工廠裝儲能：電池便宜了，項目為什麼仍未必賺錢？

01 | Start with the service the battery delivers

Time-of-use arbitrage values energy; demand management depends on the tariff’s billed power and measurement window. One system can serve several purposes, but power, stored energy and backup reserve at the same time cannot be committed twice. Establish billing and production constraints before sizing.

A cell quote, a pack quote, an operating system and a commissioned project are different cost boundaries. Compare converters, thermal management, fire protection, connection works, civil works, controls, testing, warranty and service. Distinguish nameplate capacity from usable DC energy and AC energy actually delivered to load.

IEA’s 2026 review records growth in global storage deployment during 2025. Market growth does not establish an exploitable price spread for a particular factory. A commercial project should begin with bills and interval load, not an industry-average cost. [1]

02 | Adjust the spread for losses and utilisation

Delivering d kWh AC to load at round-trip efficiency η requires about d/η kWh of charging energy. With charging price pL and displaced purchase price pH, the simplified gross margin per delivered kWh is pH−pL/η, not pH−pL. The efficiency boundary must align with the quote and meter.

The 2025 NLR commercial-storage ATB uses 85% as a representative round-trip efficiency. This is a technology-scenario assumption, not a guarantee for every system. Our calculator defaults to an adjustable hypothetical 90%; switch to 85% to inspect sensitivity. [2]

Effective cycles are not simply 365 days multiplied by planned dispatches. Weekend shutdowns, seasonal production, insufficient charging windows, maintenance and reserve requirements can reduce delivered energy. Rated cycle endurance and annual commercially useful cycles are different quantities.

03 | A 2026 case: one shaved peak is not annual savings

In its updated 2026 E.S. Fox case, WATTMORE reports a February event reducing a reconstructed 206 kW load peak to a 156 kW metered peak: a 50 kW reduction. It describes roughly halving monthly peaks across the year as a modelled expectation. Keep the recorded event separate from the annual projection. [3]

This is a controls-vendor disclosure using meter and battery power to reconstruct load, not an independently audited year of counterfactual bills. It supports the importance of identifying and controlling billing peaks, not importing the saving percentage into Chinese factories, other tariffs or every month.

If storage misses the interval that sets the monthly bill, substantial discharge at other times may not lower demand charges. NLR’s REopt manual also warns that mismatched time resolution can distort demand savings. Match load intervals to the billing window rather than replacing peak information with daily energy totals. [4]

04 | A transparent scenario: which inputs change the result?

The default scenario delivers 500 kWh AC per cycle over 300 effective cycles a year, charges at CNY 0.40/kWh, displaces purchases at CNY 1.20/kWh, uses 90% efficiency, includes CNY 15,000 annual operating/provision costs and CNY 600,000 initial project cost. Additional net demand benefit defaults to zero. These are assumptions, not a quote or a current local tariff.

On that simplified basis, annual avoided purchases are CNY 180,000 and charging costs about CNY 66,667. After CNY 15,000, net operating savings are about CNY 98,333, with simple payback of 6.10 years. At 85% efficiency payback is about 6.36 years; at 150 cycles with other defaults unchanged, it is 14.40 years. The latter does not establish recovery within asset life.

The break-even peak price covers annual operating balance, not capital recovery. Simple payback omits discounting, finance, tax, year-by-year degradation, replacements, residual value and decommissioning. With nonpositive savings the tool reports no positive payback; with zero throughput it does not calculate a per-kWh threshold.

05 | What turns a forecast into an accountable delivery?

Prepare interval load covering seasons and shifts, actual bills and tariff contracts, connection/export constraints, critical loads and reserve needs. Compare without-storage and with-storage cases on consistent boundaries across production, price and availability scenarios.

Any manually entered demand benefit must be incremental and net of associated costs, verified under the same dispatch plan. Do not add the full benefit of separate arbitrage, peak-shaving and backup strategies as if capacity were unlimited. Resilience value is also not necessarily cash collected through a bill reduction.

Acceptance should state delivered usable energy, efficiency-test boundaries, response and billing windows, control strategy, data retention and variance explanations. Uptime is one condition; the question is whether the agreed service was delivered. Predefine how the baseline will be adjusted when production changes.

06 | When is the next assessment justified?

Proceed with arbitrage assessment where spreads, charging windows and high-price load are usable. Assess peak shaving where billed peaks are short and predictable with enough energy to cover them. If only monthly energy is available, obtain interval data first. If outages are the problem, establish critical loads, transfer and backup requirements first.

Mgreen Lab can discuss requirements research, decision tools and visualisation, procurement comparison and operating-interface prototypes. Electrical design, fire engineering, connection and finance require the relevant specialists. Begin with anonymised bill structure, load sampling intervals and the desired service, then define scope. This scenario calculator is not a bankable feasibility study.

07 | Sources, formulas and limits

Public evidence checked through 23 October 2026. The commercial case remains a vendor disclosure. Defaults and sensitivity values are author scenarios, not market statistics. Source code, parameters and boundary checks are saved in the research bundle. Inputs are calculated locally in the browser and are not automatically submitted to the company.

