【钠瞳 Sodium Pupil】五：我们证明了什么，没证明什么
YeE · 发布于 2026/10/19 · 更新于 2026/10/19
YeE2026/10/19
五、我们证明了什么，没证明什么
| 结论 | ✅ 本模型可以支持 | ❌ 本模型不能支持 |
|---|---|---|
| 能量预算是否成立 | 三种策略在能量守恒约束下的差异稳定存在 | 真实光伏与积雪分布的统计特征 |
| 策略差异是否可信 | 同输入对照，且参数扰动下不翻转 | 真实场景中的识别准确率与误报谱 |
| 钠电是否必要 | 低温充电与低温可用功率上的量化差异 | 常规工况下的优势；需实测电池数据 |
| 可否直接部署 | 样机阶段已有验收口径 | 实际部署可靠性与维护成本 |
5.1 一个否定性结论
固定钠瞳策略，只把电池换成磷酸铁锂，常规日的捕获率完全相同。
我们最初以为钠电会带来全面优势，实验结果并不是。所以我们不说"钠电更好"，只说它在低温充电和低温可用功率两处成立——而这两处，恰恰是最冷、最容易死的那几天。
| 温度 | 钠电充电 | 钠电可用 | 钠电启动门槛 | 锂电充电 | 锂电可用 | 锂电启动门槛 |
|---|---|---|---|---|---|---|
| -10℃ | 可充电 | 95% | 1.6% | 禁充 | 80% | 1.9% |
| -20℃ | 禁充 | 85% | 1.8% | 禁充 | 55% | 2.7% |
| -30℃ | 禁充 | 70% | 2.1% | 禁充 | 35% | 4.3% |
| -40℃ | 禁充 | 55% | 2.7% | 禁充 | 22% | 6.8% |
5.2 敏感性：结论会不会翻
| 条件 | 捕获率均值 | 标准差 | 饿死率 |
|---|---|---|---|
| 基准（20 次重复） | 89.8% | 4.4 | 0% |
| 识别功耗 ×2 | 89.2% | — | 0% |
| 误报率 ×2 | 90.3% | — | 0% |
把最容易被质疑的两个参数各放大一倍，波动只有几个百分点，远小于策略之间 30 个百分点以上的差距。参数扰动下，饿死率始终为 0%。
V. What We Proved, and What We Did Not
| Conclusion | ✅ What this model can support | ❌ What this model cannot support |
|---|---|---|
| Is the energy budget valid | The difference among the three strategies under the energy-conservation constraint is stable | The statistical characteristics of real solar and snow-cover distributions |
| Is the strategy difference credible | The comparison uses identical inputs and does not flip under parameter perturbation | The recognition accuracy and false-positive spectrum in real scenes |
| Is sodium-ion necessary | Quantified differences in low-temperature charging and low-temperature available power | Any advantage under normal conditions; real battery data are required |
| Can it be deployed directly | The acceptance criteria for the prototype stage are already defined | Actual deployment reliability and maintenance cost |
5.1 A negative result
Fixing the Sodium Pupil strategy and replacing only the battery with LFP, the capture rate on a normal day is exactly the same.
We initially assumed that sodium-ion would bring a comprehensive advantage. The experiment showed otherwise. So we do not claim that "sodium-ion is better"; we only claim that it holds in two places — low-temperature charging and low-temperature available power — and these two places happen to be the coldest days, the days on which the device is most likely to die.
| Temperature | Sodium-ion charging | Sodium-ion usable | Sodium-ion start threshold | LFP charging | LFP usable | LFP start threshold |
|---|---|---|---|---|---|---|
| -10℃ | Allowed | 95% | 1.6% | Forbidden | 80% | 1.9% |
| -20℃ | Forbidden | 85% | 1.8% | Forbidden | 55% | 2.7% |
| -30℃ | Forbidden | 70% | 2.1% | Forbidden | 35% | 4.3% |
| -40℃ | Forbidden | 55% | 2.7% | Forbidden | 22% | 6.8% |
5.2 Sensitivity: does the conclusion flip?
| Condition | Mean capture rate | Standard deviation | Starvation rate |
|---|---|---|---|
| Baseline (20 repetitions) | 89.8% | 4.4 | 0% |
| Identification power ×2 | 89.2% | — | 0% |
| False-trigger rate ×2 | 90.3% | — | 0% |
Doubling the two most frequently questioned parameters changes the result by only a few percentage points, far smaller than the gap of more than 30 percentage points between the strategies. Under every perturbation the starvation rate remains 0%.
登录后可以留言