Cr(Ⅵ)has been identified as a significant contaminant in the environment,exhibiting properties of high toxicity,easy migration and strong carcinogenicity.These characteristics render it a substantial hazard to both water environment safety and human health.The efficient removal of Cr(Ⅵ)has consistently been a focal point in the field of environmental remediation.Biochar has been demonstrated to possess a good ability to adsorb heavy metal ions.The preparation and modification of biochar from agricultural waste is not only an effective approach to realize the resource utilization of agricultural waste,but also can effectively enhance the removal efficiency of Cr(Ⅵ)by biochar.In this study,peanut shell and sesame straw were utilised as raw materials to prepare original biochar,and consequently SF-HBC and PF-ZBC were prepared by composite modification methods of"ferric chloride+sodium alginate"and"ferric chloride+chloroapatite",respectively.Then,the experiments were conducted to investigate the impacts of biochar type,pH value and biochar dosage on the adsorption efficiency of Cr(Ⅵ)in water.Combined with the theories of adsorption kinetics,adsorption isotherms and adsorption thermodynamics,the adsorption processes of Cr(Ⅵ)by two kinds of composite-modified biochars(SF-HBC,PF-ZBC)were analyzed.At the same time,by comprehensively applying characterization methods such as SEM,FTIR,XRD and XPS,the adsorption mechanisms and differences of Cr(Ⅵ)in water by SF-HBC and PF-ZBC were analyzed.The research results showed that acidic conditions were more conducive to the adsorption of Cr(Ⅵ)in water by SF-HBC and PF-ZBC.In comparison with PF-ZBC,SF-HBC exhibited a broader pH application range.The removal efficiency of Cr(Ⅵ)by SF-HBC in water maintained at 63%at pH=8,which was significantly higher than that of PF-ZBC(17.5%).The adsorption processes of Cr(Ⅵ)by SF-HBC and PF-ZBC both conformed to the pseudo-second-order kinetic model and the Langmuir isothermal adsorption model,indicating that the adsorption processes of Cr(Ⅵ)by the two were mainly chemical adsorption of monolayer and were spontaneous endothermic processes.A comparison was made for the pseudo-second-order kinetic parameter k2 and the Langmuir isothermal adsorption parameter KL.It was determined that SF-HBC exhibited a faster adsorption rate,while PF-ZBC demonstrated a stronger binding force with Cr(Ⅵ),indicating that there were significant differences in the adsorption behaviors between the two.The mechanism study demonstrated that the adsorption mechanism of Cr(Ⅵ)by SF-HBC was a multifaceted process involving physical adsorption,redox and complexation.The adsorption mechanism of Cr(Ⅵ)by PF-ZBC was found to be physical adsorption,redox,complexation and co-precipitation.It was determined that the adsorption of Cr(Ⅵ)by SF-HBC was predominantly governed by the complexation of organic functional groups and redox reactions,while PF-ZBC mainly exhibited a more efficacious removal of Cr(Ⅵ)through the combination of redox,complexation,and phosphate co-precipitation mechanisms.A significant disparity was observed in the adsorption mechanism of Cr(Ⅵ)between SF-HBC and PF-ZBC.The study findings can provide theoretical basis and technical reference for the efficient resource utilisation of agricultural waste and the efficient treatment of Cr(Ⅵ),which is of positive significance for promoting the treatment of heavy metal pollution and the development of green environmental functional materials.
