feat(WastewaterTreatment): add BSM1 aeration control benchmark

TASK_DETAILS.md

@@ -351,5 +351,10 @@ We welcome new engineering problem ideas — even without complete verification
       <td><code>DiffSimThermalControl</code></td>
       <td>Process optimization in additive manufacturing via differentiable simulation</td>
     </tr>
+    <tr>
+      <td><b>WastewaterTreatment</b></td>
+      <td><code>BSM1AerationControl</code></td>
+      <td>Feedback control of activated-sludge aeration and internal recycle across dry, rain, and storm operation</td>
+    </tr>
   </tbody>
 </table>

TASK_DETAILS_zh-CN.md

@@ -351,5 +351,10 @@ Frontier-Eng 目前已覆盖以下领域的任务。每个任务均配有可运
       <td><code>DiffSimThermalControl</code></td>
       <td>基于可微仿真的增材制造工艺优化</td>
     </tr>
+    <tr>
+      <td><b>WastewaterTreatment</b></td>
+      <td><code>BSM1AerationControl</code></td>
+      <td>在旱天、降雨和暴雨工况下反馈控制活性污泥曝气与内回流</td>
+    </tr>
   </tbody>
 </table>

benchmarks/WastewaterTreatment/BSM1AerationControl/README.md

@@ -0,0 +1,53 @@
+# BSM1 Aeration Control
+
+Design a deterministic feedback controller for an activated-sludge plant derived from the
+IWA Benchmark Simulation Model No. 1 (BSM1). The controller sets oxygen-transfer coefficients
+in the three aerobic reactors and the internal recycle flow. It must balance effluent quality,
+energy, compliance, and actuator smoothness across dry, rain, and storm scenarios.
+
+Edit only the EVOLVE-BLOCK in `scripts/init.py`, preserving:
+
+```python
+def reset_controller(scenario: dict) -> None: ...
+def control(observation: dict) -> dict: ...
+```
+
+## Setup
+
+The task is offline and CPU-only:
+
+```bash
+python -m pip install -r verification/requirements.txt
+```
+
+No raw IWA influent files are redistributed. The evaluator generates deterministic trajectories
+from published BSM1 averages and weather-event descriptions. A complete baseline run takes about
+45 seconds on a laptop.
+
+## Direct evaluation
+
+```bash
+python verification/evaluator.py scripts/init.py --metrics-out metrics.json --artifacts-out artifacts.json
+```
+
+## Regression tests
+
+```bash
+python -m unittest discover -s verification -p "test_*.py" -v
+```
+
+## Unified evaluation
+
+From the repository root:
+
+```bash
+python -m frontier_eval task=unified task.benchmark=WastewaterTreatment/BSM1AerationControl algorithm=openevolve algorithm.iterations=0
+```
+
+The ranking metric is `combined_score` (higher is better). `metrics.json` contains the aggregate
+score and validity fields; `artifacts.json` contains scenario-level engineering metrics and daily
+effluent samples. Candidate code runs in a separate, bounded JSON-lines worker process. This is
+process isolation, not an operating-system security sandbox.
+
+See `Task.md` for the exact interface and scoring model, and
+`references/design_notes.md` for provenance, validation, and modelling limitations.

benchmarks/WastewaterTreatment/BSM1AerationControl/README_zh-CN.md

@@ -0,0 +1,44 @@
+# BSM1 曝气控制
+
+为基于 IWA 一号基准仿真模型（BSM1）的活性污泥处理厂设计确定性反馈控制器。控制器设置三个好氧反应池的氧传质系数与内回流量，并在旱天、降雨和暴雨场景下权衡出水质量、能耗、达标情况和执行器平滑性。
+
+只修改 `scripts/init.py` 中的 EVOLVE-BLOCK，并保持以下接口：
+
+```python
+def reset_controller(scenario: dict) -> None: ...
+def control(observation: dict) -> dict: ...
+```
+
+## 环境安装
+
+本任务离线运行且只需要 CPU：
+
+```bash
+python -m pip install -r verification/requirements.txt
+```
+
+仓库不再分发 IWA 原始进水文件。评测器依据已发表的 BSM1 平均值与天气事件说明生成确定性轨迹。在普通笔记本上完整基准评测约需 45 秒。
+
+## 直接评测
+
+```bash
+python verification/evaluator.py scripts/init.py --metrics-out metrics.json --artifacts-out artifacts.json
+```
+
+## 回归测试
+
+```bash
+python -m unittest discover -s verification -p "test_*.py" -v
+```
+
+## 统一评测
+
+在仓库根目录运行：
+
+```bash
+python -m frontier_eval task=unified task.benchmark=WastewaterTreatment/BSM1AerationControl algorithm=openevolve algorithm.iterations=0
+```
+
+排名指标为 `combined_score`，越高越好。`metrics.json` 给出聚合分数与有效性；`artifacts.json` 给出逐场景工程指标和每日出水样本。候选代码运行在带时限的独立 JSON-lines 工作进程中；这是进程隔离，不是操作系统级安全沙箱。
+
+完整接口和评分模型见 `Task_zh-CN.md`，来源、验证和模型限制见 `references/design_notes.md`。

