Optimal Load Distribution of CHP Based on Combined Deep Learning and Genetic Algorithm

In an effort to address the load adjustment time in thermal and electrical distribution of power plant units, we propose optimal method based on prediction among multiple units plants. The proposed utilizes optimization by attention fine-tune a deep convolutional long-short-term memory network (CNN-LSTM-A) model for accurately predicting heat supply two 30 MW extraction back pressure units. First, inherent relationship between unit parameters is qualitatively analyzed, influencing factors are screened data-driven analysis. Then, mathematical established analyzing fitting unit’s energy consumption characteristic curves boiler turbine sides. Subsequently, using randomly chosen operating point as example, genetic algorithm used optimize loads results showed that combined learning has high accuracy, with mean absolute percentage error (MAPE) less than 1.3%. By variations, preparedness adjustments done advance. At same time, this helps reduce real-time response while enhancing load’s overall competitiveness. After that, optimizes distribution, steam rate from generation side reduced 0.488 t/MWh. Consequently, coal decreases 0.197 kg (coal)/t (steam). These described changes can greatly increase plant’s revenue CNY 6.2673 million per year. case study Zhejiang Province, China.