Reservoir Inflow Forecasting Using Ensemble Models Based on Neural Networks, Wavelet Analysis and Bootstrap Method

To make full use of inflow forecasts with different lead times, a new reservoir operation model that considers the long-, medium- and short-term inflow forecasts (LMS-BSDP) for the real-time operation of hydropower stations is presented in this paper. First, a hybrid model, including a multiple linear regression model and the Xinanjiang model, is developed to obtain the 10-day inflow forecasts, and ANN models with the circulation indexes as inputs are developed to obtain the seasonal inflow forecasts. Then, the 10-day inflow forecast is divided into two segments, the first 5 days and the second 5 days, and the seasonal inflow forecast is deemed as the long-term forecast. Next, the three inflow forecasts are coupled using the Bayesian theory to develop LMS-BSDP model and the operation policies are obtained. Finally, the decision processes for the first 5 days and the entire 10 days are made according to their operation policies and the three inflow forecasts, respectively. The newly developed model is tested with the Huanren hydropower station located in China and compared with three other stochastic dynamic programming models. The simulation results demonstrate that LMS-BSDP performs best with higher power generation due to its employment of the long-term runoff forecast. The novelties of the present study lies in that it develops a new reservoir operation model that can use the long-, medium- and short-term inflow forecasts, which is a further study about the combined use of the inflow forecasts with different lead times based on the existed achievements.

Water is considered as an essential commodity in human society. It has been observed that the basic standard of living of any Country can be measure through the effective portable water supply for consumption, sanitation and hygiene for the citizen. Its production and supply in the Federal Capital Territory (FCT), Abuja, Nigeria is a challenge, as dominant percentages of lower- class workforce live outside the FCTWB existing water network, and depend on borehole or tanker for water. The demand on the water supply system in FCT, Abuja has exceeded its initial design population, and the system is incomplete and operating with only 720 million liter of water per day (MLD) functioning plant. On this basis, this study seeks to identify the best model that fit daily water supply data in FCT, Abuja and to forecast the future daily water supply in this area. The comparing models were artificial neural, multiple linear and non-linear regression model were applied to the data set of daily water production, daily water treated, daily water supply, daily water leakage, Number of household connected to Federal Capital Territory Water board water network, Number of Commercial area connected to Federal Capital Territory Water board water network, and the Estimated population of Federal Capital Territory ,Abuja from (2013 to 2020), and it was emerged that the most adequate model for the data was Multiple linear regression model (MLR), with (R2 =1.000), (RMSE = 0.00282), and (MAE = 6.540). Artificial neural network records the high value of (RMSE = 432.4321), (MAE = 549.444) and (R2 =0.921), and multiple nonlinear regression record the highest value of (RMSE = 806.687), (MAE = 666.421) and (R2 = 0.999). From the statistical measure of goodness of fit, the three models compared results indicate that MLR model forecast are closely matching with actual observations values, artificial neural network model over-forecast some days volume of daily water supply by FCTWB, cooperation and multiple nonlinear regression model under-forecast some days volume of daily water supply by FCTWB, corporation. The three models compared depicts the influential variables of FCTWB Corporation daily water supply, (ANN = DWP, DWT), (MLR = DWT, DWL) and (MNLR = DWP, DWL), and their forecast graphs shows downward daily water supply in FCT, Abuja. It is recommended that the FCT, Abuja water Board Corporation should consider multiple linear regression (MLR) model and its forecasted values as a road map in its operations and planning toward water supply sustainability in FCT, Abuja.

Objectives : In this study, a performance prediction model for a pilot-scale VOC adsorption column was developed using ANN algorithm. We compared the prediction accuracy of the mathematical models (Thomas model and Yan model) and the multiple linear regression model with that of ANN. This study showed the applicability of the ANN model for predicting the performance of activated carbon columns.Methods : The adsorption module contained 79.8 kg/module of wood-based activated carbon. The gas with 800 ppm-THC of toluene flowed downward from the top at about 5,700 m<sup>3</sup>/h. The breakthrough point was taken as 200 ppm-THC, the same as VOC emission regulation. The desorption was carried out using 130 m<sup>3</sup>/h of hot gas flowing upwards with reduced pressure (-150 to -200 mbar) and high heat (170℃). Adsorption and desorption cycles were conducted 6 times using 3 batches of activated carbon modules. Thomas model, Yan model, multiple linear regression model, and ANN model were developed to predict the breakthrough of <i>C<sub>out</sub>/C<sub>in</sub></i> .Results and Discussion : The Thomas model and the Yan model provided the R<sup>2</sup> values of 0.25 and 0.28, respectively, for predicting the <i>C<sub>out</sub>/C<sub>in</sub></i> of all adsorption module batches and cycles, and the prediction accuracies were low. This could be because these two models do not consider temperature and pressure change operating conditions in the models. Also, the prediction accuracy of <i>C<sub>out</sub>/C<sub>in</sub></i> was low when the initial inlet concentration and flow rate conditions were different for each batch. The multiple linear regression model considers all operating factors in the model, but the prediction accuracy of <i>C<sub>out</sub>/C<sub>in</sub></i> was low as R<sup>2</sup> of 0.45. On the other hand, the ANN model predicted the <i>C<sub>out</sub>/C<sub>in</sub></i> with R<sup>2</sup> higher than 0.97 for all adsorption module batches. In particular, even with the non-ideal data, the ANN model derived a breakthrough of <i>C<sub>out</sub>/C<sub>in</sub></i> close to the experimental value.Conclusion : The ANN model provided high prediction performance for the breakthrough of <i>C<sub>out</sub>/C<sub>in</sub></i> even under non-ideal operation conditions and was expected to be helpful for actual THC adsorption column operation. The accuracy of the ANN model will be further improved if data are accumulated under various conditions.

