Optimized battery control

October 8, 2025

As solar generation increases in modern grids, solar producers are faced with a problem: they generate energy when the sun is shining, at the same time as every other solar plant. This means that in the middle of the day, there is too much electricity generated, so the value of this electricity decrease substantially.

The problem

In some areas of Japan, it is already common for prices to reach the lowest allowed limit of 0.01 yen/kWh on the JEPX day-ahead electricity market for several hours a day.

This is creating revenue pressure for solar producers, further amplified by the current regulation for solar subsidies. Under the current Feed-in Premium (FIP) subsidy system in Japan, solar plant operators are compensated only if the price is above the lower limit of 0.01 yen/kWh. This means that not only producers have to sell their energy for almost nothing, they also don’t get the subsidy.

Another problem that is becoming more pressing for solar producers is curtailment: when solar demand is excessive, the grid operator is forced to limit the amount of energy produced to avoid overloading the grid. For solar operators this means not being able to generate electricity that the plant could have produced, effectively wasting it.

The role of batteries

Batteries can solve all these problems:

they can charge during the day when prices are low, and allow to sell electricity at a higher price later in the evening (price arbitrage)
they can charge when prices are their lowest (0.01 yen/kWh) and discharge later, thus ensuring that plant operators receive FIP subsidy payments
they can charge during times of curtailment, to save the energy from being wasted

Batteries are therefore becoming an important driver of revenue for solar producers in Japan, and it is becoming more common to install a battery alongside a solar power plant to increase plant revenues.

However, many solar portfolios in Japan consist of low-voltage, small-scale, distributed power plants, where adding batteries could significantly complicate operations: a battery needs to be actively controlled, as opposed to a solar panel that just produces energy when the sun shines. To effectively control (or schedule) a battery, we need to have an estimate of future prices and solar generation, so that we can plan when to charge and when to discharge. There are fundamentally two ways of obtaining these estimates: by looking at past data to extract simple rules, or to use forecasts.

For solar generators with small, distributed portfolios, generating a solar forecast for each location is complicated and often cost-prohibitive. This leads to many taking shortcuts to save costs. For example, instead of creating individual forecasts for each plant, they duplicate the same forecast across similar plants, or assume the solar generation for tomorrow is going to be the same as today. This is an extremely simple assumption which is not able to account for weather variation day by day. For example, if today is sunny and tomorrow is cloudy, this would overestimate tomorrow's generation. This is exacerbated by the fact that weather patterns are highly correlated for the same grid zone, so that errors don’t tend to cancel each other out. As for predicting electricity market prices, this is even more complex for small producers. Again, the same technique is often used: assuming that the prices tomorrow are going to be the same as today, sacrificing accuracy for costs.

Battery deployment scenarios

Once we have an estimate of the solar generation and prices, we need to find the best scheduling for charging and discharging the battery. Here we will analyze two ways of doing this:

A simple scheduling algorithm that start charging in the morning at 10am until the battery is full, and start discharging at 6pm until the battery is empty. This is the way most batteries are scheduled in small-scale plants.
Our optimized scheduling based on an analytical optimization model, which uses our solar and price forecasts.

To test the effects of these two types of scheduling, we ran simulations over a year (from July 2024 to June 2025) using data of a real solar plant we manage. The plant is 2 MW AC/2.5 MWp with a FIP rate of 36 JPY/kWh. We assume a 4-hour battery with AC capacity of 1.5 MW and 6 MWh of storage as the market seems to be converging to 4-hour battery systems as the gold standard.

We tried to make these simulations as realistic as possible:

For our optimized scheduling, we use the real solar and price forecasts for the plant available at the time. These are real forecasts that were created for the asset over the last year. For the simple scheduling, we assume the operator uses the generation on the previous day as a “forecast” for next-day generation.
We simulate sending day-ahead plans for supply to the market, and we calculate the imbalance resulting from forecast errors and changes in the plans.
We simulate the actual operation of the battery, considering SoC limits, charging and discharging efficiency, and cycling cost.

