Optimizing biochar quality and economics across feedstocks and pyrolysis temperatures in a 50,000 t/year carbonization demonstration

Biomass carbonization plays a critical role in industrial fuel substitution and efficient carbon sequestration, yet commercial-scale evidence linking biochar quality and economic performance based on actual operating data remains limited. Herein, based on an established biomass carbonization demonstration with an annual processing capacity of 50,000 t/year, 8 typical biomass feedstocks sourced within the collection radius (800 km) of the demonstration were selected as raw materials. Focusing on pyrolysis temperature, which is the key parameter governing biochar production, the quality and economic viability of biochar produced under the demonstration were systematically evaluated. Actual production data and economic parameters were analyzed using techno-economic assessment, correlation analysis, Random Forest prediction, Shapley interpretation, and Pareto optimization. At 500 °C, the fixed carbon contents and higher heating values of the 8 biochars ranged from 52.17% to 80.49% and from 5225 to 7525 kcal/kg, respectively. Only wood chips and bamboo chips exceeded 75% fixed carbon, reaching 75.85% and 80.49%, with corresponding higher heating values of 7373 and 7525 kcal/kg. The mixed biomass production line, using rice husk, sunflower seed shells, leafy firewood, pine branches, and green mountain wood, achieved the best economic performance, with an IRR of 8.19% and an NPV of 22.44 million Chinese yuan (CNY); in contrast, the bamboo-chip and wood-powder lines had negative NPVs and IRRs below the 2.5% discount rate. Pearson analysis linked Ash content, Volatiles matter content and temperature to heating value ( r = 0.77, 0.80, 0.31) and cost ( r = −0.55, 0.65, 0.38). Machine-learning-based Pareto optimization predicted an optimal trade-off for pine branches pyrolyzed at 375–425 °C, yielding 6821.60–6830.35 kcal/kg at 861.63–862.35 CNY. This work has developed an optimization strategy for industrial applications through experimental evaluation and machine learning feedback based on actual demonstration, thereby providing the theoretical foundation for the design and control of biochar production.