benchmarks/WastewaterTreatment/BSM1AerationControl/Task.md

@@ -0,0 +1,92 @@
+# Task: BSM1 Aeration Control
+
+## Engineering setting
+
+Activated-sludge plants must remove carbon and nitrogen while limiting blower and pumping energy.
+Influent flow and composition vary diurnally and become more difficult during rain and storm
+events. This task exposes feedback control rather than a one-shot parameter fit: every 15 minutes,
+the policy observes noisy process measurements and selects aeration and internal recycle settings.
+
+## Process model and scenarios
+
+The evaluator implements the standard BSM1 layout: five completely mixed ASM1 reactors (two
+anoxic, three aerobic) followed by a ten-layer Takacs secondary settler. Fixed return-activated
+sludge and waste-sludge flows are 18,446 and 385 m3/day. The plant is simulated for 14 days and
+the final seven days are scored.
+
+Three deterministic scenarios share the same diurnal base load:
+
+- `dry`: diurnal and weekend load variation;
+- `rain`: a sustained dilution-water event from day 8.35 to day 10.44;
+- `storm`: two shorter hydraulic pulses centred near days 8.87 and 11.18.
+
+The rain and storm series are deterministic IWA-derived engineering trajectories, not byte-for-byte
+copies of the official BSM1 dynamic influent files. Sensor noise uses fixed, scenario-specific seeds.
+
+## Observation
+
+`control(observation)` receives only current or past information:
+
+- `scenario_id`, `weather`, `time_day`, and `step_minutes`;
+- influent flow and ammonium;
+- dissolved oxygen in reactors 3, 4, and 5;
+- nitrate in reactor 2;
+- effluent ammonium and total nitrogen;
+- `previous_action`.
+
+The dissolved-oxygen, nitrate, and ammonium measurements contain deterministic sensor noise. The
+policy does not receive future influent, process state arrays, random seeds, or evaluator internals.
+
+## Action interface and hard constraints
+
+Return exactly these four finite numeric fields:
+
+```python
+{
+    "kla3_per_day": float,                    # [0, 360]
+    "kla4_per_day": float,                    # [0, 360]
+    "kla5_per_day": float,                    # [0, 360]
+    "internal_recycle_m3_per_day": float,     # [0, 92230]
+}
+```
+
+At one 15-minute step, each KLa may change by at most 120/day and internal recycle by at most
+30,000 m3/day. Missing/extra fields, booleans, non-finite values, range errors, slew errors,
+exceptions, protocol errors, or timeouts make the complete candidate invalid with score zero.
+`reset_controller(scenario)` must reset all candidate-owned state before each scenario.
+
+## Metrics and score
+
+For every scored step, the evaluator computes the official BSM1 effluent quality index (EQI),
+aeration energy, pumping energy, mixing energy, actuator switching, and normalized exceedance of
+five standard limits: NH4-N 4, total nitrogen 18, COD 100, TSS 30, and BOD5 10 g/m3.
+
+Each objective is mapped to a dimensionless utility:
+
+```text
+Uq = exp(-EQI / 6000)
+Ua = exp(-aeration_energy / 5000)
+Up = exp(-pumping_energy / 1200)
+Um = exp(-mixing_energy / 600)
+Us = exp(-4 * switching_index)
+
+base = 100 * (0.45 Uq + 0.20 Ua + 0.10 Up + 0.10 Um + 0.15 Us)
+scenario_score = base * exp(-6 * violation_index)
+```
+
+The aggregate rewards typical and worst-case behavior:
+
+```text
+combined_score = 0.75 * mean(scenario_scores) + 0.25 * min(scenario_scores)
+```
+
+The score is absolute and contains no frozen candidate baseline. A fixed published BSM1 operating
+point is supplied only as editable starter code and as a reproducible comparison in the result log.
+
+## Candidate process boundary
+
+The evaluator copies the candidate to a temporary directory and imports it in a persistent worker.
+Communication is JSON-only; import, each call, cumulative response time, and response size are
+bounded. Candidate stdout is discarded and credentials are removed from its environment. This
+prevents ordinary Python-level mutation of the parent evaluator; it is not a sandbox against
+hostile native code or unrestricted filesystem/network access.

