Key Takeaways & Executive Findings
- •• • PSO-BP neural network models achieved MAPE of 0.278 for lithium conversion rate and 0.284 for natural gas consumption intensity, outperforming standard BP and enabling accurate prediction under wide process fluctuations. • • NSGA-II multi-objective optimization yielded Pareto-optimal solutions that stabilize lithium conversion rate between 82.45% and 87.96%, an average increase of 3.61 percentage points over baseline, while reducing natural gas consumption intensity to as low as 53.7 m3 per ton of clinker. • • For a 3.2×105 t/a lepidolite processing plant, optimized parameters enable an additional 127.1 tons of lithium metal recovery, a reduction of 1,964,912 m3 in natural gas consumption, and a CO2 emission cut of 3,763.84 tons annually, demonstrating significant resource and environmental benefits. • • Grey relational analysis identified kiln head temperature (correlation 0.98) and natural gas flow rate (correlation 0.97) as the most influential factors on lithium conversion rate and energy intensity, respectively, guiding input variable selection for the neural network models.
Abstract
The rotary kiln roasting of lepidolite for lithium extraction faces challenges of unstable lithium conversion rates and high energy consumption. To address this, a multi-objective optimization method coupling improved neural network simulation with a multi-objective genetic algorithm was proposed, targeting the synergistic optimization of lithium conversion rate (TRLi) and natural gas consumption intensity (EIng). Using long-term industrial time-series data of batching parameters and kiln operating variables, back-propagation (BP) neural network and its particle swarm optimization (PSO) improved variant were developed to model TRLi and EIng. The PSO-BP model demonstrated superior accuracy in capturing the complex nonlinear relationships, reducing mean absolute percentage errors (MAPE) to 0.278 and 0.284 for TRLi and EIng, respectively. Subsequently, the non-dominated sorting genetic algorithm II (NSGA-II) was employed to construct a multi-objective optimization model, yielding a Pareto-optimal set of process parameters that maximize TRLi and minimize EIng. The results revealed that under NSGA-II optimized conditions, TRLi could be stabilized between 82.45% and 87.96%, an average increase of 3.61 percentage points over baseline operations, while EIng could be reduced to 53.7 m3 per ton of clinker. For an annual processing capacity of 3.2×105 tons of lepidolite concentrate and sulfate mixture, this corresponds to an additional 127.1 tons of lithium metal recovery, a reduction of 1,964,912 m3 in natural gas consumption, and a decrease of 3,763.84 tons in CO2 emissions annually. This study provides theoretical and technical support for the green, high-quality, and low-carbon supply of critical raw materials for the lithium battery new energy industry.
1. Introduction
The industrial extraction of lithium from lepidolite via sulfate mixed salt roasting in rotary kilns is a cornerstone of China's lithium supply chain, yet it is plagued by inherent instability in lithium conversion rates and excessive energy consumption. The complexity arises from the highly variable composition of lepidolite concentrates, particularly the fluctuating levels of Li2O, Ca, Na, and K, which disrupt the stoichiometry of the sulfate additives and lead to suboptimal reaction conditions. Traditional mechanistic models and laboratory-scale experiments fail to capture the multivariate, strongly coupled, nonlinear, and time-delayed dynamics of the industrial process, leaving operators with empirical adjustments that often sacrifice either lithium yield or energy efficiency. This operational bottleneck not only undermines the economic viability of lepidolite-based lithium production but also conflicts with the industry's urgent need to reduce carbon emissions under China's carbon neutrality goals.
To overcome these limitations, this study introduces a hybrid approach that couples an improved neural network—specifically a particle swarm optimization-enhanced back-propagation (PSO-BP) model—with the non-dominated sorting genetic algorithm II (NSGA-II) for multi-objective optimization. By leveraging a substantial dataset of 106 continuous industrial batches, the PSO-BP model accurately maps the nonlinear relationships between 12 process variables (including batching parameters and kiln operating conditions) and the twin objectives of lithium conversion rate and natural gas consumption intensity. The NSGA-II algorithm then explores the Pareto frontier to identify optimal parameter sets that simultaneously maximize lithium recovery and minimize energy use. This data-driven methodology provides a robust framework for real-time process optimization, offering a practical pathway to achieve high-quality, low-carbon lithium production at scale.
Loading authentic research manuscript (Pages 1–5)...
