Emerging IoT applications are transitioning from battery-powered to grid-powered nodes. DRP, a contention-based data dissemination protocol, was developed for these applications. Traditional contention-based protocols resolve collisions through control packet exchanges, significantly reducing goodput. DRP mitigates this issue by employing a distributed delay timer mechanism that assigns transmission-start delays based on the average link quality between a sender and its children, prioritizing highly connected nodes for early transmission. However, our in-field experiments reveal that DRP is unable to accommodate real-world link quality fluctuations, leading to overlapping transmissions from multiple senders. This overlap triggers CSMA's random back-off delays, ultimately degrading the goodput performance. To address these shortcomings, we first conduct a theoretical analysis that characterizes the design requirements induced by real-world link quality fluctuations and DRP's passive acknowledgments. Guided by this analysis, we design EDRP, which integrates two novel components: (i) Link-Quality Aware CSMA (LQ-CSMA) and (ii) a Machine Learning-based Block Size Selection (ML-BSS) algorithm for rateless codes. LQ-CSMA dynamically restricts the back-off delay range based on real-time link quality estimates, ensuring that nodes with stronger connectivity experience shorter delays. ML-BSS algorithm predicts future link quality conditions and optimally adjusts the block size for rateless coding, reducing overhead and enhancing goodput. In-field evaluations of EDRP demonstrate an average goodput improvement of 39.43% than the competing protocols.

This study aimed to investigate the combined effects of photoperiod (L6h: D18h, L12h: D12h, and L18h: D6h) and light intensity (200, 400, and 600 lx) on the growth performance, physiological stress, and gonadal development of mandarin fish (Siniperca chuatsi) in recirculating aquaculture systems. The results demonstrated that compared with other photoperiods, the long photoperiod was associated with reduced weight gain rate (WGR), condition factor (CF), and survival rate (SR), as well as an increased hepatic vacuolization ratio and metabolic stress. The activities of liver function-related enzymes, alkaline phosphatase (ALP) and aspartate aminotransferase (AST), were significantly influenced by the interaction of photoperiod and light intensity. Concurrently, the long photoperiod strongly directed energy allocation toward reproduction, as evidenced by significantly elevated testicular index, proportion of spermatocytes and spermatids, and intragonadal levels of testosterone (T) and estradiol (E2). The ovarian index was also regulated by photoperiod and its interaction with light intensity. In contrast, light intensity alone primarily affected alanine aminotransferase (ALT) activity and, through interaction with photoperiod, co-regulated antioxidant defense. Multi-omics analysis of testicular tissues revealed that this growth-reproduction trade-off process involved the activation of the Neuroactive ligandreceptor interaction signaling pathway and coordinated regulation of pathways including ABC transporters, Arachidonic acid metabolism, and Glycerophospholipid metabolism. In conclusion, photoperiod acts as the dominant environmental cue, whereas light intensity plays an auxiliary fine-tuning role. The long photoperiod induces energy allocation from growth toward reproduction via stress signaling. These findings provide a critical theoretical foundation for the precise management of the light environment in the recirculating aquaculture of mandarin fish.

Large-scale multi-hop wireless IoT networks are increasingly being deployed to support critical applications with stringent QoS requirements. Robust fault detection in such networks is vital for ensuring normal network operations and enabling proactive maintenance especially in adversarial environments. However, the existing researches primarily focus on data-driven attack detection, leaving natural network fault detection largely unexplored. Due to the absence of dedicated routers, multi-hop IoT networks require network nodes to transmit their own data and also relay data for others, which introduces inherent and cascaded anomalies, making it difficult to locate anomaly sources. Furthermore, the underlying CSMA/CA based communication protocols adopted in IoT networks incur unpredictable delays in transmissions, thereby complicating timely anomaly detection and accurate diagnosis. This paper proposes innovative network fault detection technologies tailored for large-scale multi-hop IoT networks. We introduce a spatio-temporal graph neural network (STGNN) model capable of accurately detecting both node-level and edge-level faults. Our model takes both temporal data and metadata for context-aware forecasting and reconstruction. To address the lack of labeled data, we adopt an unsupervised learning paradigm. We evaluated our model on various large-scale network topologies and transaction log datasets, and results demonstrate that our model consistently outperforms baseline models across most cases.

