To address local crushing, slippage, and force-transmission failure of TBM grippers when tunneling in weak and fractured surrounding rock, this study investigates the engineering feasibility of using polycarbonate (PC) plates as a novel reusable load-transfer cushion for grippers, aiming to improve the adaptability and stability of the gripper support system under unfavorable surrounding rock conditions. Uniaxial compression, rebound, and three-point bending tests were conducted to examine the mechanical properties, deformation characteristics, and failure behaviors of single-layer and double-layer PC plates, with particular focus on their compressive bearing capacity, elastic recovery performance, flexural resistance, and interlayer collaborative bearing characteristics. The results show that the single-layer PC cushion exhibits excellent mechanical performance, with a compressive strength of 341–357 MPa, an elastic modulus of 1.28–1.62 GPa, a rebound rate of nearly 95% within the elastic limit, and an average flexural strength of 117.57 MPa. These properties ensure stable bearing capacity and reversible deformation under repeated loading–unloading cycles of the gripper. Within the elastic range, the PC plate can recover most of its deformation after unloading, which facilitates repeated reuse in engineering applications. The single-layer PC plate exhibits a ductile failure mode with large-deformation capacity, providing sufficient deformation redundancy for the gripper under eccentric loading and local bending conditions and reducing the risk of sudden brittle failure during force transfer. The double-layer composite structure exhibits compressive bearing capacity comparable to that of the single-layer plate, with a peak stress of 356.20–363.87 MPa. No interfacial debonding or cracking occurs during loading, and effective collaborative bearing is achieved between the two PC layers, indicating that the bonding interface remains stable under compression. The double-layer structure achieves a flexural strength of 369.81 MPa, which is 3.18 times that of the single-layer plate, indicating that the laminated configuration can significantly improve flexural bearing capacity while maintaining stable compressive performance. The epoxy adhesive can efficiently transfer interfacial shear stress, restrict relative sliding between the two PC layers, and enable the composite structure to behave as an integral solid beam, thereby fully utilizing the collaborative bearing capacity of the double-layer structure. These results demonstrate that PC plates possess favorable compressive bearing capacity, elastic recovery ability, ductile deformation characteristics, and interlayer collaborative performance. Field tests were also carried out in a deep-buried water-diversion tunnel over multiple tunneling cycles. Compared with wood-block cushions, the test results demonstrate the prominent advantages of the PC cushion in load-transfer performance, reusability, and operational efficiency. This study provides a technical basis for optimizing the number of layers, thickness, bonding process, and structural configuration of TBM gripper cushions for tunneling in weak and fractured surrounding rock.
The degradation of riverine physical habitats and its impact on aquatic communities have become a central concern in river ecology. Channel regulation measures, such as bank hardening, bed leveling, and dredging, ensure flood safety but simultaneously destroy natural habitats, triggering habitat fragmentation, biodiversity decline, and ecological-function degradation. How the decline in habitat quality affects the health of benthic macroinvertebrate communities, and the quantitative relationship between the two, remain insufficiently understood, particularly in basins where water quality is generally good and water-chemistry factors no longer limit communities. Clarifying this relationship is essential for identifying ecological weaknesses and optimizing river restoration and watershed management. The Baipu River, a first-order tributary on the left bank of the Dongjiang River, is located in Zijin County, Heyuan City, Guangdong Province, China, with a main-channel length of 68.49 km and a drainage area of 445.42 km². The basin is mountainous, sparsely populated, and agriculture-dominated; owing to recent rural water-environment remediation, its water quality remains at Class II-III of the Chinese surface-water standard throughout the year. Twelve sampling sites were established, comprising eight impaired sites on the mainstem (intensive riparian development and frequent human disturbance) and four reference sites on forested, nearly pristine tributaries. Benthic macroinvertebrates were collected with a 60-mesh D-frame net in January 2023 during the dry season, sieved, fixed in 10% formalin, and identified mostly to the species or genus level. Habitat quality was assessed with ten indicators scored across four grades (good, fair, moderate, and poor). The Benthic-Index of Biotic Integrity (B-IBI) was constructed from 33 candidate metrics covering species richness, community composition, pollution tolerance, and feeding groups; core metrics were selected through range analysis, discriminative-power analysis (IQ ≥ 2), and redundancy analysis (|r| > 0.75), yielding five core metrics (M1, M17, M21, M25, and M27). A ratio-scoring method was applied, and health grades were defined relative to the 25th percentile of reference-site B-IBI values (excellent ≥ 3.82, good 2.86-3.82, moderate 1.91-2.86, poor 0.95-1.91, and very poor < 0.95). The results showed that reference sites had B-IBI values of 3.29-4.21 (good to excellent health) and habitat scores of 150-159, whereas impaired sites had B-IBI values of 1.76-3.30 (poor to good) and habitat scores of 107-133. Poor health accounted for 25% of impaired sites, and an independent-samples t-test confirmed that B-IBI significantly discriminated reference from impaired sites (P < 0.01). B-IBI showed an extremely significant positive correlation with the total habitat-quality score (r = 0.821, P < 0.01). Among the ten habitat factors, human-activity intensity (r = 0.832, P < 0.01), riparian land use (r = 0.774, P < 0.01), and river-channel alteration (r = 0.768, P < 0.01) were most strongly correlated with B-IBI, followed by vegetation diversity (ρ= 0.681, P < 0.05) and apparent water-quality status (ρ= 0.648, P < 0.05). These findings indicate that, in basins with generally good water quality, physical habitat quality is a dominant driver of benthic community health, and that human activities, operating through riparian land-use change and channel regulation, are the principal source of habitat degradation. It is recommended that biological-habitat restoration and management measures be incorporated into river regulation to sustain river ecological health. The study has limitations, including a relatively small sample size (n = 12), inherent differences between mainstem and tributary sites, and a single dry-season survey; multi-site, multi-season, and multi-year validation is therefore needed in future research.