To clarify the long-term variation of ozone pollution and its driving mechanisms in northern Henan,this study used air pollution observations from Anyang,Xinxiang,Jiaozuo,Puyang,and Hebi during 2015 to 2024,together with ERA5 meteorological reanalysis data and OMI satellite tropospheric column data as study data.A random forest based meteorological normalization method was applied to quantify the contributions of anthropogenic emissions and meteorological conditions to changes in the daily maximum 8 h average ozone concentration,or MDA8-O3.The spatiotemporal evolution of ozone formation sensitivity was further analyzed by using the HCHO/NO2 column ratio.The results showed that MDA8-O3 in northern Henan increased continuously during 2015 to 2024,with a growth rate of 2.2 to 4.3 μg/(m3·a).The number of exceedance days also increased clearly.During the same period,surface NO2 concentration and tropospheric NO2 column concentration decreased significantly,with annual mean declines of 5.7%to 8.1%and 4.8%to 7.2%,respectively.In contrast,the HCHO column changed only slightly.This indicated that NOx emission reduction was effective,while VOCs control lagged behind.The random forest model showed good performance.Meteorological factors explained 80.6%to 86.7%of ozone variability.Solar radiation,air temperature,and boundary layer height were the main meteorological drivers.Regional transport also played an important role in ozone enhancement.The meteorological normalization results showed that emission changes contributed 55.9%to 75.8%of the increasing ozone trend and were the dominant cause of ozone growth,while meteorological conditions played an overall promoting role.A stage-by-stage analysis showed that both emission-related and meteorology-related ozone increased faster during 2015 to 2018 than during 2019 to 2024.This suggested that the recent slowdown in ozone growth resulted from the combined effects of emission control measures and meteorological conditions.As NOx emissions continued to decline,the HCHO/NO2 column ratio increased by 0.10 to 0.15 per year.Ozone formation sensitivity gradually shifted from a VOCs-limited regime to a transitional or NOx-limited regime.These results provide a scientific basis for coordinated ozone control and differentiated emission reduction in northern Henan.
In this study,the leading modal characteristics of precipitation during the autumn flood season and their relationship with sea surface temperature(SST)anomalies in the Yellow River Basin are analyzed using statistical methods.Precipitation data from 319 meteorological stations in the Yellow River Basin for the autumn flood season(September-October)from 1961 to 2022,NCEP/NCAR reanalysis circulation data,and NOAA linearly optimum Interpolation(OIV)global SST data are employed.The results show that precipitation during the autumn flood season in the basin exhibits prominent interdecadal variability and is currently embedded in an above-average interdecadal background.The leading mode of autumn flood precipitation features a spatially consistent pattern of above-or below-average precipitation,with a variance contribution rate of 45.81%.This leading mode displays alternating interannual and interdecadal variations:from the 1960s to the mid-1980s and after 2000,it is dominated by quasi-2-3-year cycles with notable interannual variability;from the mid-1980s to the 1990s,it is mainly characterized by interdecadal changes.Circulation analyses indicate that the leading mode of basin precipitation is mainly modulated by the intensity and meridional displacement of the western Pacific subtropical high at 500 hPa,as well as the corresponding anomalous cyclonic/anticyclonic circulation near the Japanese Islands at 850 hPa.Such anomalous circulation patterns are likely induced by anomalous vertical motion and meridional circulation over the tropical western Pacific,which are triggered by anomalous SST patterns over the tropical Pacific.It is further revealed that the leading mode of autumn flood precipitation in the basin is closely linked to the evolution and persistence of SST anomalies over the equatorial central-eastern Pacific from the preceding summer to the following autumn.When the equatorial central-eastern Pacific is uniformly warm(cold),precipitation during the autumn flood season in the basin tends to be below average(above average).
The stability of roadways in underlying coal seams is significantly affected by residual coal pillars in overlying seams.This study investigates the reasonable layout and support technology of roadways under the influence of overlying coal pillars.Based on limit equilibrium theory and elastic mechanics theory,the stress transfer in the floor beneath residual coal pillars and the lateral stress distribution of the goaf are analyzed.A bimodal stress distribution caused by stress superposition is revealed.FLAC3D numerical simulation is used to compare the stress distribution,plastic zone evolution,and stability of surrounding rock under different coal pillar widths.A corresponding support scheme is proposed.The results show that the elastic core width of the 30 m residual coal pillar is 22.26 m.Obvious stress concentration occurs in the 4-1 coal seam.The vertical stress beneath the coal pillar remains 20.80 MPa at a depth of 47 m.The superimposed stress reaches a peak value of 26.85 MPa due to the lateral stress from the underlying goaf,which has a significant negative effect on roadway stability.Theoretical analysis indicates that the minimum width of the protective coal pillar is 10.09 m.Combined with numerical simulation,11 m is determined as the optimal width considering both stability and resource recovery.Field monitoring results show that the maximum deformation of the roof and floor is 181 mm,and that of the two sides is 153 mm after adopting a bolt-cable support system.The deformation is well controlled and meets the requirements of safe production.The findings provide a reference for roadway layout and support design under similar conditions.