This study investigated the effects of dietary n-3 highly unsaturated fatty acids (n-3 HUFA) on growth performance, antioxidant capacity, lipid metabolism, intestinal morphology, and testes development in male Macrobrachium rosenbergii, aiming to determine the optimal dietary inclusion level of n-3 HUFA. Prawns with an initial weight of 17.49 +/- 0.14 g were fed six isonitrogenous and isolipidic diets containing graded levels of n-3 HUFA: 0.12% (CON), 0.31% (T1), 0.60% (T2), 0.83% (T3), 1.10% (T4), and 1.31% (T5). The T2 group showed the best growth performance, with significantly higher final body length (FBL) and specific growth rate (SGR) (P<0.05). Broken-line regression analysis indicated that 0.57% n-3 HUFA is optimal for maximizing weight gain. The antioxidant enzyme results showed that the T2 group exhibited higher activities of total superoxide dismutase (T-SOD), total antioxidant capacity (T-AOC), and catalase (CAT), together with the lowest malondialdehyde (MDA) content, indicating the strongest antioxidant capacity in this group. Increasing dietary n-3 HUFA levels resulted in reduced triglyceride (TG) and elevated high-density lipoprotein cholesterol (HDL) contents, with T5 showing the most improved lipid profile. Histological analysis revealed more intact hepatopancreatic structure in the T1 and T2 groups, and superior intestinal fold height and width in T2, while the T2 and T3 groups showed better testes development. Analysis of the intestinal microbiota based on 16S rRNA gene sequencing revealed that the T2 group exhibited higher intestinal alpha-diversity, with ASVs, ACE, and Chao1 indices all being significantly higher than those in the other groups (P<0.05). Concurrently, the relative abundance of potentially pathogenic bacterial families, namely Enterobacteriaceae and Mycoplasmataceae, demonstrated a decreasing trend with increasing dietary n-3 HUFA levels. These results suggest that dietary n-3 HUFA at 0.57-0.83% enhances growth, testes development, antioxidant status, lipid metabolism, and intestinal health in male M. rosenbergii.

The rapid development of marine recirculating aquaculture systems (RASs) worldwide offers an efficient and sustainable approach to aquaculture. However, the slow start-up of the nitrification process under low-temperature conditions remains a significant challenge. This study evaluated multiple start-up strategies for moving bed biofilm reactors (MBBRs) operating at 13–15 °C. Among them, the salinity-gradient (SG) strategy exhibited the best performance, reducing the start-up time by 38 days compared to the control, with microbial richness (Chao1 index) reaching 396 and diversity (Shannon index) of 4.89. Inoculation with mature biofilm (MBI) also showed excellent results, shortening the start-up period by 26 days and achieving a stable total ammonia nitrogen (TAN) effluent concentration below 0.5 mg/L within 132 days. MBI exhibited the highest microbial richness (Chao1 index = 808) and diversity (Shannon index = 5.55), significantly higher than those of the control (Chao1 index = 279, Shannon index = 3.90) and other treatments. The hydraulic retention time-gradient (HRT) strategy contributed to performance improvement as well, with a 24-day reduction in start-up time and a Chao1 index of 663 and a Shannon index is 4.69. In contrast, nitrifying bacteria addition (NBA) and carrier adhesion layer modification (CALM) had limited effects on start-up efficiency or microbial diversity, with Chao1 indices of only 255 and 228, and Shannon indices were both 3.24, respectively. Overall, the results indicate that salinity acclimation, mature biofilm inoculation, and extended HRT are effective approaches for promoting microbial community adaptation and enhancing MBBR start-up under low-temperature marine conditions.

This paper explores the transformative potential of the IoT paradigm in promoting smart agriculture. Key challenges lie in how to connect agriculture sensors to remote cloud servers in the absence of feasible communication infrastructure and the unreliable wireless links in rural areas. To address these issues, we propose an innovative two-tier smart agriculture architecture: an Unmanned Aerial Vehicle (UAV) aided agriculture network model, which leverages UAVs as intermediaries to collect and route data from agriculture sensors to cloud servers. This novel architecture leads to two particular problems, i.e., data packet scheduling in the first-tier networks and multi-hop routing in the second-tier UAV mesh network. To that end, we present formal Markov decision process (MDP) based problem formulations for both tiers, with a primary focus on the more challenging multi-hop routing problem in the second-tier network. This problem is approached as a multi-agent reinforcement learning (MARL) framework, for which we introduce a novel distributed algorithm - Focus Coordination: attention-guided Multi-Agent Deep Deterministic Policy Gradient (FC-MADDPG). This algorithm reduces communication overhead and mitigates the risks associated with single-node failures. We evaluated the performance of the proposed FC-MADDPG algorithm, demonstrating its efficacy in enhancing data transmission reliability and efficiency.

