into account in the policy making process itself. We claim that another perspective should be considered in modeling policy issues, that is the global perspective. Each public body has global objectives, constraints and guidelines that have to be combined to take decisions. The policy making process should be at the same time consistent with constraints, optimal with respect to given objectives and assessed to avoid negative impacts on the environment, economy and society. We propose in this paper a constraint-based model for the global policy making process and we apply the devised model to the regional planning activity. A case study in the field of energy plan is used to evaluate the proposed model. Clearly an interaction with agent-based simulation is desirable and could provide important feedback to the global model. This aspect is the subject of current research.
Public policy issues are extremely complex, occur in rapidly changing
environments characterized by uncertainty, and involve conflicts among different
interests. Our society is ever more complex due to globalisation, enlargement and the
changing geo-political situation. This means that political activity and
intervention become more widespread, and so the effects of its interventions become more
difficult to assess, while at the same time it is becoming ever more important
to ensure that actions are effectively tackling the real challenges that this
increasing complexity entails. Thus, those responsible for creating, implementing,
and enforcing policies must be able to reach decisions about ill-defined problem
situations that are not well understood, have no single correct answer, involve
? An extended version of this paper appeared in [
In this paper we consider policy issues related to regional planning, the
science of the efficient placement of activities and infrastructures for the sustainable
growth of a region. Regional plans are classified into types, such as Agriculture,
Forest, Fishing, Energy, Industry, Transport, Waste, Water, Telecommunication,
Tourism, Urban Development and Environment to name a few. Each plan
defines activities that should be carried out during the plan implementation. On
the regional plan, the policy maker has to take into account impacts on the
environment, the economy and the society. The procedure aimed to assess the
impacts of a regional plan is called Strategic Environmental Assessment [
One important aspect to be considered for supporting policy makers with
Computational Intelligence approaches is the definition of formal policy models.
In the literature, the majority of policy models rely on agent based simulation
[
In addition to agent-based simulation models, which provide “individual level models”, we claim that the policy planning activity needs a global perspective: in the case of regional planning, we need “a regional perspective” that faces the problem at a global level while tightly interacting with the individual level model. Thus rather than proposing an alternative approach with respect to simulation, we claim that the two approaches should be properly combined as they represent two different perspectives of the same problem: the individual and the global perspective. This integration is the subject of our current research activity. In this setting, regional planning activities can be cast into complex combinatorial optimization problems. The policy maker has to take decisions satisfying a set of constraints while at the same time achieving a set of (possibly conflicting) objectives such as reducing negative impacts and enhancing positive impacts on the environment, the society and the economy. For this reason, impact assessment should be integrated into the policy model so as to improve the current procedure performed a posteriori.
In previous work [
In this work, we extend the CLP model used for the assessment, and apply it
to the planning problem, i.e., deciding which actions should be taken in a plan.
In the model, decision variables represent political decisions (e.g., the magnitude
of a given activity in the regional plan), potential outcomes are associated with
each decision, constraints limit possible combination of assignments of decision
variables, and objectives (also referred to as criteria) can be either used to
evaluate alternative solutions or translated into additional constraints. The model has
been solved with CLP [
Further constraint based approaches have been proposed for narrower
problems in the field of energy, such as locating biomass power plants in positions
that are both economically affordable [
Regional Planning is the result of the main policy making activity of European regions. Each region has a budget distributed by the Operational Programme (OP): an OP sets out each region’s priorities for delivering the funds. On the basis of these funds, the region has to define its priorities: in the field of energy, one example of priority is increasing the use of renewable energy sources. Then, a region should decide which activities to insert in the plan. Activities may be roughly divided into six types: infrastructures and plants; buildings and land use transformations; resource extraction; modifications of hydraulic regime; industrial transformations; environmental management. Also, a magnitude for each activity should be decided describing how much of a given activity is performed.
Each activity has an outcome (such as the amount of energy produced or consumed) and a cost. We have two vectors O = (o1, . . . , oNa ) and C = (c1, . . . , cNa ) where each element is associated to a specific activity and represents the outcome and cost per unit of an activity. There are constraints linking activities: for instance if a regional plan decides to build three biomass power plants (primary activities for an energy plan), each of these plants should be equipped with proper infrastructures (streets, sewage or possibly a small village nearby, power lines) also called secondary activities. We thus have a matrix of dependencies between activities. In particular, we have a Na × Na square matrix D where each element dij represents the magnitude of activity j per unit of activity i.