Accurate extraction of the spatial distribution information of agricultural greenhouses,a vital component of modern agriculture,is of significant importance for agricultural management and environmental monitoring.This study took Yaojia Town in Zhongmu County,Zhengzhou City as the research area,utilized high-resolution GF-2 satellite imagery acquired on January 4,2024 as the primary data source,and investigated the capability of different machine learning models for agricultural greenhouse extraction.Four classification models were constructed:convolutional neural network(CNN),random forest(RF),support vector machine(SVM),and an optimized convolutional neural network incorporating an attention mechanism(CBAM-CNN).These models were applied to extract agricultural greenhouse information from the study area.The classification results from each method were qualitatively and quantitatively compared and analyzed using visual interpretation data and confusion matrices.The results indicated that the CBAM-CNN model achieved the best extraction performance,with an overall accuracy of 94.26%and a Kappa coefficient of 91.24%.The resulting classified patches were complete with clear boundaries,demonstrating the strongest robustness in suppressing salt-and-pepper noise and background interference.The CNN and RF models followed,with overall accuracies of 91.46%and 86.24%,respectively,achieving good extraction results but slightly inferior to CBAM-CNN.The SVM model yielded the lowest overall accuracy at 80.25%,with notable misclassifications in its results.This study validates the effectiveness and superiority of deep learning models incorporating attention mechanisms for the high-precision extraction of agricultural greenhouses from high-resolution remote sensing imagery,providing a scientific methodological foundation for intelligent monitoring and management of agricultural resources.
To elucidate the long-term evolution characteristics and stage-specific change patterns of land use/land cover(LUCC)in typical coastal small and medium-sized cities,and to support refined territorial spatial management,this study took Beihai City as a case study.Based on long-term Landsat remote sensing imagery from 2000 to 2023,land use/land cover classification was conducted using a random forest algorithm on the Google Earth Engine(GEE)platform,and spatio-temporal evolution characteristics were analysed using land use transition matrices.The results indicated that:① The classification accuracy remained consistently high across all periods,with overall accuracy(OA)exceeding 0.88 and Kappa coefficients ranging from 0.87 to 0.90;② Construction land expanded from the urban core and major transportation corridors towards coastal zones and peripheral areas,with the spatial pattern evolving from a single-centre structure to a multi-polar configuration,accompanied by a continuous reduction in cultivated land and the encroachment of certain ecological land types in urban expansion zones;③ From 2000 to 2023,construction land and forest land increased by 73.75 km² and 396.77 km²,respectively,while cultivated land decreased by 458.99 km².Specifically,the period from 2000 to 2010 was characterised by ecological structure adjustment,urban expansion accelerated markedly during 2010-2020,and land use transition intensity weakened significantly after 2020,with construction land shifting from extensive outward expansion to internal optimisation.Overall,the land use pattern had become increasingly stable and intensive.The integration of the GEE platform with machine learning methods demonstrates strong applicability for long-term LUCC monitoring in coastal cities,providing reliable technical support for land resource management and territorial spatial planning.