benchmarks/WastewaterTreatment/BSM1AerationControl/Task_zh-CN.md

@@ -0,0 +1,58 @@
+# 任务：BSM1 曝气控制
+
+## 工程背景
+
+活性污泥处理厂既要去除有机物和氮，又要控制鼓风与泵送能耗。进水流量和组分具有日周期变化，降雨与暴雨会进一步增加水力冲击。本任务是反馈控制而非一次性参数拟合：策略每 15 分钟读取带噪过程测量，并设置曝气与内回流。
+
+## 工艺模型与场景
+
+评测器实现标准 BSM1 布局：五个完全混合 ASM1 反应池（前两个缺氧、后三个好氧）以及十层 Takacs 二沉池。固定污泥回流和排泥流量分别为 18,446 与 385 m3/day。每个场景仿真 14 天，最后 7 天计分。
+
+三个确定性场景共享同一日周期基础负荷：
+
+- `dry`：日周期和周末负荷变化；
+- `rain`：第 8.35 至 10.44 天的持续稀释水事件；
+- `storm`：中心约在第 8.87 与 11.18 天的两次短时水力脉冲。
+
+降雨与暴雨序列是依据 IWA 资料构造的确定性工程轨迹，并非官方 BSM1 动态进水文件的逐字节副本。传感器噪声使用固定且按场景区分的种子。
+
+## 观测
+
+`control(observation)` 只接收当前或历史信息：场景、天气、时间、步长；进水流量与氨氮；3—5 号池溶解氧；2 号池硝酸盐；出水氨氮与总氮；以及 `previous_action`。溶解氧、硝酸盐和氨氮测量含确定性噪声。策略不能获得未来进水、完整过程状态、随机种子或评测器内部信息。
+
+## 动作接口与硬约束
+
+必须恰好返回四个有限数值字段：
+
+```python
+{
+    "kla3_per_day": float,                    # [0, 360]
+    "kla4_per_day": float,                    # [0, 360]
+    "kla5_per_day": float,                    # [0, 360]
+    "internal_recycle_m3_per_day": float,     # [0, 92230]
+}
+```
+
+相邻 15 分钟内，每个 KLa 的最大变化为 120/day，内回流最大变化为 30,000 m3/day。字段缺失或多余、布尔值、非有限值、越界、变化率超限、异常、协议错误或超时都会使整个候选无效并得零分。`reset_controller(scenario)` 必须在每个场景前重置策略自身状态。
+
+## 指标与评分
+
+计分期的每一步都会计算 BSM1 出水质量指数（EQI）、曝气/泵送/搅拌能耗、执行器切换，以及五项标准限值的归一化超标量：NH4-N 4、总氮 18、COD 100、TSS 30、BOD5 10 g/m3。
+
+```text
+Uq = exp(-EQI / 6000)
+Ua = exp(-曝气能耗 / 5000)
+Up = exp(-泵送能耗 / 1200)
+Um = exp(-搅拌能耗 / 600)
+Us = exp(-4 * switching_index)
+
+base = 100 * (0.45 Uq + 0.20 Ua + 0.10 Up + 0.10 Um + 0.15 Us)
+场景分 = base * exp(-6 * violation_index)
+combined_score = 0.75 * 场景均分 + 0.25 * 最差场景分
+```
+
+该分数是绝对工程分，不依赖冻结的候选基线。公开的 BSM1 固定工况仅作为可编辑起始程序和结果日志中的可复现实验对照。
+
+## 候选进程边界
+
+评测器把候选复制到临时目录并在持久工作进程中导入。通信仅使用 JSON；导入、单次调用、累计响应时间和响应大小均受限。候选标准输出会被丢弃，环境中的凭据会被移除。该边界可阻止普通 Python 代码修改父评测器，但不能替代针对恶意原生代码、文件系统或网络访问的操作系统沙箱。