Annual water balances can conceal monthly supply–demand mismatches and tail shortage risks in cities dependent on Yellow River water. Taking Shizuishan as a case, this study examines how dependence assumptions, tail-risk objectives, sectoral reliability constraints, and monthly allocation schedules affect water-resource investment portfolios and their supply–demand risk boundaries. A two-stage monthly water-allocation model was developed using public water-resources bulletins, statistical yearbooks, and official scheduling plans. Copula models fitted to the 2010–2023 sample generated joint scenarios of Yellow River water supply and agricultural and industrial activity factors. A total of 50,000 scenarios were reduced to 100 representative scenarios for optimization, while portfolio risks and sectoral reliability were evaluated using the full scenario set. Agricultural water saving and nominal annual reclaimed-water supply capacity were first-stage decisions. Monthly water allocation followed domestic, ecological, industrial, and agricultural priorities. Objectives based on the expected shortage rate and a weighted combination of the expected shortage rate and conditional value-at-risk (CVaR) were compared under a common budget. A separate cost-minimization problem imposed agricultural and industrial reliability thresholds of 75% and 95%, respectively. Reliability was defined as the probability of no shortage in any of the 12 months. Seven combinations of sample period, demand proxy, and dependence model were examined, followed by tests using alternative planned monthly allocations. Under the common annualized budget of CNY 37.88 million, all four combinations of independent/Gaussian dependence settings and the two risk objectives selected agricultural water saving of 0.36×10⁸ m³/a without reclaimed-water expansion. Sectoral reliability constraints brought reclaimed water into the portfolio. The portfolio comprises agricultural water saving of 0.59×108 m³/a and reclaimed-water expansion of 0.1095×108 m³/a had an annualized cost of approximately CNY 76.89 million. Under the Gaussian reference model, it reduced the expected shortage rate and CVaR at a confidence level of 0.90 from 1.92% and 10.39% without intervention to 0.56% and 5.32%, respectively. This portfolio met both sectoral thresholds in all seven statistical settings under the 2023 planned monthly allocation. Agricultural reliability declined to 74.45% under the 2025 allocation and remained below 75%, at 74.93%, when agricultural water saving increased to 0.604×108 m³/a. At baseline demand, the boundaries corresponding to expected shortage rates of 5% and 10% required scaling factors applied to the Yellow River water-supply distribution of at least 0.89 and 0.83, respectively. These thresholds increased to 0.98 and 0.92 when agricultural pressure increased by 10%. Sectoral reliability requirements can change investment choices even when dependence assumptions and tail-risk objectives select the same portfolio under a common budget. Engineering capacity should be determined jointly with irrigation timing and monthly supply schedules. The identified boundaries support comparisons of allocation plans and contingency water-supply plans under specified annual and monthly supply–demand conditions.