DUAN Yunlong, LI Huiquan, LIU Changfeng, ZHANG Chenmu, SHI Yao, LIU Shanjun, YAN Chengsheng, FAN Bo, YU Xiuyuan, LI Zhihong, SHI Jingjing, ZHANG Jinlei (2026). Multi-objective optimization of high-quality lithium extraction from lepidolite roasting based on neural network coupled modeling. Chinese Journal of Environmental Engineering. https://doi.org/10.12030/j.cjee.202506020
Research & Educational Purpose Only: The translations, structured abstracts, analytical annotations, and data reports provided by SinoGreenTechare intended exclusively for academic research, internal corporate R&D, and educational benchmarking. They do not constitute formal engineering, chemical safety, legal, or professional advice.
Copyright & Intellectual Property Notice: Original copyright of the underlying source articles and experimental data remains with the respective authors, institutions, and original publishing journals. SinoGreenTech claims intellectual property only over its proprietary translations, analytical syntheses, and AEO structured enhancements in accordance with international fair use and academic citation principles.
Frequently Asked Questions
What are the key process variables that most significantly influence lithium conversion rate and energy consumption in lepidolite roasting, and how were they identified?
Grey relational analysis was employed to quantify the influence of 12 process variables on the two objectives. The kiln head temperature showed the highest correlation (0.98) with lithium conversion rate, while natural gas flow rate had the highest correlation (0.97) with natural gas consumption intensity. All variables exhibited correlations above 0.6, justifying their inclusion as inputs to the neural network models.
How does the PSO-BP model improve prediction accuracy compared to standard BP neural networks, and what are the specific error metrics?
The PSO-BP model optimizes the initial weights and biases of the BP network using particle swarm optimization, which mitigates the risk of converging to local minima and enhances generalization. The mean absolute percentage errors (MAPE) for lithium conversion rate and natural gas consumption intensity were reduced to 0.278 and 0.284, respectively, compared to higher errors for standard BP, demonstrating superior capability in capturing the complex nonlinear dynamics of the process.
What are the practical benefits of the NSGA-II optimized parameter set in terms of lithium recovery and energy savings for a commercial-scale plant?
For a plant processing 3.2×105 tons of lepidolite concentrate annually, the optimized parameters enable lithium conversion rates to stabilize between 82.45% and 87.96%, an average increase of 3.61 percentage points over baseline. This translates to an additional 127.1 tons of lithium metal recovered per year. Simultaneously, natural gas consumption intensity can be reduced to as low as 53.7 m3 per ton of clinker, leading to annual savings of 1,964,912 m3 of natural gas and a corresponding reduction of 3,763.84 tons of CO2 emissions.
How robust is the proposed model to the wide fluctuations in raw material composition and operating conditions typical of industrial lepidolite roasting?
The model was trained and validated on 106 consecutive industrial batches, which inherently include significant variability in feed composition (e.g., Li2O content often below 2%) and operating parameters. The PSO-BP model achieved low MAPE values (0.278 and 0.284) despite this variability, indicating strong robustness. The NSGA-II optimization further provides a Pareto front that allows operators to select parameter sets that are less sensitive to fluctuations, as the optimized batching parameters are more centralized and exhibit gentler shifts compared to baseline operations, reducing the risk of process upsets.
What are the limitations of the current study and what future work is suggested to further improve the optimization framework?
The study relies on historical data from a single industrial site, which may limit the generalizability to other lepidolite sources or kiln designs. Future work could incorporate online learning to adapt to real-time changes, integrate additional process variables such as gas composition or kiln rotation speed, and extend the optimization to include other sustainability metrics like NOx emissions. Additionally, validation of the Pareto-optimal solutions in live plant trials would be essential to confirm the predicted benefits under real-world conditions.