Henan Province,one of China's most important core grain-producing regions,has experienced profound land-use transformation under the combined influence of rapid urbanization and ecological civilization construction.As a key agricultural province undergoing simultaneous pressures from cultivated land protection,urban expansion,and ecological restoration,Henan provides a representative case for understanding the spatiotemporal evolution of regional land-use patterns and their driving mechanisms.Taking Henan Province as the study area,this study employed the long-term China Land Cover Dataset(CLCD)from 2000 to 2023 and selected multiple time nodes to analyze land-use changes over the past 23 years.By integrating a land-use transition matrix,centroid migration analysis,elevation-gradient statistics,and the geographic detector model,this study systematically quantified the structural transformation,spatial differentiation,and dominant drivers of land use in Henan Province.The results showed that the land-use structure of Henan Province underwent significant reorganization during 2000-2023,characterized by continuous cropland shrinkage,rapid impervious-surface expansion,and steady forest growth.Cropland,as the dominant matrix landscape,showed a net decrease of 9 604.96 km²,while impervious surfaces and forest land increased by 49.16%and 11.15%,respectively.Land conversion exhibited marked asymmetry.The large-scale one-way conversion from cropland to impervious surfaces was the primary cause of cropland loss,indicating the rigid occupation of high-quality cultivated land by urbanization and infrastructure development.In contrast,forest expansion mainly resulted from the conversion of cropland and grassland,reflecting the long-term effects of ecological restoration and vegetation recovery policies.These results reveal an evident dual process of"urban encroachment"and"ecological compensation"in regional land-use transition.Distinct centroid migration trajectories were observed among different land-use types,indicating clear spatial heterogeneity.The centroid of cropland shifted 2.139 km toward the northeast,suggesting the influence of agricultural restructuring and farmland redistribution.Impervious surfaces showed the smallest migration distance,only 0.374 1 km,reflecting the relative spatial stability and agglomeration of the Central Plains urban system.In contrast,bare land exhibited the largest centroid migration distance,reaching 14.334 km,implying strong sensitivity to ecological restoration and land remediation processes.Meanwhile,elevation-gradient analysis further indicated that cropland stability generally declined with increasing elevation,whereas forest land showed greater persistence in higher-altitude areas.The geographic detector results demonstrated that socioeconomic factors gradually became the dominant drivers of land-use spatial differentiation in Henan Province.The explanatory power of human-related factors,including the proportions of the secondary and tertiary industries,total population,and the industrial upgrading index,increased continuously over time.Among them,the proportion of the secondary industry consistently played the leading role in impervious-surface expansion,with its q value increasing from 0.887 in 2000 to 0.910 in 2023.By contrast,the influence of natural factors such as annual mean temperature and annual precipitation generally weakened,although they still exerted stage-specific regulatory effects on certain vegetation-related land-use types.Overall,this study reveals the dynamic trade-offs among food security,urban development,and ecological conservation during land-use transformation,and provides a scientific basis for optimizing territorial spatial planning and promoting sustainable regional development.
Slopes containing weak interlayers are prone to instability and sliding under prolonged rainfall,and a deeper understanding of their mechanical response mechanisms and sliding evolution processes is therefore required.Taking the Qingmingshan landslide area as a case study,nine groups of remolded weak interlayer specimens with a water content of 20%were prepared.Repeated direct shear tests,direct shear tests with different resting times,and scanning electron microscopy(SEM)observations were conducted to systematically investigate the evolution of the shear behavior of weak interlayer materials under rainfall infiltration conditions.In addition,a rainfall infiltration numerical model of a slope containing a weak interlayer was established using the finite element method to analyze the evolution of slope displacement and plastic strain under different rainfall intensities and durations.The results indicate that the shear behavior of the weak interlayer gradually changes from peak-strength-controlled behavior to residual-strength-controlled behavior.After repeated shearing,particles within the shear band become preferentially aligned along the shear direction and the shear surface becomes smoother,revealing the microstructural mechanisms responsible for shear strength degradation.The resting time mainly affects the shear strength level but has a limited influence on the shear failure mode.With increasing resting time,the structural contribution to shear resistance gradually decreases,and the shear response tends to be dominated by sliding behavior.Numerical simulations further demonstrate that the spatial position of the weak interlayer and its preferential seepage-path characteristics play a dominant role in controlling slope deformation and plastic evolution.These findings provide a reference for the stability assessment and mitigation of slopes containing weak interlayers under rainfall conditions.