The Pinglu Canal, located in the southern Guangxi Zhuang Autonomous Region, is a major infrastructure project. The construction of the Pinglu Canal involves watershed excavation and channel widening, which have reshaped the local topography and drainage network relationships. These large-scale engineering activities, particularly channel excavation, have caused substantial changes in the local terrain. As the data acquisition dates of commonly used open-source global digital elevation models (DEMs) predate the construction of the canal, these datasets cannot accurately represent the newly formed terrain. This study used AW3D30, with a spatial resolution of approximately 30 m, as the base DEM and integrated it with engineering design information obtained from environmental impact assessment reports published by government agencies. Sentinel-1 and Sentinel-2 images acquired during canal-wide integrated wet commissioning following full water filling were also used. To delineate the canal water extent from Pingtangjiangkou to the Shajing Bridge over the Qinjiang River, a random forest classifier was trained using a total of 15 Sentinel-1 and Sentinel-2 bands and derived indices at a spatial resolution of 10 m. The samples were divided into 70% for training and 30% for validation. For areas without valid Sentinel-2 observations, a Sentinel-1-based classifier was used to fill the resulting gaps. Segment-specific designed riverbed elevations were then burned into AW3D30, followed by bank-slope smoothing to ensure a continuous transition at the channel boundary. Finally, the drainage network was extracted from the reconstructed DEM using a contributing-area threshold of 1 km². The main results were as follows: the overall classification accuracy was 93.33%, with a Kappa coefficient of 0.92, while the coverage rate for the independent water sample areas was 88.22%. Riverbed burning covered an area of approximately 17.17 km². The mean elevation decreases in the Pingtangjiangkou–Madao Hub, Madao Hub–Qishi Hub, Qishi Hub–Qingnian Hub, and Qingnian Hub–Shajing Bridge over the Qinjiang River sections were 17.08, 17.66, 17.71, and 11.77 m, respectively. The overall mean elevation decrease was 16.29 m, and the maximum decrease reached 99.70 m near the Madao Hub. Compared with the drainage network derived from the original DEM, the reconstructed network reduced unreasonable flow diversions and spatial offsets and more effectively represented the drainage relationships of the Pinglu Canal. Its horizontal positional errors also remained within an acceptable range. The confluence path from Pingtangjiangkou to the Shajing Bridge over the Qinjiang River had an overall sinuosity of 1.25. The sinuosity was 1.15 for the section from the Pingtangjiangkou to the Madao Hub, 1.16 from the Madao Hub to the Qishi Hub, 1.22 from the Qishi Hub to the Qingnian Hub, and 1.31 from the Qingnian Hub to the Shajing Bridge over the Qinjiang River. The results can provide fundamental data support for navigation safety and unified water resources regulation of the Pinglu Canal.
Accurate discharge forecasting in tidal river networks is challenging because river flow is jointly controlled by upstream runoff and ocean tides, resulting in pronounced periodicity, non-stationarity, flow reversal, and complex spatiotemporal interactions among interconnected channels. To improve the representation of periodic information in data-driven forecasting, this study proposes a multi-station discharge forecasting method based on a gated recurrent unit (GRU) network enhanced with Fourier temporal features. The lower tidal river network of the Dongjiang River is used as the study area. A one-dimensional hydrodynamic model is first established using the Hydrologic Engineering Center's River Analysis System (HEC-RAS) to generate continuous and physically consistent hourly discharge series at 235 computational stations. The hydrodynamic model is calibrated and independently validated against observed water levels at Shilong Hydrological Station. Based on the simulated discharge series, historical observations are organized using sliding time windows, while daily and semi-daily periodic information is explicitly encoded through sine and cosine Fourier features and combined with the original discharge inputs. The resulting sequences are then used by the GRU network for joint multi-station and multi-step forecasting. The results demonstrate that the GRU model substantially outperforms the conventional autoregressive integrated moving average (ARIMA) model. Under the representative forecasting configuration, the Nash–Sutcliffe efficiency (NSE) increases from approximately 0.00 for ARIMA to 0.77 for GRU, while the root mean square error (RMSE) and mean absolute error (MAE) decrease to 242.32 and 103.74 m³/s, respectively. Incorporating Fourier temporal features further improves the overall forecasting performance, reducing RMSE and MAE to 240.68 and 101.88 m³/s and increasing NSE to 0.78. Analysis across the 235 stations shows substantial spatial heterogeneity in forecasting skill, with a median station-level NSE of 0.605 and 61.7% of stations achieving NSE values of at least 0.5. Scatter-density analysis further indicates that predictive agreement gradually decreases with increasing forecast horizon, with coefficients of determination (R²) of 0.87, 0.83, and 0.77 for 1, 2, and 3 h forecasts, respectively.Further analysis reveals that model performance varies markedly among different discharge regimes. Both models perform relatively well under low-flow conditions, whereas the medium-flow range exhibits dynamic-range compression and the high-flow range shows larger errors and systematic underestimation. Fourier temporal features provide clear improvements under low- and high-flow conditions, particularly during high-flow periods, although their contribution is limited in the transitional medium-flow regime. Sensitivity experiments show that relatively short input windows of 6–12 h are more effective than longer historical sequences, while forecasting accuracy progressively deteriorates as the forecast horizon increases from 3 to 24 h. Amplitude analysis also demonstrates that the conventional GRU tends to underestimate discharge variability; Fourier temporal encoding partially alleviates this compression by improving the representation of tidal periodicity and phase information. Overall, the proposed framework provides an efficient approach for short-term multi-station discharge forecasting in tidal river networks and demonstrates that explicitly incorporating periodic temporal information can improve the representation of complex hydrodynamic processes characterized by the coexistence of periodic tidal forcing and non-stationary runoff disturbances.