Related Chinese Research & Cross-Citations
Analysis of National and Local Policies for Medical Waste Treatment and Disposal in China
The escalating generation of medical waste, driven by healthcare expansion and frequent medical activities, poses significant environmental and public health risks. Under the framework of ecological civilization, China is developing a comprehensive policy system for medical waste treatment and disposal, yet the current framework remains nascent and exhibits inconsistencies between national and local policies. This study systematically analyzes the status of national and local policies from 2003 to 2024, collecting 413 policy documents (166 from national ministries and 247 from provincial governments). The analysis examines temporal evolution, regional distribution, and policy focus, alongside the influence of medical waste output, treatment technologies, facility infrastructure, and major epidemic responses. Findings reveal distinct policy phases: initial self-disposal, exploratory management, foundational system building, and rapid development. Regional disparities are pronounced, with eastern coastal areas showing more advanced policies due to greater technical and financial resources. The surge in medical waste, particularly during the COVID-19 pandemic, underscores the need for enhanced regulatory guidance. Non-incineration technologies are gaining traction for their environmental and cost benefits, and facility coverage has improved but remains uneven. The study proposes five policy principles to foster technological innovation and industrial upgrading, ensuring safe medical waste management and environmental protection.
Kinetic Analysis and Simulation of Pollutant Removal in Sewage Networks
This study investigates pollutant removal characteristics and kinetic behaviors in sewage networks, and analyzes their impact on the carbon-to-nitrogen ratio (C/N, as COD/TN) of influent to wastewater treatment plants. Source water quality sampling at drainage outlets revealed spatial and temporal variations in C/N, with domestic sewage exhibiting higher C/N than industrial sewage, and diurnal peaks reaching 6.92 versus 4.71 during off-peak hours. Using a pilot-scale adjustable sewage network system in Kunshan, experiments were conducted under high (0.491 m·s−1) and low (0.089 m·s−1) flow velocities, monitoring pollutant removal over 144 hours. Pseudo-first-order kinetics were applied to model COD and TN removal. Results showed that COD (including SCOD and PCOD), BOD5, and SS achieved approximately 80% removal within 144 h, with higher removal at low flow velocity. TN, NH3-N, and TP exhibited lower overall removal rates. Kinetic fitting revealed that COD removal rate constants (kCOD) were significantly higher than those for TN (kTN), and both decreased with increasing flow velocity: at low velocity, kCOD=0.0167 h−1 and kTN=0.0029 h−1; at high velocity, kCOD=0.0127 h−1 and kTN=0.0020 h−1. Simulations based on actual source pollutant concentrations indicated that the time for C/N to drop to the denitrification critical value of 4.50 was 12.24 h at high velocity, but shortened to 9.49 h at low velocity. These findings demonstrate that increasing flow velocity effectively retards the decline of C/N. Therefore, regulating network flow velocity to reduce hydraulic retention time is a key strategy for maintaining adequate C/N at the terminal and ensuring denitrification efficiency in wastewater treatment plants.
Preparation of Trimetallic-Carbon Composite Catalysts and Their Application in Catalytic Ozonation of Industrial Wastewater
Advanced oxidation processes (AOPs) are promising for degrading organic pollutants in water treatment. Heterogeneous catalytic ozonation (HCO) has gained attention due to its high oxidation efficiency, strong interference resistance, and low secondary pollution. In this study, a series of trimetallic-carbon composite ozone catalysts were prepared via an organic precursor calcination method using γ-Al2O3 as support. This method enhanced catalytic activity and mechanical strength while overcoming the limitations of carbon materials (low mechanical strength) and metal-based materials (poor mass transfer). The optimized catalyst, CA-FeCoCu, comprising Fe, Co, Cu, carbon, and alumina, exhibited excellent performance in phenol degradation and real industrial wastewater treatment. Characterization revealed that the synergistic effect of trimetals and the introduction of multiple carbon types increased specific surface area and hydroxyl radical (·OH) generation. In a pilot-scale fixed-bed reactor, the CA-FeCoCu/O3 system reduced COD from 120 mg·L−1 to below 40 mg·L−1, with an O3 consumption ratio (O/C) of less than 1, effectively lowering operational costs. This work provides a new strategy for developing efficient and stable heterogeneous O3 catalysts and offers a reference for the practical application of HCO in industrial wastewater treatment.