01｜先问电池替你完成什么任务

峰谷套利按电量计价，需量管理则取决于合同的计费功率与测量窗口。一个系统可以支持多种服务，但同一时刻的功率、可用电量和备用储备不能重复承诺。先确定计费制度和生产约束，再决定容量。

电芯报价、电池包报价、可运行储能系统和完整交付项目不是同一分母。比较时应列出变流器、温控、消防、电气接入、土建、控制、测试、质保和运维边界；还要区分铭牌容量、可用直流电量与交付到负荷端的交流电量。

IEA 的2026年综述记录了2025年全球储能新增部署增长，但产业规模增长不证明某家工厂有可用的价差。商业项目应从账单和负荷曲线开始，不从行业平均成本开始。[1]

02｜价差要扣掉充电损耗，再谈利用率

若每次向负荷交付 d kWh交流电，往返效率为 η，则需要约 d/η kWh购电。低价充电单价 pL、高价时段被替代的购电单价 pH，每交付一度电的简化毛差为 pH−pL/η，不是 pH−pL。效率边界应包含与报价和电表一致的损耗。

NLR的2025年商业储能ATB采用85%作为代表性往返效率，这是技术情景口径，不是每台设备的保证。本文计算器默认90%只是可调整假设；读者可以切换至85%，检验结果对损耗是否敏感。[2]

有效循环不是把365天乘以计划次数。周末停产、季节负荷、充电窗口不足、检修和为备用保留电量，都可能降低实际交付。设备可循环多少次，是耐久能力；一年真能完成多少次有价值循环，是运营问题。

03｜2026年实地案例：一次削峰不等于全年收益

WATTMORE在2026年更新的E.S. Fox案例中披露，一次2026年2月事件把重建的206 kW负荷峰值降为156 kW电表峰值，减少50 kW；同时把全年月峰值降低约一半表述为模型预测。必须把事件记录与全年预测分开。[3]

这是控制供应商披露，重建负荷来自电表和电池功率，不是独立审计的全年反事实账单。这个案例支持“识别并控制计费峰值很重要”，不支持把同一节省比例套到中国工厂、其他电价制度或所有月份。

如果储能错过某次真正决定月账单的峰值，其他时段放出很多电也可能无法降低需量费。NLR的REopt手册还提示，时间分辨率不匹配会影响需量节省估计。因此原始负荷间隔应与计费窗口相符，不能用日总电量替代峰值信息。[4]

04｜原创情景：哪些变量真正改变结果？

默认情景为每循环交付500 kWh交流电、每年300次有效循环、充电价0.40元/kWh、替代购电价1.20元/kWh、效率90%、年运维及预留费用15,000元、完整初始投入600,000元，额外净需量收益默认零。它不是实际项目报价，也不引用某地当期电价。

按该简化口径，年度替代购电价值为180,000元，充电购电费约66,667元，扣除15,000元后净运行节省约98,333元，简单回收期约6.10年。效率改为85%时约6.36年；其他默认值不变、循环减至150次时约14.40年。后一个结果不能当作寿命期内可回收的承诺。

模型显示的盈亏平衡峰时购电价只覆盖年运行收支，不包含资本回收。简单回收也没有折现、融资、税费、逐年衰减、更换、残值或退役成本。若净节省为零或负数，工具不输出虚假的正回收期；零吞吐时也不计算每度电阈值。

05｜把预测转成交付承诺，需要补哪些证据？

采购前至少准备覆盖季节与生产班次的间隔负荷、真实账单和电价合同、接入与反送电约束、关键负荷与备用需求。然后用一致的边界比较“无储能”与“有储能”，检验不同生产量、价格与可用率情景。

需量收益若手动计入本工具，必须是经同一调度方案核验、扣除相关增量成本后的额外净值。不能把电费套利方案的满额收益、独立削峰方案的满额收益和备用价值直接相加。备用价值也不等于可从电费账单兑现的现金收入。

验收条款应写清负荷端可用电量、效率测试边界、响应与计费窗口、控制策略、数据保留和偏差解释。设备在线率只是条件之一，最终要知道是否完成了约定服务。对生产变化造成的偏差，也需要预先约定重建基线的方法。

06｜什么时候值得做下一步评估？

有持续可利用的价差、可调充电窗口和足够高价用电时，可以继续评估套利；计费峰短而可预测、且有足够电量支持时，可以评估削峰。若只有月电量而没有间隔负荷，先补数据；若主要痛点是停电，则先明确关键负荷、切换和备用运行要求。

Mgreen Lab 的切入点可以是需求研究、决策工具与数据可视化、采购比较和运维界面原型；电气设计、消防、接入和项目融资需要对应专业方。讨论时先带匿名账单结构、负荷采样间隔与目标任务，再界定工作范围。这里不把情景计算器包装成可融资的可研报告。

07｜来源、公式与限制

公开资料核查截至2026年10月23日。商业案例保留供应商披露属性；默认参数与敏感性表均为作者情景，不是市场统计。完整源码、参数与边界验算保存于研究包，输入在浏览器本地计算，不会因使用本工具自动发送给公司。