Taking as an example the Emilia-Romagna Regional Energy Plan approved in 2007, some objectives of the policy makers are the production of a given amount of energy (400 additional MW from renewable energy sources), while reducing the current greenhouse gas emission percentage by 6.5% with respect to 2003. In addition, the budget constraint limiting the amount of money allocated to the energy plan by the Regional Operational Programme was 30.5Me in 2007.
The policy maker also takes into account impacts on the environment, the economy and the society, as defined by a Strategic Environmental Assessment that relates activities defined in the plan to environmental and economic impacts. Each activity has impacts on the environment in terms of positive and negative pressures. An example of positive pressure is the increased availability of energy, while an example of a negative pressure is the production of pollutants. Pressures are further linked to environmental receptors such as the quality of the air or of surface water. On both pressures and receptors, there are constraints: for example the maximum amount of greenhouse gas emissions of the overall plan.
One of the instruments used for assessing a regional plan in Emilia-Romagna
are the so called coaxial matrices [
One matrix M defines the dependencies between the above mentioned activities impacts (also called pressures) on the environment. Each element mij of the matrix M defines a qualitative dependency between the activity i and the impact j. The dependency can be high, medium, low or null. Examples of negative impacts are energy, water and land consumption, variation of water flows, water and air pollution and so on. Examples of positive impacts are reduction of water/air pollution, reduction of greenhouse gas emission, reduction of noise, natural resources saving, creation of new ecosystems etc.
The second matrix N defines the dependencies between the impacts and environmental receptors. Each element nij of the matrix N defines a qualitative dependency between the impact i and an environmental receptor j. Again the dependency can be high, medium, low or null. Examples of environmental receptors are the quality of surface water and groundwater, the quality of landscapes, energy availability, wildlife wellness and so on.
The matrices used in Emilia-Romagna contain 93 activities, 29 negative impacts, 19 positive impacts and 23 receptors, and assess 11 types of plans. 2
The regional planning activity is now performed by human experts that build a single plan, considering strategic regional objectives that follow national and EU guidelines. After the plan has been devised, the agency for environmental protection is asked to assess the plan from an environmental point of view. Typically, there is no feedback: the assessment can state that the devised plan is environmentally friendly or not, but it cannot change the plan. In rare cases, it can propose corrective countermeasures, that can only mitigate the negative impact of wrong planning decisions. Moreover, although regulations state that a significant environmental assessment should compare two or more options (different plans), this is rarely done in Europe, because the assessment is typically hand made and requires a long work. Even in the few cases in which two options are considered, usually one is the plan and the other is the absence of a plan.
Constraint based modeling overcomes the limitation of a hand made process for a number of reasons. First, it provides a tool that automatically performs planning decisions, considering both the budget allocated to the plan by the Regional Operative Plan, and national/EU guidelines.
Second, it takes environmental aspects into consideration during plan construction, avoiding trial-and-error schemes.
Third, constraint reasoning provides a powerful tool in the hand of a policy maker as the generation of alternative scenarios is extremely easy and their comparison and evaluation comes for free. Adjustments can be performed onthe-fly in the case that the results do not satisfy policy makers or environmental experts. For example, in the field of energy regional plan, by changing the bounds on the amount of energy each source can provide, we can adjust the plan by considering market trends and also the potential receptivity of the region. 3
To design a constraint-based model for the regional planning activity, we have to define variables, constraints and objectives. Variables represent decisions that have to be taken. Given a vector of activities A = (a1, . . . , aNa ), we associate to each activity a variable Gi that defines its magnitude. The magnitude could be represented either in an absolute way, as the amount of a given activity, or in a relative way, as a percentage with respect to the existing quantity of the same activity. We use in this paper the absolute representation.
As stated above, we distinguish primary from secondary activities: let AP be the set of indexes of primary activities and AS the set of indexes of secondary activities. The distinction is motivated by the fact that some activities are of primary importance in a given plan. Secondary activities are those supporting the primary activities by providing the needed infrastructures. The dependencies between primary and secondary activities are considered by the constraint: ∀j ∈ AS
Gj =
X dij Gi i∈AP Given a budget BP lan available for a given plan, we have a constraint limiting the overall plan cost as follows
Na X Gi ci ≤ BP lan i=1 Na X Gi oi ≥ oP lan.
i=1 Such constraint can be imposed either on the overall plan or on parts of it. For example, if the budget is partitioned into chapters, we can impose constraint (1) on activities of a given chapter.
Moreover, given an expected outcome oP lan of the plan, we have a constraint ensuring to reach the outcome:
For example, in an energy plan the outcome can be to have more energy available in the region, so oP lan could be the increased availability of electrical power (e.g., in kilo-TOE, Tonnes of Oil Equivalent). In such a case, oi will be the production in kTOE for each unit of activity ai.
