Large Language Models (LLMs) and Evolutionary Computation (EC) are increasingly being combined to support automated optimization, algorithm design, and adaptive decision-making. This survey reviews the bidirectional interaction between these two paradigms and examines how their complementary strengths can be leveraged in hybrid intelligent systems. First, we analyze how EC can enhance LLM-based systems through prompt optimization, hyperparameter tuning, and architecture search. Second, we review how LLMs can im- prove EC by supporting metaheuristic design, surrogate reasoning, adaptive operator control, and heuristic generation. We further discuss emerging co-adaptive frameworks in which LLMs and EC interact through iterative feedback loops. Beyond summarizing recent developments, the survey provides a structured perspec- tive on interaction mechanisms, application patterns, and methodological challenges, including computational cost, reproducibility, interpretability, benchmarking, and generalization. The paper concludes by outlining open research questions and future directions for developing more robust, transparent, and scalable LLM-EC systems.
This study assessed annual phosphorus (P) and potassium (K) balances in twelve traditional olive orchards in Andalusia, Spain, ranging from conventional (types A and B) to nature-based systems (types C and D). Inputs, outputs, internal flows, and ecological indicators such as the cycling index (CI) and nutrient input/output footprints per liter of olive oil were quantified. Nature-based systems recycled more nutrients, with average efficiencies of 72% for P and 59% for K, compared to 57% and 47% in conventional orchards. A regression model based on fruit yield and canopy area explained 69% of annual P demand and 76% of K demand. Without considering erosion, annual balances were neutral or slightly positive, up to + 4.22 kg P ha-1 y-1 and + 12.57 kg K ha-1 y-1 , with no significant differences between systems due to high within-group variability. Including erosion drastically altered K balances, from + 16.17 to -377.15 kg K ha-1 y-1 on average, reflecting high soil K content, while P remained relatively stable due to its lower abundance and geochemical immobilization. Overall, nature-based management enhances nutrient efficiency but cannot prevent severe losses under erosion, highlighting the need to combine agroecological practices with soil conservation to sustain fertility and long-term productivity.
Public comment platforms in civil aviation receive large volumes of semantically diverse and repetitive messages, creating challenges for timely response recommendation. This study proposes a semantic-enhanced attention-based LSTM-Seq2Seq framework. Chinese comments are segmented and encoded using a pretrained sentence model; salient and non-redundant keywords are selected through cosine relevance and Maximal Marginal Relevance, while K-Means clustering derives topic labels as global semantic cues. The encoder uses only the original comment, comment keywords, and topic label, whereas the official reply serves solely as the decoder target during training. From an initial corpus of 6425 archived comment-reply pairs, 6333 aligned pairs were retained and divided into 4446 training, 1259 validation, and 628 held-out test pairs. All clustering, vocabulary construction, and model training were restricted to the training set. On the held-out test set, the proposed framework achieved BLEU and semantic similarity scores of 0.2862 and 0.8722, respectively, compared with 0.2315 and 0.8480 for the LDA+TF-IDF baseline. These results represent relative improvements of 23.61% and 2.85%, both statistically significant under 5000 paired-bootstrap resamples ( p < 0.001 ). Although the baseline produced higher model confidence, it generated fewer distinct responses and showed greater repetition. The results demonstrate that semantic enhancement and attentive generation improve lexical agreement, semantic alignment, and response diversity.
The performance of a solar photovoltaic (PV) module or array is significantly affected by partial shading conditions (PSC), which reduce the output power and shorten the lifespan of the PV system due to the formation of hotspots on the PV cells. These conditions lead to multiple peaks on the power-voltage (P–V) curve, which cannot be accurately tracked using classical maximum power point tracking (MPPT) methods. Consequently, metaheuristic-based optimization techniques are employed to locate the global maximum power point (GMPP). This paper addresses the challenges posed by PSC in a standalone PV system by implementing a recent optimization algorithm known as the tumoral angiogenesis optimizer (TAO). The proposed MPPT approach is validated under various partial shading scenarios and benchmarked against existing MPPT algorithms, including particle swarm optimization (PSO), horse herd optimization algorithm (HOA), salp swarm algorithm (SSA), and osprey optimization algorithm (OOA). Simulations are conducted using MATLAB R2021. The performance is evaluated in terms of output power, voltage, and duty cycle, demonstrating the TAO algorithm’s capability to rapidly and accurately track the GMPP. Moreover, the tracking efficiency, convergence time, and power oscillations of each MPPT method are analyzed. The proposed method achieves an average tracking efficiency of 99.5
Accurate and early prediction of crop yield is essential for agricultural management, economic planning, and market stability, particularly for high-value products such as wine. This study presents a multi-temporal modeling framework to predict grapevine yield (in kilograms) in the province of Cádiz, Spain, a region with a deep-rooted winemaking heritage. Using a 12-year dataset that includes historical harvest records, meteorological variables, and time series of remotely sensed vegetation indices, Machine Learning (ML) regression models were developed and evaluated at three key phenological stages: post-harvest (December), post-dormancy (March), and post-flowering (June). The methodology employs a rigorous Leave-One-Year-Out (LOYO) cross-validation approach to assess model performance in the context of short time series. Results show a progressive and significant improvement in prediction accuracy throughout the growing season: the Mean Absolute Percentage Error (MAPE) decreases from 13.9