benchmarks/WastewaterTreatment/BSM1AerationControl/baseline/result_log.txt

@@ -0,0 +1,24 @@
+Direct command:
+python verification/evaluator.py scripts/init.py --metrics-out metrics.json --artifacts-out artifacts.json
+
+Environment:
+Windows 11, Python 3.12.13, CPU-only
+
+Result:
+=== BSM1 Aeration Control ===
+scenario=dry score=53.457 eqi=5488.86 ae=3341.39 violations=0.011330
+scenario=rain score=44.609 eqi=7535.45 ae=3341.39 violations=0.025575
+scenario=storm score=48.513 eqi=6780.93 ae=3341.39 violations=0.017011
+---
+completed_scenarios: 3/3
+diagnostic_score: 47.7973
+combined_score: 47.7973
+
+Regression:
+python -m unittest discover -s verification -p "test_*.py" -v
+Ran 12 tests in 3.266s
+OK
+
+Unified evaluation:
+python -m frontier_eval task=unified task.benchmark=WastewaterTreatment/BSM1AerationControl algorithm=openevolve algorithm.iterations=0
+combined_score=47.7973, valid=1.0000, completed_scenarios=3.0000, benchmark_returncode=0.0000

benchmarks/WastewaterTreatment/BSM1AerationControl/baseline/solution.py

@@ -0,0 +1,16 @@
+"""Reference fixed-operating-point policy for BSM1AerationControl."""
+
+from __future__ import annotations
+
+
+def reset_controller(scenario: dict) -> None:
+    pass
+
+
+def control(observation: dict) -> dict:
+    return {
+        "kla3_per_day": 240.0,
+        "kla4_per_day": 240.0,
+        "kla5_per_day": 84.0,
+        "internal_recycle_m3_per_day": 55338.0,
+    }

benchmarks/WastewaterTreatment/BSM1AerationControl/frontier_eval/agent_files.txt

@@ -0,0 +1,7 @@
+README.md
+README_zh-CN.md
+Task.md
+Task_zh-CN.md
+scripts/init.py
+references/design_notes.md
+frontier_eval/constraints.txt

benchmarks/WastewaterTreatment/BSM1AerationControl/frontier_eval/artifact_files.txt

@@ -0,0 +1,2 @@
+metrics.json
+artifacts.json

benchmarks/WastewaterTreatment/BSM1AerationControl/frontier_eval/candidate_destination.txt

@@ -0,0 +1 @@
+scripts/init.py

benchmarks/WastewaterTreatment/BSM1AerationControl/frontier_eval/constraints.txt

@@ -0,0 +1,9 @@
+BSM1 aeration-control constraints:
+1) Edit only `scripts/init.py` inside the EVOLVE-BLOCK markers.
+2) Keep `reset_controller(scenario) -> None` and `control(observation) -> dict` working.
+3) Return exactly the four finite numeric fields documented in Task.md.
+4) Keep each KLa in [0, 360]/day and internal recycle in [0, 92230] m3/day.
+5) Respect per-step slew limits: 120/day for each KLa and 30000 m3/day for recycle.
+6) Reset all candidate-owned state before each dry, rain, or storm scenario.
+7) Keep the controller deterministic, self-contained, and within the 0.10-second call limit.
+8) Do not read, write, import, or modify evaluator, model, result, or environment-secret files.

benchmarks/WastewaterTreatment/BSM1AerationControl/frontier_eval/copy_files.txt

@@ -0,0 +1,12 @@
+README.md
+README_zh-CN.md
+Task.md
+Task_zh-CN.md
+scripts/init.py
+verification/evaluator.py
+verification/bsm1_model.py
+verification/policy_runtime.py
+verification/policy_worker.py
+verification/requirements.txt
+references/config.json
+references/BSD-3-Clause-bsm2-python.txt

benchmarks/WastewaterTreatment/BSM1AerationControl/frontier_eval/eval_command.txt

@@ -0,0 +1 @@
+{python} verification/evaluator.py {candidate} --metrics-out metrics.json --artifacts-out artifacts.json

benchmarks/WastewaterTreatment/BSM1AerationControl/frontier_eval/initial_program.txt

@@ -0,0 +1 @@
+scripts/init.py

benchmarks/WastewaterTreatment/BSM1AerationControl/frontier_eval/readonly_files.txt

@@ -0,0 +1,15 @@
+README.md
+README_zh-CN.md
+Task.md
+Task_zh-CN.md
+verification/evaluator.py
+verification/bsm1_model.py
+verification/policy_runtime.py
+verification/policy_worker.py
+verification/test_bsm1_model.py
+verification/test_evaluator.py
+verification/test_policy_runtime.py
+verification/requirements.txt
+references/config.json
+references/design_notes.md
+references/BSD-3-Clause-bsm2-python.txt

benchmarks/WastewaterTreatment/BSM1AerationControl/references/BSD-3-Clause-bsm2-python.txt

@@ -0,0 +1,28 @@
+BSD 3-Clause License
+
+Copyright (c) 2025, FAU-EVT
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+3. Neither the name of the copyright holder nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.