Microgrid Design

Overview

Optimizes the sizing and dispatch of distributed energy resources for grid-connected and islanded microgrids. The skill follows an enumerate, simulate, rank methodology: it builds a feasible set of system configurations (solar photovoltaic capacity, battery storage, diesel genset, grid connection), simulates 8,760-hour dispatch for each, screens against renewable-fraction and reliability constraints, and ranks survivors by net present cost. It then evaluates islanding resilience, computes bankable financial metrics, and produces feasibility deliverables (single-line diagram, dispatch charts, equipment schedule, and a written report). Use it for feasibility studies, interconnection applications, and investment decisions.

Workflow

<Workflow - Microgrid Design and Feasibility description="Size and dispatch a microgrid, evaluate resilience and financials, and produce feasibility deliverables." tools=[get_current_time, web_search, url_fetch, file_read, run_python, run_python_with_write, open_in_session_tab] triggers=["User asks to design or size a microgrid", "User asks to optimize solar plus storage or a hybrid energy system", "User asks for a microgrid feasibility study, LCOE, or resilience analysis"] preferred_model=smart preferred_thinking=high

1. [Agent] Establish the current date with get_current_time, then verify reference data. List the time-sensitive values the study needs (fuel price, tariffs and demand charges, export compensation, tax credits and depreciation, LCOE benchmarks, permitting limits). For each, fetch the current value from a source in references/data-sources.md using web_search or url_fetch. Validate: Every time-sensitive value has a source verified this session or supplied by the user. If fails: Per Rule 2, stop and ask the user to provide or confirm the missing value before continuing.

2. [Agent] Ingest and characterize the load profile. If {{load_data_path}} is given, read it with file_read and confirm it covers at least 8,760 hours. Separate critical load from total load. Compute annual energy, peak demand, load factor, and daily and seasonal patterns. Validate: An 8,760-hour load series and a critical-load series exist with plausible annual energy and peak. If fails: If data is missing or shorter than a year, ask for at least monthly consumption plus building type, area, and climate zone, and build a synthetic hourly profile; flag the added uncertainty.

3. [Agent] Assess the solar resource and build a PV production profile. Use {{latitude}} and {{longitude}} to fetch site solar-resource data from a verified service in references/data-sources.md, or implement the geometry in references/component-models.md with numpy. Produce an 8,760-hour AC profile per installed kW, apply year-1 losses, and compute the capacity factor. Validate: An 8,760-hour per-kW AC profile exists and its capacity factor is within a plausible range for the site. If fails: If site data is unavailable, use a regional average irradiance, mark the result as uncertain, and tell the user.

4. [Agent] Define the search space and enumerate candidates per references/methodology.md. Set PV, battery, and genset ranges from the bounds there, decide grid options, pre-screen obvious constraint violations, and target 50 to 200 candidates. Validate: A feasible candidate set of roughly 50 to 200 configurations exists after pre-screening. If fails: Coarsen the increments for a first pass, then refine around the emerging optimum.

5. [Agent] Simulate 8,760-hour dispatch for each candidate with the chosen strategy (load-following or cycle-charging) per references/methodology.md and the component models in references/component-models.md. Track hourly PV, battery state of charge, genset runtime and fuel, grid flows, and unserved load. Compute annual fuel, throughput, renewable fraction, and reliability. Discard configurations that violate hard constraints. Checkpoint partial results to {{output_dir}} with run_python_with_write to stay within the execution cap. Validate: Every surviving candidate has annual metrics and no hard-constraint violations, and results are checkpointed. If fails: Reduce the candidate count or use simplified dispatch for screening, then re-simulate the shortlist at full fidelity.

6. [Agent] Evaluate islanding performance per references/metrics.md. Simulate design-duration outages from worst-case conditions (winter evening, battery at minimum reserve, peak critical load), compute hours of autonomy, and run the Monte Carlo LOLE if a reliability target is set. Verify critical load is served throughout the design outage. Validate: Each shortlisted configuration has hours of autonomy and, if required, an LOLE that meets the target. If fails: Increase battery or genset size to meet the resilience target and report the cost premium of resilience.

7. [Agent] Compute financial metrics for the configurations that pass all constraints using the verified inputs from step 1 and the formulas in references/metrics.md. Rank by NPC, then report LCOE, simple payback versus a grid-only baseline, and internal rate of return. Apply verified incentives and run a sensitivity analysis on fuel price, solar cost, and battery cost. Validate: The top 3 to 5 configurations have NPC, LCOE, payback, and a sensitivity range, all built on verified inputs. If fails: Report a partial financial model with every assumption stated explicitly and flag which inputs are unverified.

8. [Agent] Produce system-architecture deliverables per references/architecture.md: a single-line diagram, an equipment schedule (use canvas_xlsx), dispatch charts for a sunny week, a cloudy week, and an outage event, an annual energy-flow Sankey, and a monthly generation breakdown (use highcharts with html_design). Write all files to {{output_dir}} or the location confirmed with the user. Validate: Each deliverable file is written to the target location. If fails: If chart rendering is unavailable, deliver a text-based architecture description with the key specifications rather than fabricating an image.

9. [Agent] Compile the feasibility report with canvas_docx: executive summary (recommended configuration, LCOE, NPC, renewable fraction, resilience hours), technical design, financial pro-forma with the sensitivity results, an implementation roadmap, and a risk register. Include the Rule 1 liability disclaimer. Open the report and the charts with open_in_session_tab. Validate: The report exists at the target location, cites its verified sources, carries the disclaimer, and is opened for review. If fails: Deliver the completed sections with clearly marked placeholders for anything missing, and tell the user what remains.

</Workflow - Microgrid Design and Feasibility>