Concerning the impacts, we sum up the contributions of all the activities and obtain the estimate of the impact on each environmental pressure: ∀j ∈ {1, . . . , Np}
Na pj = X mij Gi.
i=1
(1)
(2)
(3)
The qualitative values in the matrices have been converted into quantitative
values mij in the [
In the same way, given the vector of environmental pressures P = (p1, . . . , pNp ), one can estimate their influence on the environmental receptor ri by means of the matrix N , that relates pressures with receptors: ∀j ∈ {1, . . . , Nr} rj =
Np X nij pi. i=1 Moreover we can have constraints on receptors and pressures. For example, “Greenhouse gas emission” (that is a negative pressure) should not exceed a given threshold.
Concerning objectives, there are a number of possibilities suggested by planning experts. From an economic perspective, one can decide to minimize the overall cost of the plan (that is anyway subject to budget constraints). Clearly in this case, the most economic energy sources are preferred, despite their potentially negative environmental effects (which could be anyway constrained). On the other hand, one could maintain a fixed budget and maximize the produced energy. In this case, the most efficient energy sources will be pushed forward. Or the planner could prefer a green plan and optimize environmental receptors. For example, one can maximize, say, the air quality, or the quality of the surface water. In this case, the produced plan decisions are less intuitive and the system we propose is particularly useful. The link between decisions on primary and secondary activities and consequences on the environment are extremely complex to be manually considered. Clearly, more complex objectives can be pursued, by properly combining the above mentioned aspects. 3.1
We can now describe how to cast the general model for regional planning described above into the model for designing a regional energy plan. The first step is to identify primary and secondary activities. In the context of a regional energy plan, the environmental and planning experts defined the following distinction. Primary activities are those capable of producing energy, namely renewable and non-renewable power plants. Secondary activities are those supporting the energy production, such as activities for energy transportations (e.g., power lines), and infrastructures supporting the primary activities (e.g., dams, yards).
One important aspect to be taken into account when designing a regional energy plan is the energy source diversification: this means that funds should not be directed toward a single energy source, but should cover both renewable and non renewable energy sources. This requirement comes from fluctuations of the price and availability of the various resources. For this reason, we have constraints on the minimal fraction Fi of the total energy produced by each source i: ∀i ∈ AP
Gioi ≥ FiT o where the total outcome T o is simply obtained as
T o =
X Gj oj .
j∈AP In addition, each region has its own geo-physical characteristics. For instance, some regions are particularly windy, while some others are not. Hydroelectric power plants can be built with a very careful consideration of environmental impacts, the most obvious being the flooding of vast areas of land. This poses constraints on the maximum energy Ui that can be produced by a given energy source i ∀i ∈ AP
Gioi ≤ Ui.
Finally, the region priorities should be compliant with European guidelines, such as the 20-20-20 initiative, that aims at achieving three ambitious targets by 2020: reducing by 20% greenhouse gas emissions, having a 20% share of the final energy consumption produced by renewable sources, and improving by 20% its energy efficiency. For this reason, we can impose constraints on the minimum amount of energy Lren produced by renewable energy sources whose set of activities is referred to as AP ren. The constraint that we can impose is
X i∈AP ren
Gioi ≥ Lren. 4
The constraint-based model described in previous sections has been used in the
planning of the regional energy plan for 2011-2013. The system is implemented
in the Constraint Logic Programming language ECLiPSe [
The regional energy plan had the objective of paving the way to reach the ambitious goal of the 20-20-20 directive, in particular having 20% of energy in 2020 produced by renewable sources. This amount does not consider only electric power, but the whole energy balance in the region, including thermal energy, and transports.
Transports can use renewable energy by using renewable fuels, like biogas (methane produced from the fermentation of vegetable or animal wastes) or oil produced from various types of crops. Currently, we do not consider this issue.
Thermal energy can be used e.g. for home heating; renewable sources in this case are thermal solar panels (that produce hot water for domestic use), geothermal pumps (that are used to heat or to refresh houses), biomass plants, that produce hot water used to heat neighboring houses during winter.
The considered electric power plants that produce energy from renewable sources are hydroelectric plants, photovoltaic plants, thermodynamic solar plants, wind generators and, again, biomass power plants.
For each energy source, the plan should provide: the installed power, in MW; the total energy produced in a year, in kTOE (TOE stands for Tonne of Oil Equivalent); the total cost, in Me. The ratio between installed power and total produced energy is mainly influenced by the availability of the source: while a biomass plant can (at least in theory) produce energy 24/7, the sun is available only during the day, and the wind only occasionally. For unreliable sources an average for the whole year is taken.