Insufficient lighting at subway construction sites often leads to performance degradation of traditional computer vision algorithms,resulting in elevated false positive and miss rates.Addressing the prevalent issues of low-Light,high noise,and insufficient contrast in such environments,this paper proposes a helmet detection method based on EnlightenGAN and an improved YOLOv8.First,EnlightenGAN is employed for image preprocessing.This unsupervised generative adversarial network adaptively enhances low-light images,effectively preserving structural details and geometric consistency while significantly improving overall brightness and contrast.Subsequently,a LightSE module is integrated after the C2f module in the YOLOv8 backbone network.This module adaptively recalibrates brightness-related information within the feature layers and emphasizes edge and texture responses in dark scenes,thereby enhancing the discriminative power and robustness of feature representations.Experiments were conducted on a dataset comprising 2315 images collected from subway construction sites in Q City.Comparative evaluations involving combinations of YOLOv8,EnlightenGAN,and LightSE demonstrate that the proposed model achieves optimal performance across multiple metrics.Specifically,the Precision,Recall,and mAP@0.5 reached 92.66%,86.56%,and 92.57%,respectively,representing improvements of 5.95%,4.92%,and 4.61%over the origial YOLOv8.The results indicate that the dual-stage synergistic strategy comprising EnlightenGAN adaptive enhancement and LightSE-YOLOv8 optimized detection significantly mitigates the issue of missed detections under low-light conditions.The method maintains high detection accuracy and robustness without a significant increase in false positives,while ensuring excellent real-time performance(127.3 fps).These research findings offer substantial application value for the intelligent safety management of typical low-light scenarios,including nighttime operations,tunnel construction,and underground worksites.
To investigate the strength degradation behavior of flowable solidified soil under the coupled effects of salt erosion and freeze-thaw cycles,flowable solidified soil specimens were prepared using cement blended with slag and phosphogypsum.Laboratory tests were conducted to simulate freeze-thaw cycles in fresh water,sodium chloride solution,and sodium sulfate solution.The strength evolution,mass variation,and deterioration characteristics of different solidification systems were analyzed.The results show that with increasing freeze-thaw cycles,the compressive strength of all specimens exhibits a three-stage degradation pattern,including slow reduction,rapid reduction,and gradual stabilization.The incorporation of slag significantly improves resistance to salt erosion and freeze-thaw damage.After ten freeze-thaw cycles,the strength retention rate of slag modified specimens ranges from 36%to 54%,while the mass change rate is between 0.25%and 0.60%,indicating the best durability performance.Under coupled salt erosion and freeze-thaw conditions,the cement slag system experiences greater strength loss in sodium sulfate solution than in sodium chloride solution,with an increase of 28%to 39%.In contrast,the cement phosphogypsum system shows better resistance in the sodium sulfate environment.Based on the experimental results,a strength degradation model suitable for coupled salt erosion and freeze-thaw conditions is established through regression analysis.The proposed model can be used to predict long term strength evolution of flowable solidified soil in harsh environments and provides reference for engineering applications in cold and saline regions.
The secure,stable,and reliable operation of power transmission systems depends on the real-time perception and precise detection of defects in critical components.Traditional manual inspection methodologies are frequently characterized by low operational efficiency,high economic costs,and the presence of significant safety risks.In stark contrast,automated inspection technology based on unmanned aerial vehicles(UAVs),leveraging its superior advantages of high safety standards and low operational costs,has successfully emerged as the mainstream direction in the field of electric power operation and maintenance.However,in the context of practical engineering applications,the images acquired by drones from high-altitude aerial perspectives frequently encounter substantial challenges,such as the extremely diminutive scale of targets and highly complex background environments.When processing such challenging tasks,existing mainstream general-purpose target detection algorithms are severely limited by the continuous strided convolution downsampling mechanisms inherent in their backbone networks.This structural limitation inevitably leads to a situation where the fine-grained features of tiny defects such as insulator cracks and corrosion are excessively compressed or even completely lost within deep feature maps,thereby directly resulting in missed detections.To effectively address and mitigate this specific problem,this paper proposes an improved YOLOv8 detection model based on a collaborative enhancement strategy.This model constructs a structural coupling mechanism that integrates frontend feature fidelity with backend fine-grained parsing within the multi-scale feature fusion network.To be specific,at the frontend of the network architecture,the non-destructive downsampling module,known as SPD-Conv,is introduced to replace the traditional strided convolution layers.By mapping information from the spatial dimension non-destructively directly into the channel dimension,this module is able to meticulously preserve the fine-grained features of tiny defects while simultaneously reducing the resolution.At the backend of the network,a dedicated high-resolution detection head is designed and incorporated.This additional detection head utilizes the bidirectional fusion capabilities of the feature pyramid network and the path aggregation network to specifically parse and reconstruct the detailed information that has been preserved by the frontend.This design is explicitly intended to achieve the end-to-end,precise capture of tiny targets through the synergistic effect generated by the collaboration of both components.Extensive empirical experiments conducted on the VisDrone2019 benchmark dataset demonstrate that the proposed model exhibits superior comprehensive performance.Compared with the baseline YOLOv8 model,the improved model achieves substantial improvements of 3%,1.57%,and 1.54%in precision,recall,and mean average precision(mAP),respectively,ultimately reaching an mAP of 38.75%.Furthermore,with only about 4.24 million parameters,its detection accuracy outperforms mainstream models such as TIB-Net,YOLOv5s,and YOLOv7,while the inference speed reaches 233 FPS.Featuring both low parameter count and high frame rate,it meets the practical needs for deployment on edge devices.In conclusion,this collaborative enhancement strategy can effectively boost the detection performance for tiny defects in power systems,possessing strong and robust potential for practical engineering applications.