For reference, we also tested the case with perfect knowledge, that is, assuming perfect forecasts for solar generation and prices. This will give us an upper bound for the possible improvements that scheduling optimization can bring, independently of the quality of the forecasts. In addition to this, we ran the same simulation for the solar plant without battery, to have an idea of what the impact of the battery is compared to the solar generation alone.

Annual revenue by case

While the total revenue is largely driven by the FIP subsidy, the true value of the battery lies in its marginal revenue contribution. By analyzing the revenue gained only from the battery's operation (that is, the total revenue minus the revenue from the solar plant alone) we can clearly see the impact of our optimization. As shown in the graph, our optimized scheduling consistently delivers a greater revenue increase than the basic scheduling strategy.

What if we remove the effect of the solar generation? We can do this by looking at only the marginal improvements that we can have by adding a battery, thus subtracting the case without the battery. This is actually the important number, since it is the amount that matters when judging whether to make the battery investment or not. In this case the results are clearer:

Revenue gain from battery deployment

The optimized cases have higher total “battery” revenues compared to the basic scheduling cases. We find these results:

For the case with perfect information (our upper bound): the optimized case increase total revenues by 18.47%
When we use our forecasts, the optimized scheduling increase total revenues by 7.40%, which is expected since the forecasts can never be perfect.
If we include the effect of curtailment, our optimization is able to increase total revenue by 10.33%

While these gains may seem small, given the current tight economics of batteries a 10% revenue increase may be the difference between a negative and positive IRR.

The optimized scheduling is able to better take advantage of price differentials at the right time. For example, this is an average day for the two systems:

Basic scheduling, perfect forecasts

The basic scheduling starts discharging the battery at 6pm, no matter what the price is.

Optimized scheduling, perfect forecasts

The optimization is instead able to discharge exactly at the slots with the highest prices, thus maximizing the revenue.

We also analyzed the case with a stand-alone battery. In this case, an optimized scheduling has a much bigger advantage compared to a simple scheduling algorithm: all of the revenue in this case comes from price arbitrage, without fixed revenue streams such as FIP, which do not change much with different battery behaviors. This case therefore better represents the advantage of an optimized schedule.

For the stand-alone battery case, we see a 43.4% increase in revenue for the optimized case compared to the simple scheduling (charge from 10am, discharge from 6pm). This is mostly because the optimized schedule is able to charge and discharge at the best time to maximize the price difference. Also, the optimization is not limited to one cycle per day: during the test year, the basic scheduling cycles 365 times (exactly once per day), while the optimization managed 598 cycles (about 1.64 cycles per day).

Advantages of optimization over time

Note that the battery optimization takes into account the battery cycle cost and round-trip efficiency too, so it will only cycle the battery when the potential revenue is worth the cost.

Including a battery with a solar plant can increase revenues by over 20%. This is mostly because of the increased FIP payment thanks to the avoidance of zero-yen slots discharge. Out of this revenue increase, the contribution of price arbitrage (ie. changing the time of discharge to take advantage of higher prices) is relatively small, which means that battery scheduling does not have a huge impact compared to a simple strategy designed to charge during likely zero-yen slots. However, because the economics of batteries right now are quite tight, every small improvement is essential to make this investment work. Optimized scheduling can increase the battery return on investment by about 10%.

For assets with lower FIP (assets coming online today have FIP less than a third of what we tested here), the effect of price arbitrage is much stronger, thus optimization becomes a more important driver of revenue. Moreover, as solar penetration in the Japanese grid increases and more fossil fuel plants go offline, we expect to see more extreme daily price differentials, which again makes arbitrage relatively more important.

For the case of a stand-alone battery, optimized scheduling is already essential. This can increase revenue by over 40% compared to a simple scheduling strategy.

Authors

Riccardo Iacobucci

Principal Energy & Data Scientist