The cost of the plant, instead, depends mainly on the installed power: a solar plant has an installation cost that depends on the square meters of installed panels, which on their turn can provide some maximum power (peak power).
It is worth noting that the considered cost is the total cost of the plant for
the regional system, which is not the same as the cost for the taxpayers of the
Emilia-Romagna region. In fact, the region can enforce policies in many ways,
convincing private stakeholders to invest in power production. This can be done
with financial leverage, or by giving favorable conditions (either economic or
other) to investors. Some power sources are economically profitable, so there is
no need for the region to give subsidies. For example, currently in Italy biomasses
are economically advantageous for investors, so privates are proposing projects to
build biomasses plants. On the other hand, biomasses also produce pollutants,
they are not always sustainable (see [
Technicians in the region estimated (considering current energy requirements, growth trends, foreseen energy savings) the total energy requirements for 2020; out of this, 20% should be provided by renewable sources. They also proposed for this amount a percentage to be provided during the plan 2011-2013: about 177kTOE of electrical energy and 296kTOE of thermal energy.
Starting from these data, they developed a plan for electrical energy and one for thermal energy.
We used the model presented in Section 3 considering initially only “extreme” cases, in which only one type of energy source is used. The application provides the optimal plan, together with its environmental assessment, namely an evaluation of the environmental receptors used by the environmental protection agency.
In order to understand the individual contributions of the various energy forms, we plotted all the plans that use a single type of energy in Figure 1, together with the plan developed by the region’s experts (we consider here electrical energy sources). On the x-axis, we chose the receptor Air quality because it is probably the most sensitive receptor in the Emilia-Romagna region. On the y-axis we plotted the cost of the plan. As explained previously, all plans provide the same energy in kTOE, while they can require different installation power (in MW).
First of all, we notice that some of the energy types improve the air quality (positive values on the x-axis), while others worsen it (negative values). Of course, no power plant can improve the air quality by itself (as it cannot remove pollutants from the air). The point is that the plant provides electrical energy without introducing new pollutants; if such energy would not have been provided to the electrical network, it would have been imported from neighboring regions. In such a case, the required energy would be produced with the same mixture of energy sources as in the national production, including those emitting pollutants, so the net contribution is positive for the air quality. Note also that the different energy sources have different impacts on the air quality not only due to the emissions of the power plants, but also to the impact of the secondary activities required by the various sources.
Finally, the “extreme” plans are usually not feasible, in the sense that the constraint on the real availability of the energy source in the region was relaxed. For example, wind turbines provide a very good air quality at a low cost, but the amount required in the corresponding extreme plan is not possible in the region considering the average availability of wind and of land for installing turbines.
The plan proposed by the region’s experts is more balanced: it considers the real availability of the energy source in the region, and provides a mixture of all the different renewable types of energy. This is very important in particular for renewable sources, that are often discontinuous: wind power is only available when the wind is blowing at a sufficient speed, solar power is only available during sunny days, etc., so having a mixture of different sources can provide an energy availability more continuous during the day.
Beside assessing the plan proposed by the experts, we also provided new, alternative plans. In particular, we searched for optimal plans, both with respect to the cost, and to the air quality. Since we have two objective functions, we plotted the Pareto-optimal frontier: each point of the frontier is a point such that one cannot improve one of the objectives without sacrificing the other. In our case, the air quality cannot be improved without raising the cost, and, vice-versa, it is impossible to reduce the cost without sacrificing the air quality. The Pareto frontier is shown in Figure 2, together with the experts’ plan. The objective function is a weighted sum of single criteria so our formulation of the problem is linear and we can compute the Pareto frontier by changing coefficients in the weighted sum.
The picture shows that, although the plan devised by the experts is close to the frontier, it can be improved. In particular, we identified on the frontier two solutions that dominate the experts’ plan: one has the same cost, but better air quality, while the other has same air quality, but a lower cost.
Table 1 contains the plan developed by the region’s experts, while Table 2 shows the plan on the Pareto curve that has the same air quality as the plan of the experts. The energy produced by wind generators is almost doubled (as they provide a very convenient ratio (air quality)/cost, see Figure 1), we have a slight increase in the cheap biomass energy, while the other energy sources reduce accordingly. Extracting plans from the solution of the CLP model is trivial: we simply extract the values assigned to decision variables by the linear solver.