Indoor positioning, aimed at monitoring human behavior, overseeing IoT device operations, and self-localization of indoor mobile entities, has garnered significant attention. One of the positioning methods introduced to the market is the IEEE 802.11 FTM (Fine Time Measurement) protocol, which estimates indoor location by measuring the distance based on the propagation time between Wi-Fi Access Points (AP) and stations (STA). In this paper, we present an indoor positioning method by measuring FTM and RSSI in an indoor environment, and apply machine learning to the collected data. In actual indoor environments, it is conceivable that obstacles may alter or block the radio propagation paths, affecting the propagation time of FTM and RSSI used in machine learning, thereby impacting the accuracy of position estimation. We created scenarios with and without obstacles that block radio propagation indoors, collected data, and performed position estimation using supervised machine learning. Our results indicate that if the input data includes at least one FTM-enabled Wi-Fi AP, the accuracy of position estimation using machine learning improves compared to using only RSSI. Additionally, the results showcase that the proposed method is effective for position estimation even in environments with indoor obstacles.

Indoor positioning methods using wireless signal propagation data have attracted significant interest for applications such as monitoring IoT devices and tracking human behavior. The IEEE 802.11 FTM (Fine Timing Measurement) protocol, used for Wi-Fi location, was introduced to the market in 2016. The FTM protocol can measure the distance between Wi-Fi access point (AP) and station (STA). By measuring the distances between multiple Wi-Fi APs and STAs, the indoor position of the STA can be estimated. In this paper, we collected FTM and RSSI measurements in an indoor environment using FTM-enabled Wi-Fi APs and STAs, and evaluated indoor positioning accuracy through geometric calculations based on FTM data as well as a machine learning approach using both FTM and RSSI data as inputs. Furthermore, for the machine learning approach, we also assessed the impact of varying the number of Wi-Fi AP data elements supplied to the model in increments of AP count. The results demonstrated that positioning accuracy achieved by the machine learning approach surpassed that of geometric calculations. Moreover, even when the number of input data elements to the machine learning model was limited, utilizing FTM data obtained from at least one AP mitigated the degradation in positioning accuracy within the machine learning framework.

As the number of devices with multiple communication interfaces increases, the connection redundancy is being considered for efficient bandwidth utilization and improved reliability. Accordingly, multipath transport technologies are attracting attention. This paper investigates the Multipath Quick UDP Internet Connection (MPQUIC) transport protocol with the goal of enhancing data throughput and packet transmission efficiency. We introduce a novel Blocking Probability (BLP) path scheduler, which leverages a learned blocking threshold for probability-based blocking decision-making, and an innovative congestion controller, which not only dynamically adjusts congestion window size but also optimizes packet size. Our simulations demonstrate that BLP significantly improves network throughput, packet delay time, and overall transmission efficiency in both homogeneous and heterogeneous network environments. By outperforming benchmark schedulers, BLP showcases the potential of advanced scheduling strategies to adapt to network dynamics, ensuring reliable and efficient data delivery.

Liver fibrosis represents an important pathological stage during chronic hepatopathy development, posing a significant threat to human health. Hepatic stellate cells (HSCs), an essential hepatic non-parenchymal cells, have a key effect on fibrogenesis, with their activation being a hallmark of liver fibrosis. MicroRNAs (miRNAs), the small non-coding RNAs, become the critical biomarkers and regulatory molecules in fibrotic processes. Among them, miR-125a-5p is implicated in cancer and inflammatory pathways, yet its functional role and mechanistic involvement in HSC activation remain poorly understood. According to our findings, miR-125a-5p expression was significantly decreased in TGF-β-activated HSC-T6 cells. Notably, ectopic miR-125a-5p overexpression effectively inhibited TGF-β-mediated HSC-T6 activation. Further mechanistic investigations revealed that miR-125a-5p attenuated HSC activation while ameliorating liver fibrosis through regulating the TGF-β/Smad2/3 pathway and autophagy. Additionally, TGFβR1 was miR-125a-5p’s target gene. Collectively, miR-125a-5p negatively regulates HSC activation in liver fibrosis, exerting its anti-fibrotic activities through suppressing the TGF-β/Smad2/3 pathway and autophagy modulation.

