Skip to main content

Research & Development
Industrial AI blog

Operation planning method using convolutional neural network for combined heat and power system

15 April 2022

Tetsushi Ono

Tetsushi Ono
Research & Development Group, Hitachi, Ltd.


The energy efficiency of a combined heat and power (CHP) can reach about 85%, whereas conventional thermal power plants operate only at 45% efficiency or lower. CHPs perform better mainly because the heat from generators can be used as a energy source to meet heat demands or power refrigerators to generate cold water, in other words the “waste” heat is used and not wasted. Therefore, a growing number of factories and commercial buildings are installing combined heat and power (CHP) systems that include various energy storage devices (e.g., heat storage tanks) (Figure 1). To reduce the energy cost of CHPs, optimal operation plans to satisfy time-varying energy demands with minimum energy cost are required. However, conventional operation planning methods using optimized calculation have an issue with long computing time. Especially these days, operation plans need to be generated within a few minutes or even seconds to make up for output of renewable energy sources.

Figure 1. Configuration of target CHP system

Figure 1. Configuration of target CHP system


To reduce computing time, we developed an operation planning technique using a pre-trained convolutional neural network (CNN) model (see Figure 2). The CNN model, which is a kind of machine learning method, can automatically build mathematical models through a training process using input-output pairs called a training dataset. Since the trained model can estimate the optimal operation plans without iterative calculation, operation plans can be generated in just a few seconds.

To learn the time-series dependency of optimal operation plans for energy storage devices, the architecture of the proposed CNN has two key features:

the application of one-dimensional convolutional filters to extract the time-series features of input data.
the elimination of pooling layers that compress the input data.

Figure 2. Proposed CNN architecture for operation planning

Figure 2. Proposed CNN architecture for operation planning

Meanwhile, the operation plan generated by the pre-trained CNN model has the potential to violate constraints. Especially, capacity constraints of storage devices are difficult to satisfy because the energy amount in storage devices is determined by time integration of the charging/dis-charging amount. Thus, we designed rule-based flows to remove constraint violations from the operation plans.


Our evaluation using the typical demand trends of office buildings showed that the average error when using the proposed operation planning method and linear programming was 3.2%, which is much lower than the 8.0% error of the CNN with pooling calculation. In addition, we found that applying the rule-based flows can remove constraint violations in plans generated by the proposed CNN model with the average error of 3.3% (see Figure 3).

Figure 3. Model performances

Figure 2. Model performances


With the increasingly widespread use of renewable energy sources, CHP operation plans help minimize energy costs but it is essential to reduce the computation time to generate these plans. In this study, we showed how this can be accomplished by developing a planning method that uses a CNN trained with a dataset of input-output pairs for optimized calculation. To find out more, we encourage you to read our paper [1].


Tetsushi Ono,Tsutomu Kawamura, and Ryosuke Nakamura. "Operation Planning Method Using Convolutional Neural Network for Combined Heat and Power System," IEEJ Transactions on Electrical and Electronic Engineering, vol.16, issue.10, pp. 1319-1327, October 2021, doi: 10.1002/tee.23431.