Two-Stage DEA-AHP Approach for Solar PV Power Plant Site Selection in Taiwan

1. Introduction

This paper addresses the critical challenge of selecting optimal sites for Solar Photovoltaic (PV) power plants in Taiwan. The urgency is driven by the global need to transition from fossil fuels to renewable energy, a shift amplified by the Covid-19 pandemic and climate change imperatives. Taiwan, heavily reliant on imported fossil fuels and located in a seismically active zone, views solar energy development as pivotal for energy security and economic sustainability.

1.1 Global Renewable Energy Situation

The paper contextualizes the study within global efforts like the Paris Agreement and the European Green Deal, aiming for net-zero emissions. It highlights the resilience of renewable energy during the Covid-19 crisis, with electricity generation from renewables increasing by 5% in 2020 despite disruptions.

1.2 The Potential of Solar Energy

Solar energy is identified as the most suitable renewable source for Taiwan due to its geographical and climatic conditions. However, land constraints, policy challenges, and scale issues hinder development, making systematic site selection essential.

2. Methodology: Two-Stage MCDM Framework

The core contribution is a novel two-stage Multiple Criteria Decision Making (MCDM) approach combining Data Envelopment Analysis (DEA) and the Analytic Hierarchy Process (AHP).

2.1 Stage 1: Data Envelopment Analysis (DEA)

DEA is used as an initial filter to evaluate the natural resource efficiency of 20 potential cities/counties. It treats locations as Decision Making Units (DMUs).

• Inputs: Temperature, Wind Speed, Humidity, Precipitation, Air Pressure.
• Outputs: Sunshine Hours, Insolation.

Locations achieving a perfect efficiency score of 1.0 proceed to the next stage.

2.2 Stage 2: Analytic Hierarchy Process (AHP)

AHP is employed to rank the efficient locations from Stage 1 based on a broader set of socio-techno-economic-environmental criteria. It involves pairwise comparisons to derive criterion weights and final location scores.

2.3 Criteria and Sub-Criteria Hierarchy

The AHP model is structured with five main criteria and 15 sub-criteria:

1. Site Characteristics: Land slope, Land use type, Distance to grid.
2. Technical: Solar radiation, Sunshine hours, Temperature.
3. Economic: Investment cost, Operation & Maintenance cost, Electricity transmission cost, Support mechanisms (e.g., feed-in tariffs).
4. Social: Public acceptance, Job creation, Electricity consumption demand.
5. Environmental: Carbon emission reduction, Ecological impact.

3. Case Study: Taiwan

3.1 Data Collection & Potential Sites

The study evaluated 20 major cities and counties across Taiwan. Meteorological data (inputs/outputs for DEA) and socio-economic data (for AHP) were collected from official Taiwanese sources like the Central Weather Bureau and the Ministry of Economic Affairs.

3.2 DEA Efficiency Analysis Results

The DEA model filtered out locations with sub-optimal natural resource efficiency. Only cities/counties that efficiently converted climatic inputs (like moderate temperature and low humidity) into solar energy outputs (high sunshine and insolation) received a score of 1.0. This step reduced the candidate pool for the more detailed AHP analysis.

3.3 AHP Weighting & Final Ranking

The AHP pairwise comparison revealed the relative importance of criteria. The top three most influential sub-criteria were:

This underscores that policy and economic factors (support, cost) and local demand are more decisive than pure solar resource potential in the final ranking.

4. Results & Discussion

4.1 Key Findings

The hybrid DEA-AHP approach successfully identified and prioritized sites. The two-stage process's strength lies in first ensuring natural resource viability (DEA) before evaluating broader feasibility (AHP), preventing resource-rich but otherwise infeasible locations from ranking highly.

4.2 Top-Ranked Locations

The final AHP ranking identified the top three most suitable locations for large-scale solar PV farm development in Taiwan:

1. Tainan City
2. Changhua County
3. Kaohsiung City

These areas combine strong solar resources with favorable economic conditions (e.g., existing support mechanisms), lower relative transmission costs, and high local electricity demand.

5. Technical Details & Mathematical Formulation

DEA Formulation (CCR Model): The efficiency score $\theta_k$ for DMU $k$ is obtained by solving the linear program: $$\text{Max } \theta_k = \sum_{r=1}^{s} u_r y_{rk}$$ $$\text{subject to: } \sum_{i=1}^{m} v_i x_{ik} = 1$$ $$\sum_{r=1}^{s} u_r y_{rj} - \sum_{i=1}^{m} v_i x_{ij} \leq 0, \quad j=1,...,n$$ $$u_r, v_i \geq \epsilon > 0$$ where $x_{ij}$ are inputs, $y_{rj}$ are outputs, $v_i$ and $u_r$ are weights, and $\epsilon$ is a non-Archimedean infinitesimal.

AHP Consistency Check: A critical step is ensuring the pairwise comparison matrix $A$ is consistent. The Consistency Index ($CI$) and Consistency Ratio ($CR$) are calculated: $$CI = \frac{\lambda_{max} - n}{n-1}$$ $$CR = \frac{CI}{RI}$$ where $\lambda_{max}$ is the principal eigenvalue, $n$ is matrix size, and $RI$ is the Random Index. A $CR < 0.1$ is acceptable.