Emerging smart agriculture is critical for optimizing crop quality and quantity. However, its realization faces significant challenges, particularly the lack of feasible communication infrastructure and poor wireless connectivity in rural areas. This paper presents a novel Unmanned Aerial Vehicle (UAV) assisted two-tier agriculture network architecture to address these issues, where UAVs act as intermediaries between agriculture sensors and cloud servers. Our key innovation is a Large Language Mode (LLM)-based approach for context-aware semantic mapping, introducing an innovative Semantic Criticality Index (SCI) that dynamically assesses the importance of agricultural data. This novel SCI drives our formulation of the agricultural sensor data collection scheduling problem as an optimization problem to minimize energy use in sensors and UAVs, solved using a proposed Semantic-Guided Deep Q-Network (SG-DQN) algorithm that optimizes energy consumption and resource allocation based on semantic context. Simulations using public agricultural datasets show significant improvements over traditional methods in energy efficiency and data classification accuracy.

The long-chain polyunsaturated fatty acids (LC-PUFA) synthesis characteristics in fish, including the identification of key genes and their functional activities, are crucial for developing targeted nutritional strategies. Although the endogenous LC-PUFA synthesis capacity of largemouth bass (Micropterus salmoides) has been preliminarily reported, the key genes involved and their functional activities remain uncharacterized. In this study, we systematically validated the LC-PUFA synthesis pathway in largemouth bass by comparing diets supplemented with LC-PUFA (LD) or non-LC-PUFA (NLD). The results demonstrated that largemouth bass can endogenously synthesize LC-PUFA from C18 PUFA, albeit with compromised growth performance and hepatic health under NLD feeding. Furthermore, we characterised the genes potentially involved in LC-PUFA biosynthesis, including two fatty acid desaturases (fads2a and fads2b) and three fatty acid elongases (elovl5, elovl4a and elovl4l). Bioinformatics analysis confirmed the evolutionary conservation of these genes and revealed the unique evolutionary divergence mechanism of the fads2s in largemouth bass. The LC-PUFA biosynthesis-related genes knockdown in primary hepatocyte and yeast heterologous expression confirmed that Fads2a has Delta 4, Delta 5, Delta 8 desaturase activities and Fads2b Delta 5, Delta 6, Delta 8 desaturase activities. And Elovl5 primarily participated in the elongation of C18 and C20 PUFA, whereas Elovl4a and Elovl4l were involved in the elongation of C18, C22 and longer-chain PUFA (> C24). Overall, we constructed a complete LC-PUFA synthesis pathway in the largemouth bass, thereby enhancing the understanding of the characteristics and evolutionary mechanisms of LC-PUFA synthesis in freshwater carnivorous fish.

In China, a newly emerging disease, White Syndrome (WS), has been identified in Macrobrachium rosenbergii. This study investigates the impact of WS, characterized by systemic whitening, abdominal muscle and hepatopancreatic atrophy, and the presence of blisters. Histopathological and transcriptomic analyses were performed on hepatopancreas and abdominal muscle tissues of affected prawns. Histological examinations revealed damaged hepatopancreatic tubules with enlarged intercellular spaces, while muscle tissues exhibited signs of atrophy. A total of 191 differentially expressed genes (DEGs) were identified from the hepatopancreas, along with 233 KEGG pathways. The primary pathways associated with WS include Lysosome, Glycosaminoglycan degradation, Pentose and glucuronate interconversions, Starch and sucrose metabolism, Biosynthesis of unsaturated fatty acids, Sphingolipid metabolism, Steroid biosynthesis, Fatty acid elongation and Retinol metabolism. Similarly, 191 DEGs were identified from the abdominal muscle, with 203 KEGG pathways. WS-related pathways included Proteasome, Glycolysis/Gluconeogenesis, and Starch and sucrose metabolism. These findings suggest that WS in M. rosenbergii may be associated with nutritional deficiencies linked to excessive reproductive activity in female prawns. This study provides critical insights for future research on WS and its management in M. rosenbergii.