The precise prediction,evaluation,and improvement of tunnel blasting effects represent the core topic of safe and efficient construction in underground engineering.With the deep integration of numerical simulation technology,intelligent algorithms,and new material technologies,the prediction of tunnel blasting effects has been significantly developing toward refinement and intelligence.Important improvements have been achieved in the prediction accuracy of key indicators,such as blasting fragment size distribution,peak vibration control,and surrounding rock damage range assessment.Precise prediction,evaluation,and improvement are of fundamental significance for optimizing blasting design,improving construction quality,ensuring project safety,and controlling costs.The study focused on core evaluation indicators,including blasthole residual hole rate,overbreak and underbreak control,profile flatness,and blasting advance.It systematically reviewed the latest research progress on prediction,evaluation,and improvement technologies for tunnel blasting effects at home and abroad.The scientific guiding value of numerical simulation methods and artificial intelligence prediction technologies in blasting parameter optimization was deeply analyzed.By comparing the performance characteristics and applicable conditions of different prediction models and combining with typical engineering cases,the internal mapping rules between blasting parameter adjustment and effect improvement were clarified.The significant effect of key parameter optimization on the improvement of blasting quality was verified.On this basis,the key technical development directions for intelligent blasting were put forward,providing path guidance for improving the blasting control level under complex geological and construction environments.
Against the backdrop of the accelerating integration of the digital economy and the real economy,together with the continuous advancement of the"dual carbon"goals,exploring how intelligent transformation promotes corporate green upgrading and productivity improvement has become a crucial issue for achieving high-quality economic development.As a strategic emerging industry supporting the green and low-carbon transition,the photovoltaic(PV)sector is characterized by high technological intensity,substantial capital investment,and a strong environmental orientation.These distinctive features make it an ideal empirical context for examining the economic consequences of intelligent transformation.Based on panel data of Chinese A-share listed photovoltaic enterprises from 2011 to 2024,this study constructs a comprehensive measurement system for intelligent transformation from two dimensions:intelligent investment and intelligent technology application.Specifically,the measurement incorporates firms'tangible investment in intelligent software and hardware assets,and further extracts detailed indicators of intelligent technology adoption through precise keyword identification and in-depth semantic analysis of annual reports.The entropy method is employed to synthesize these indicators into a composite index of intelligent transformation.Meanwhile,firm-level total factor productivity(TFP)is estimated using the Levinsohn-Petrin(LP)method,which effectively mitigates endogeneity concerns arising from simultaneity bias and measurement errors.On this basis,two-way fixed effect models,instrumental variable approaches,and mechanism analysis models are applied to systematically examine the impact of intelligent transformation on the TFP of photovoltaic enterprises.The empirical results demonstrate that intelligent transformation significantly enhances the total factor productivity of PV firms.Mechanism analysis reveals that green technological innovation plays both mediating and moderating roles in the relationship between intelligent transformation and TFP.On the one hand,intelligent transformation promotes green R&D activities and accelerates the application of environmentally friendly technologies,thereby indirectly improving productivity through green innovation.On the other hand,higher levels of green technological innovation strengthen the positive effect of intelligent transformation on productivity,suggesting that the synergistic development of"intelligence+greenness"constitutes a key pathway for achieving efficiency upgrading.Heterogeneity analysis further indicates that the productivity-enhancing effect of intelligent transformation varies across different contexts.From the perspective of ownership structure,the positive effect is more pronounced among non-state-owned enterprises than state-owned enterprises.From the viewpoint of value chain segmentation,firms located in the midstream segment experience greater productivity gains from intelligent transformation compared with upstream and downstream firms.In terms of policy environment,the positive impact of intelligent transformation on TFP is stronger before the phase-out of government subsidies than afterward.Overall,this study uncovers the dual mechanism of green technological innovation in the process of intelligent transformation and extends the theoretical boundaries of research on the integration of digitalization and green development.The findings provide differentiated empirical evidence for policy formulation and enterprise practice across ownership types,value chain positions,and policy cycles.