The aim of this study is to assess the possibility of forecasting water level fluctuations in a relatively small (<100 km2), post-glacial lake located in a temperate climate zone by means of artificial neural networks and multiple linear regression. The area of study was Lake Serwy, located in northeastern Poland. Two artificial neural network (ANN) multilayer perceptron (MLP) and multiple linear regression (MLR) models were built. The following explanatory variables were considered: maximal and minimal temperature (Tmax, Tmin) wind speed (WS), vertical circulation (VC) and water level from previous periods (WL). Additionally, a binary variable describing the period of the year (winter, summer) has been considered in one of the two MLP and MLR models. The forecasting models have been assessed based on selected criteria: mean absolute percentage error (MAPE), root mean squared error (RMSE), coefficient of determination (R2) and mean biased error. Considering their values and absolute deviations from observed values it was concluded that the ANN model using an additional binary variable (MLP_B+) has the best forecasting performance. Absolute deviations from observed values were the determining factor which made this model the most efficient. In the case of the MLP_B+ model, those values were about 10% lower than in other models. The conducted analyses indicated good performance of ANN networks as a forecasting tool for relatively small lakes located in temperate climate zones. It is acknowledged that they enable water level forecasting with greater precision and lower absolute deviations than the use of multiple linear regression models.

Accurate reservoir inflow forecasting is vital for effective water resource management. Reliable forecasts enable operators to optimize storage and release strategies to meet competing sectoral demands—such as water supply, irrigation, and hydropower scheduling—while also mitigating flood and drought risks. To address this need, in this study, we propose a novel time-variant encoder–decoder (ED) model designed specifically to improve multi-step reservoir inflow forecasting, enabling accurate predictions of reservoir inflows up to seven days ahead. Unlike conventional ED-LSTM and recursive ED-LSTM models, which use fixed encoder parameters or recursively propagate predictions, our model incorporates an adaptive encoder structure that dynamically adjusts to evolving conditions at each forecast horizon. Additionally, we introduce the Expected Baseline Integrated Gradients (EB-IGs) method for variable importance analysis, enhancing interpretability of inflow by incorporating multiple baselines to capture a broader range of hydrometeorological conditions. The proposed methods are demonstrated at several diverse reservoirs across the United States. Our results show that they outperform traditional methods, particularly at longer lead times, while also offering insights into the key drivers of inflow forecasting. These advancements contribute to enhanced reservoir management through improved forecasting accuracy and practical decision-making insights under complex hydroclimatic conditions.

In this study, we developed and evaluated a hybrid framework for reservoir inflow forecast. This framework is unprecedented, which integrates new quasi-globally available observation-, satellite-, or model-based datasets using machine learing models to forecast inflow at the local scale. Under this framework, we compared random forests, gradient boosting machine, extreme learning machine, M5-cubist, elastic net, as well as their multi-model ensemble using Bayesian model averaging (BMA), and tested contributions from different input datasets, including retrospective forecast (reforecast) from florecast oriented low ocean resolution (FLOR) dynamic climate model, satellite-based hydrologic products, and climate phenomenon information. The performance was evaluated using Kling–Gupta efficiency (KGE) and correlation coefficient (R) in two headwater reservoirs, Harris reservoir in the humid Alabama–Coosa–Tallapoosa river basin and the Navajo reservoir in the arid Upper Colorado River Basin (UCRB). The results showed that for the Harris reservoir, the BMA combining five models with antecedent inflow and satellite-based hydrologic information as model inputs provided the best performance (KGE = 0.66, R = 0.76). For the Navajo reservoir, the gradient boosting machine model with all variables combined as input showed the best performance (KGE = 0.76, R = 0.83). Satellite-based soil moisture and evaporation consistently showed significant contributions to the inflow forecast. Benefits from climate indices and FLOR reforecast varied by locations, with more benefits coming from climate indices than FLOR potential evaporation reforecast at the Navajo reservoir in UCRB. Given the global coverage of the model inputs, our approach can be potentially applicable to improve reservoir inflow forecasts in different regions of the world.