Differentiated Characteristics of Suspended Particulate Matter and Their Effects on Water Quality in the Middle and East Routes of the South-to-North Water Diversion Project
This study investigates the spatiotemporal differentiation of suspended particulate matter (SPM) characteristics, sources, and their impacts on water quality between the Middle Route (closed artificial channel) and East Route (open natural water system) of the South-to-North Water Diversion Project. Thirty sampling sites (13 on the Middle Route, 17 on the East Route) were established, and samples were collected during dry and wet seasons. Water quality parameters and SPM characteristics were analyzed, including particle size distribution, total suspended solids (TSS), chlorophyll a, and stable carbon and nitrogen isotopes. Results show that the Middle Route maintains good and stable water quality, with SPM dominated by coarse particles (>63 μm, 61.43%–94.68%), total phosphorus (TP) <0.01 mg·L−1, and a significant positive correlation between chlorophyll a and coarse particles (r=0.60), indicating algal aggregation dominates particle formation. In contrast, the East Route exhibits high and fluctuating nitrogen and phosphorus concentrations, with SPM dominated by fine particles (<20 μm, 51.26%–88.61%), TP ranging from 0.03 to 1.11 mg·L−1, and a positive correlation with fine particles, suggesting significant external inputs. Carbon and nitrogen isotope analysis reveals that Middle Route SPM primarily originates from autochthonous algae (contribution >46.75%), while East Route SPM is influenced by both terrestrial C3 plants and algae. The distinct engineering and management approaches of the two routes lead to significant differences in SPM characteristics and sources, thereby affecting water quality dynamics. The Middle Route requires an 'algal reduction and hydrodynamic optimization' strategy to control algal-derived coarse particle deposition, whereas the East Route benefits from 'retention-sedimentation and wetland purification' to reduce external fine particles and pollutant inputs. This research provides theoretical support and practical guidance for differentiated SPM management in long-distance water diversion systems.
Combined Ozone Micro-Nano Bubble Oxidation and Powdered Activated Carbon Adsorption for Removal of Taste and Odor Compounds from Drinking Water
Algal-derived taste and odor compounds (2-methylisoborneol, 2-MIB, and geosmin, GSM) in drinking water sources are poorly removed by conventional treatment. This study systematically evaluated the standalone and combined performance of ozone micro-nano bubbles (O3-MNBs) oxidation and powdered activated carbon (PAC) adsorption for removing 2-MIB, GSM, and algal cells from source water. Results showed that O3-MNBs pre-oxidation achieved >97.5% removal of odorants at 400 ng·L−1 and 67.2% algal cell removal within 30 min. When applied as a deep treatment stage, the degradation rate constant (k) was 10.1%–25.6% higher than in pre-oxidation due to lower background matrix interference. Both pre-oxidation and deep treatment reduced effluent concentrations of 2-MIB and GSM to below 10 ng·L−1, with oxidation kinetics fitting pseudo-first-order models (R²>0.95). PAC adsorption of both compounds followed pseudo-second-order kinetics (R²>0.99), with GSM equilibrium adsorption capacity approximately 20.0% higher than that of 2-MIB. In pure water, adsorption capacity increased by >10.0% compared to raw water. Based on kinetic models, a quantitative prediction method was established for O3-MNBs oxidation and PAC adsorption processes, aiming to achieve efficient odorant removal and cost optimization, providing theoretical support for advanced drinking water purification and smart water plant construction.
Pulsed Electric Field Enhancement of Nitrogen Removal Performance and Microbial Community Structure Response in Anammox Granular Sludge
This study investigated the effects of a ring-shaped pulsed electric field (PEF) (1.5 V, 4 h on-time per cycle) on nitrogen removal performance and microbial community structure of anammox granular sludge (AnGS). Two anaerobic sequencing batch reactors (R1 control, R2 with PEF) were operated under stepwise increasing nitrogen loading rates (NLR). At NLR below 1,155 mg·(L·d)−1, R2 exhibited total nitrogen removal efficiency (TRE) 7.5%–17.0% higher than R1, with biomass, specific anammox activity (SAA), and extracellular polymeric substances (EPS) increased by 5%–7%, 21%–71%, and 54%–77%, respectively. However, at NLR above 1,320 mg·(L·d)−1, the toxic effect of nitrite dominated, and PEF enhancement diminished or even reversed to inhibition. Microbial community analysis revealed that at low-to-moderate NLR, PEF increased the relative abundance of Planctomycetes and key anammox bacteria (Candidatus Brocadia and Candidatus Jettenia), along with enhanced community richness (Chao1) and diversity (Shannon/Simpson indices). At high NLR, PEF decreased microbial richness compared to R1. Principal component analysis and redundancy analysis indicated that PEF was the key factor driving community differences at low-to-moderate NLR, whereas nitrite concentration became the dominant factor at high NLR. This study provides theoretical support for enhancing the resilience and engineering application of anammox processes.