To address the prevalent nonlinear and non-stationary characteristics in power load forecasting and the insufficient modeling of meteorological coupling effects and multi-scale temporal dependencies in existing algorithms,a feature engineering framework integrating the"meteorology-load"coupling mechanism and multi-scale temporal features is proposed and applied to a random forest regression model.First,two product terms,i.e."temperature×humidity"and"rainfall×average temperature",are constructed to explicitly capture the surge in cooling demand caused by high temperature and high humidity,as well as the reverse regulatory effects of rainfall on load under different temperature conditions.Second,1-to 7-day load lag features are introduced to capture weekly periodicity,while 1-to 3-day rolling average temperatures are built to reflect short-term cumulative thermal effects.Finally,fixed-window rolling statistics are employed to extract seasonal temperature inertia,avoiding model lag in response to load mutations during seasonal transitions.All features are normalized via Min-Max scaling before being fed into the random forest model for training and prediction.Experimental results on a public dataset show that the proposed method achieves a mean absolute error(MAE)of 28 293.29 kWh,a root mean square error(RMSE)of 45 921.49 kWh,a mean absolute percentage error(MAPE)of 5.03%,and a mean absolute scaled error(MASE)of 0.18%.Compared with classic models such as XGBoost,DNN-Attention-LSTM,LSTM,and ARIMA,the proposed method achieves the best results across all four evaluation metrics,validating the effectiveness of the feature engineering design.This study provides a feature construction approach that balances physical mechanisms and data characteristics,with practical significance for improving load forecasting accuracy.
The construction of highway tunnels consumes a large amount of resources and energy,resulting in substantial greenhouse gas emissions.To achieve the goals of"low carbon emission"and even"zero carbon emission"in the tunnel engineering field,it is necessary to carry out carbon emission prediction for tunnel construction.This study takes the Ludian-Qiaojia expressway tunnel project in Yunnan Province as an example,focusing on the prediction of carbon emissions in the physical construction stage of the tunnel,aiming to improve the accuracy and stability of the existing prediction methods.Based on clarifying the boundaries and methods of carbon emission calculation,the carbon emissions during the physical construction phase of the case project tunnel are calculated using the emission coefficient method,and the calculation results are used as a dataset to construct the BPNN prediction model.Four parameters that are easy to obtain and significantly related to carbon emissions in the physical construction phase of tunnel,namely surrounding rock grade,tunnel burial depth,surrounding rock quality and special geological conditions,are selected as input features.Spearman correlation analysis verifies the science and objectivity of these four parameters as input features.To address the difficulty of accurately modeling the nonlinear mapping relationship between carbon emissions in the tunnel physical construction phase and geological parameters,the SSA algorithm with the advantages of strong optimization ability,fast convergence and few parameters is innovatively introduced to optimize the initial weight and threshold of BPNN,and the SSA-BPNN model is established.The model performance evaluation results show that the R2 of the SSA-BPNN model can reach 0.974,and the fit effect is good.Compared with the single BPNN model and the BPNN models optimized by other intelligent optimization algorithms,SSA-BPNN model has the highest prediction accuracy and the lowest error,proving that the SSA-BPNN model is the optimal model in this case and also shows its superiority in the prediction of carbon emissions in the tunnel physical construction stage.This study provides an effective technical tool for the decision-making of low-carbon tunnel alignment.