This study aims to investigate the effects of dietary supplementation with protease on the growth performance, liver health, immunity, and intestinal microbiota of grass carp. Three levels of protease, 0 U/kg (P0), 6000 U/kg (P6000) and 12,000 U/kg (P12000), were formulated into diets using post-spray technology. After 9 weeks of feeding, the P6000 group had a higher weight gain rate than the P0 group, with both P6000 and P12000 groups exhibiting higher protein retention ratio and lower feed conversion ratio compared to P0. As protease levels increased, crude protein content and intestinal protease activity increased, while alkaline phosphatase (ALP) content decreased. Concurrently, the P6000 group showed higher aspartate transaminase (AST) and triglyceride (TG) levels but lower amylase content compared to P0. In terms of immune indicators, the expression levels of Intelectin and MHC-II β in the P6000 and P12000 groups were significantly lower than in P0, and IgM expression gradually increased with increasing protease levels. After challenge with Aeromonas veronii, the P12000 group showed higher survival rate. The histological results indicated that the intestinal villi length in the P6000 group was notably longer than in other groups. The intestinal microbiota analysis revealed that with increasing protease levels, Fusobacteriota abundance decreased and Actinobacteriota and Cyanobacteria abundance increased at the phylum level, while Aeromonas relative abundance decreased at the genus level. Overall, 6000 U/kg protease boosts grass carp growth and nutrient retention maximally, while 12,000 U/kg protease increases survival but may increase liver metabolic burden. Given these outcomes, the optimal protease inclusion level for grass carp aquaculture is approximately 6000 U/kg.

Alkalinity, a key environmental stressor in saline-alkali ecosystems, adversely affects the growth and survival of aquatic organisms. Genetic improvement of alkali-tolerant fish strains offers a promising strategy for utilizing saline-alkaline water resources; however, the molecular mechanisms underlying the response to alkalinity stress remain inadequately understood. This study aimed to identify novel molecular signatures of alkaline exposure in grass carp by integrating gut microbiome profiling with host transcriptome data. Histological analysis revealed significant alterations in the height of intestinal folds, muscle layer thickness, and fold width in response to NaHCO₃ exposure, along with an increased number of goblet cells under alkalinity stress. Differential gene expression (DEGs) analysis identified 1620, 6564, and 3190 genes with significant expression changes at 24, 48, and 72 h of NaHCO₃ treatment, respectively, compared to controls. Several known alkalinity-responsive genes, such as aquaporin 1a (aqp1a), carbonic anhydrase 6 (ca6), heat shock protein 30 (hsp30), prostaglandin-endoperoxide synthase 2b (ptgs2b), caspase 23 (casp23), solute carrier family 7a (slc7a), toll-like receptor 5 (tlr5), and toll-like receptor 13 (tlr13), were identified and validated through real-time quantitative PCR. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses revealed that immune- and disease-related signaling pathways played critical roles in mitigating acute alkaline stress. Furthermore, a total of 1,521,323 quality-filtered sequences with an average of 416 bp in length were generated through 16S rRNA sequencing. Bioinformatics analysis indicated that NaHCO₃ exposure reduced microbial diversity and altered the gut microbiota composition in grass carp. Notably, Fusobacteriota abundance significantly increased, while Firmicutes and Proteobacteria decreased substantially at 48 h post-alkalinity stress. Integrative analysis further highlighted strong correlations between specific bacterial taxa and alkalinity-responsive differentially expressed genes (DEGs). These findings provided valuable insights into the molecular mechanisms underlying alkalinity stress and identified potential targets for molecular breeding to enhance alkaline tolerance in grass carp.

As the number of wireless devices supporting multiple communication interfaces increases, the connection redundancy is being considered for efficient bandwidth utilization and QoS improvement. Accordingly, network technologies must adapt to emerging multi-interface devices to improve network performance. Multipath TCP (MPTCP) is default multipath transport protocol desired for networks with multi-interface devices and has achieved success in computer networks. However, it has not been well studied for wireless networks, especially for carrier sense multiple access (CSMA) based wireless networks, which present great challenges to round trip time (RTT) computation and multipath scheduling. This paper introduces MPTCP techniques for heterogeneous WiFi and 5G networks. We first model a proposed 5-state congestion control algorithm and WiFi CSMA function. We then present an innovative RTT computation method and a novel loss-ware multipath scheduling mechanism. We evaluated the proposed MPTCP techniques under varying network configurations. Our MPTCP can significantly outperform conventional MPTCP.