6. Analysis Framework: Example Case

Scenario: Evaluating two candidate sites, "City A" and "County B," after DEA pre-filtering.

Step 1 - Criterion Weighting (AHP): Experts perform pairwise comparisons. For example, comparing "Economic" vs. "Environmental" impact might yield a score of 3 (moderate importance of Economic over Environmental). This populates the comparison matrix to derive global weights (e.g., Economic: 0.35, Environmental: 0.10).

Step 2 - Site Scoring per Criterion: Rate each site against each sub-criterion on a scale (e.g., 1-9). For "Support Mechanisms," if City A has excellent feed-in tariffs (score=9) and County B has poor support (score=3), these are normalized.

Step 3 - Synthesis: Final score for City A = $\sum (\text{Sub-criterion Weight} \times \text{City A's Normalized Score})$. The site with the higher aggregate score is preferred.

This structured, quantitative framework replaces ad-hoc decision-making with transparency and traceability.

7. Application Outlook & Future Directions

• Integration with GIS: Future work should integrate this MCDM approach with Geographic Information Systems (GIS) for spatial visualization and analysis of land suitability, creating powerful decision-support tools.
• Dynamic & Probabilistic Models: Incorporating time-series data and probabilistic forecasts for climate variables and electricity prices can make the model adaptive to future changes.
• Hybrid with other MCDM methods: Combining AHP with techniques like TOPSIS or VIKOR could handle uncertainty or conflicting criteria more robustly.
• Broader Application: This two-stage framework is highly transferable to other renewable energy site selection problems (e.g., wind, geothermal) in different geographical contexts.
• Lifecycle Sustainability Integration: Expanding the environmental criterion to a full Life Cycle Assessment (LCA) would evaluate the carbon footprint of manufacturing and decommissioning PV panels.

8. References

1. Intergovernmental Panel on Climate Change (IPCC). (2021). Climate Change 2021: The Physical Science Basis. Cambridge University Press.
2. United Nations. (2015). Paris Agreement. United Nations Treaty Collection.
3. European Commission. (2019). The European Green Deal. COM(2019) 640 final.
4. International Energy Agency (IEA). (2020). World Energy Outlook 2020. OECD/IEA.
5. International Renewable Energy Agency (IRENA). (2021). Renewable Energy and Jobs – Annual Review 2021.
6. Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the efficiency of decision making units. European Journal of Operational Research, 2(6), 429-444.
7. Saaty, T. L. (1980). The Analytic Hierarchy Process. McGraw-Hill.
8. Wang, C. N., Nguyen, N. A. T., Dang, T. T., & Bayer, J. (2021). A Two-Stage Multiple Criteria Decision Making for Site Selection of Solar Photovoltaic (PV) Power Plant: A Case Study in Taiwan. IEEE Access, 9, 75509-75522. DOI: 10.1109/ACCESS.2021.3081995.

9. Expert Analysis & Critical Review

Core Insight: This paper isn't just another site selection study; it's a pragmatic blueprint for de-risking renewable energy infrastructure investment. The real insight is the sequential logic: use DEA to ruthlessly filter for natural resource efficiency first—a non-negotiable, physics-based gate—before letting the softer, policy-heavy AHP criteria determine the winner. This prevents the common pitfall of choosing a site that's politically convenient but climatically mediocre.

Logical Flow: The methodology's elegance is in its division of labor. DEA handles the "can it work here?" question based on sun, wind, and rain. AHP tackles the "should we build it here?" question based on cost, policy, and social impact. This mirrors the real-world decision process of developers and governments, moving from technical potential to project feasibility. The high weight given to "Support Mechanisms" (0.332) is a brutally honest reflection of reality: a good feed-in tariff can outweigh several percentage points of higher solar irradiance.

Strengths & Flaws: The major strength is the hybrid approach's robustness and its validation in a complex, real-world context (Taiwan). Using established, widely understood tools (DEA, AHP) enhances replicability. However, the model has notable gaps. First, it's static; it doesn't account for the temporal variability of solar resources or future climate change impacts, a critical consideration highlighted by the IPCC's latest reports. Second, the AHP's reliance on expert pairwise comparisons, while standard, introduces subjectivity. The paper would be stronger if it supplemented this with sensitivity analysis or used a fuzzy-AHP approach to handle uncertainty, as seen in advanced applications like those discussed on the RAND Corporation's methodology pages. Third, land availability and cost—often the ultimate bottleneck—seem buried within sub-criteria. In many markets, this is the primary constraint.

Actionable Insights: For policymakers in Taiwan and similar regions, the top-ranked list (Tainan, Changhua, Kaohsiung) provides a data-driven starting point for concentrating infrastructure and incentives. For developers, the framework is a ready-made due diligence checklist. The immediate next step should be to integrate this model with high-resolution GIS data to move from city-level to parcel-level analysis. Furthermore, comparing this DEA-AHP result with outcomes from machine learning-based site suitability models—like those increasingly used in wind farm planning—would be a valuable research direction to test the convergence (or divergence) of different paradigms. Ultimately, this work provides a solid, operational foundation. The future lies in making it dynamic, spatially explicit, and capable of ingesting real-time data streams.