Concerning the environmental assessment, we plot in Figure 3 the value of the receptors in significant points of the Pareto front. Each bar represents a single environmental receptor for a specific plan on the Pareto frontier of Figure 2. In this way it is easy to compare how receptors are impacted by different plans. In the Figure, the white bar is associated to the plan, on the frontier, that has the highest air quality, while bars with dark colors are associated to plans that have a low cost (and, thus, a low quality of the air). Notice that the receptors have different trends: some of them improve as we move in the frontier towards higher air quality (like climate quality, mankind wellness, value of material goods), while others improve when moving to less expensive solutions (like quality of sensitive landscapes, wellness of wildlife, soil quality). This is due to several reasons, depending both on the type of power plants installed and on the secondary activities. The application (including both the assessment and the planning) was developed in few person-months by a CLP expert. It does not have a graphical user interface yet, and it is currently usable only by CLP experts; however it produces spreadsheet files with tables having the same structure as those used for years by the region’s experts, so the output is easily understandable by the end user. We are currently developing a web-based application, to let users input the relevant data, and try themselves producing plans on-the-fly.
The assessment module [
One of the results is the ability to generate easily alternative plans with their assessment; this is required by the EU regulations, but it is widely disregarded.
Another result is the possibility to provide plans that are optimal; the optimization criteria can include the cost, or one of the various environmental receptors. The user can select two objectives, and in this case the application generates a Pareto front. This helps the experts or the regional council in doing choices that are more grounded.
We still do not know which plan the regional council will choose, neither we know if and how the directives given in the regional plan will be indeed implemented. More refined plans (at the province or municipality level) should follow the guidelines in the regional plan, but it is also possible to introduce modifications during the plan execution. However, in a perfect world, in which everything is implemented as expected, the added value of CLP in monetary terms could be the difference of the investment columns in the plans in Tables 1 and 2: 191Me saved (by the various actors, public and private, in the whole region) in three years.
Finally, the choice of Constraint Programming greatly enables model flexibility. Discussing with experts, it is often the case that they change their mind on some model constraints or on objectives. Therefore, the flexibility in dealing with side constraints and in dealing with non linear constraints facilitates knowledge acquisition making Constraint Programming the technique of choice for the problem and its future extensions. 6
Global public policy making is a complex process that is influenced by many factors. We believe that the use of constraint reasoning techniques could greatly increase the effectiveness of the process, by enabling the policy maker to analyze various aspects and to play with parameters so as to obtain alternative solutions along with their environmental assessment. Given the amount of financial, human and environmental resources that are involved in regional plans, even a small improvement can have a huge effect.
Important features of the system are: its efficiency, as a plan is returned in few milliseconds, its wide applicability to many regional plans, to provincial and urban plans and also to private and public projects. The system was used for the environmental assessment of the regional energy plan of the Emilia-Romagna region of Italy. Beside performing automatically the assessment (that was performed by hand in previous years), the assessment for the first time includes the evaluation of alternative plans: this is a requirement of EU regulations that is largely disregarded in practice. Moreover, the alternative plans were produced by optimizing the quality of the environmental receptors, together with the cost for the community of the plan itself.
This work is a first step towards a system that fully supports the decision
maker in designing public policies. To achieve this objective, the method must be
extended to take into account the individual level, by investigating the effect of
a policy over the parties affected by it. This can be achieved by integrating
constraint reasoning with simulation models that reproduce the interactions among
the parties. In our current research, we are studying how the region can choose
the form of incentives and calibrate them in order to push the energy market to
invest in the directions foreseen by the Regional Energy plan [
In turn these models can be enriched by adopting e-Participation tools that allow citizens and stakeholders to voice their concerns regarding policy decisions. To fully leverage e-Participation tools, the system must also be able to extract information from all the available data, including natural language. Thus, opinion mining techniques will be useful in this context.
At the moment the system can only be used by IT expert people. In order to turn it into a practical tool that is routinely used by decision makers, we must equip it with a user-friendly interface. In particular, we are in the process of developing a web interface to the constraint solver, in order to make it easy to use and widely accessible.
Finally, economic indicators will be used to assess the economic aspect of the plan. Up to now, only budget and few economic pressures and receptors are considered. We believe that a comprehensive system should fully incorporate this aspect. We will integrate a well established approach (UN and Eurostat Guidelines) and robust data from official statistics into the system to combine economic accounts (measured in monetary terms) and environmental accounts (measured in physical units) into a single framework useful for the evaluation of the integrated economic, environmental-social performance of regions. If the relations with economic indicators are linear, then the solution time should not increase significantly. 7
This work was partially supported by EU project ePolicy, FP7-ICT-2011-7, grant agreement 288147. Possible inaccuracies of information are under the responsibility of the project team. The text reflects solely the views of its authors. The European Commission is not liable for any use that may be made of the information contained in this paper.