The cultivation of largemouth bass (Micropterus salmoides), a species of significant economic value in aquaculture, has experienced notable growth recently. However, the deterioration of water quality seriously affects the metabolic responses of M. salmoides. While compound microbial agent (CMA) is widely utilized for ecological rehabilitation and water filtration, its application in M. salmoides has not been reported. Here, based on physio-biochemical tests and 16S rRNA sequencing, we investigated the effects of CMA (yeast, Bacillus subtilis, and lactic acid bacteria) on the water quality within the recirculating aquaculture system, along with physiological indices and gut microbiota of M. salmoides. Compared to the control and single microbial agent (yeast), CMA treatment improved the water quality by improving the dissolved oxygen and delaying the increase of pH, total nitrogen, total phosphorus, ammonia nitrogen, and nitrite. The 16s rRNA gene sequencing revealed that the water treated with CMA exhibited elevated levels of chao1, Shannon, Pd, and a larger population of dominant bacterial. Besides, higher values of ACE, chao1, Shannon, and OTU level, and lower Simpson index were found in CMA treated M. salmoides samples, suggesting that CMA treatment enhanced the species richness and diversity of gut microbiota of M. salmoides. Furthermore, CMA treatment hindered the generation and proliferation of harmful bacteria, such as the Mycoplasma mobile 163K species and the Erysipelotrichaceae family, which was associated with enhanced antioxidant enzymatic activity and decreased MDA level in both the serum and liver. These findings shed light on the essential roles of CMA in M. salmoides culturing and introduce an innovative approach to enhance the aquatic environment.

The crowdsourced information is useful to calibrate Advanced Driver Assistance Systems/Autonomous Driving (ADAS/AD) parameters for automated and autonomous vehicles. However, learning such information in vehicular networks is challenging. On the one hand, data collected by individual vehicle may be not sufficient to train a large scale machine learning model. On the other hand, uploading raw data to cloud server is likewise impractical due to enormous communication bandwidth requirement and data privacy threat. This paper seeks a solution by applying federated learning (FL). We aim to improve FL algorithm stability to increase prediction accuracy. Accordingly, we propose a variance-based and structure-aware FL (VSFL), in which a variance-based model aggregation method is introduced for FL server to make optimal model aggregation and a structure-aware model training scheme is provided for vehicle clients to tackle statistical heterogeneity without compromising performance. We first provide theoretical analysis for the proposed VSFL. We then validate the effectiveness of VSFL algorithms on vehicle trajectory prediction using both synthetic data and real data.

AIMS:Developing energy-saving and ecofriendly strategies for treating harvested Microcystis biomass. METHODS AND RESULTS:Streptomyces amritsarensis HG-16 was first reported to effectively kill various morphotypes of natural Microcystis colonies at very high cell densities. Concurrently, HG-16 grown on lysed Microcystis maintained its antagonistic activity against plant pathogenic fungus Fusarium graminearum. It could completely inhibit spore germination and destroy mycelial structure of F. graminearum. Transcriptomic analysis revealed that HG-16 attacked F. graminearum in a comprehensive way: interfering with replication, transcription, and translation processes, inhibiting primary metabolisms, hindering energy production and simultaneously destroying stress-resistant systems of F. graminearum. CONCLUSIONS:The findings of this study provide a sustainable and economical option for resource reclamation from Microcystis biomass: utilizing Microcystis slurry to propagate HG-16, which can subsequently be employed as a biocontrol agent for managing F. graminearum.

With the advent of 5G and beyond communication technologies, the consumer Internet of Things (IoT) devices are evolving from the current-generation to the next-generation. Next-generation IoT devices can support multiple communication interfaces and perform more functions. Accordingly, IoT network technologies must adapt to the emerging next-generation IoT devices. Routing is an inevitable technology in multi-hop IoT networks. However, as IoT devices become more and more diverse, IoT networks become more complex. As a result, the routing problem becomes more and more complicated for traditional protocols and mathematical optimization approaches to provide optimal solutions. Machine learning based routing techniques have been recently proposed and can outperform traditional routing methods in complex network environments. To that end, this paper presents a machine learning based routing link scheduling scheme for heterogeneous wireless IoT networks. We formulate the routing link scheduling problem as a combinatorial optimization problem, which is then parameterized for application of machine learning algorithm and the parameterized problem is solved using primal-dual approach with zero duality gap. A heterogeneous graph neural network (HetGNN) algorithm is proposed to update the primaldual problems. We evaluate the proposed HetGNN model under networks with randomly deployed heterogeneous nodes. Compared with a convolutional neural network (CNN) model and a homogeneous GNN (HomGNN) model, the proposed HetGNN model can improve network throughput, reduce link capacity violation and interference link violation.