Spatial prediction of soil organic carbon is a crucial proxy to manage and conserve natural resources, monitoring CO2 and preventing soil erosion strategies within the landscape, regional, and global scale. The objectives of this study was to evaluate capability of artificial neural network and multivariate linear regression models in order to predict soil organic carbon using terrain attributes. A study area of 24 km2 in hilly regions of Zargham Ababd in south of Semirom under natural rangeland uses, was selected and then 125 soil samples (0-10 cm depth) were collected. Soil organic carbon was measured for the collected soil samples. Topographic attributes were calculated by a digital elevation model with 10 m spacing. Finally, multiple linear regression (MLR) analysis and ANN models were developed for soil organic carbon estimation in the study area and then the developed modeless were validated by additional samples (25 points). The results showed that the MLR and ANN models explained 60 and 89 % of the total variability of SOC, respectively, in the study area using terrain attributes. Sensitivity analysis based upon the ANN models, revealed that the profile curvature, stream power index, slope, sediment transport index, wetness index, plan curvature and aspect were identified as the important topographic attributes influencing the SOC distribution within the selected hillslope. The overall results indicated that topographic attributes and hydrological process control a significant variability of SOC. Prediction of the statistical studied models in the study area resulted in mean error and root mean square error values of 0.25, 0.3 in MLR equation and 0.006, 0.027 in ANN, respectively. Therefore, the ANN model could provide superior predictive performance when compared with developed MLR model. Keywords: Soil organic carbon, Terrain attributes, Linear regression, Artificial neural network

Estimating olive (Olea europaea L.) leaf area is an important aspect of monitoring plant health and evaluating growth processes in agriculture. Accurate estimation of leaf area allows for a better understanding of processes such as water and nutrient utilization, photosynthesis efficiency, respiration, and yield potential. This study aims to determine the most accurate, easy, and reliable leaf area estimation model using the geometric properties (length and width) of olive leaves. Additionally, the predictive performances of multiple linear regression (MLR) and artificial neural network (ANN) were compared. A total of 1320 leaf samples collected from 22 olive cultivars were used in the study. Leaf length and width were taken as input parameters, and both MLR and ANN models were developed for each cultivar. Both multiple linear regression (MLR) and artificial neural network (ANN) models demonstrated high predictive accuracy for olive leaf area estimation across 22 cultivars. The MLR models explained up to 96% of the variation in leaf area using leaf length (LL) and leaf width (LW), with low root mean square errors, indicating strong reliability. When cultivar identity was modeled as a categorical factor through dummy encoding, the model captured significant cultivar-specific effects without altering the overall predictive performance. The ANN models achieved slightly higher accuracy, with determination coefficients exceeding 0.99 and minimal prediction errors, confirming their superior ability to model nonlinear relationships. Across both approaches, leaf width contributed more strongly to leaf area than leaf length. Cultivar-specific differences were statistically significant for only a few genotypes, while most cultivars exhibited comparable patterns after adjustment for multiple testing. In conclusion, both MLR and ANN models demonstrated high accuracy in predicting olive leaf area, with ANN models showing slightly superior performance. However, MLR models also yielded highly reliable results, indicating that both approaches are viable for practical applications in olive cultivation. These predictive models can be effectively used for rapid, non-destructive phenotyping, growth monitoring, and precision management in olive breeding and production systems.

Short-range reservoir inflow forecasting using hydrological and large-scale atmospheric circulation information

Purpose – This paper aims to predict the needle penetration force (NPF) in denim fabrics using the artificial neural network (ANN) and multiple linear regression (MLR) models based on the effects of various sewing parameters. Design/methodology/approach – In order to design the ANN and MLR models, four parameters including fabric weight, number of fabric layers, weave pattern, and sewing needle size are taken into account as the input parameters and NPF as the output parameter. According to these parameters, 140 samples of data were resulted. Each sample was tested five times. From these 140 data (input-output data pairs), 112 were used for training the ANN and MLR models and 28 samples were used to test the performance of ANN and MLR. Also, the NPF was measured on the Instron tensile tester to simulate sewing process. Findings – The results indicated that the NPF in denim fabrics can be well predicted in terms of sewing parameters by using ANN and MLR models, in which the ANN model exhibits greater performance than MLR (RANN=0.989> RMLR=0.901). Research limitations/implications – The NPF measurement method is limited at low speed. Originality/value – Using the ANN model for forecasting NPF in denim fabrics can help the garment manufactures to produce high-quality denim products and improve the sewing process through reducing seam damage. The NPF could be also measured in the cycle loading conditions similar to sewing machine process by using a special designed tools mounted on the Instron tensile tester.

The purpose of our study was the evaluation of the most important factors that affect the volumetric and conventional mechanical properties of produced asphalt mix and the volumetric properties of built‐in asphalt layer. Asphalt mix design follows the standard procedure (Marshall procedure). We were interested not only in the quantity of bitumen specified by the Marshall procedure, but also in the quantity of stone aggregate fractions, temperatures of production and properties of bitumen that is used. The influence of these factors was investigated with several models. For the building of models we used 444 asphalt samples, analysed by one laboratory. To select the most important factors, several multiple linear regression (MLR) models, partial least squares (PLS) regression models and counterpropagation neural network models were made. Obtained models were tested with leave‐one‐out (LOO) and leave‐10%‐out cross‐validation procedures. The results of MLR and PLS models show that the independent variables are closely related. Among 21 variables there is only one found as less important. MLR and PLS models show better predictive ability than counterpropagation neural network models. The best MLR models will be employed for the preparation of the asphalt mix design (recipe) with some unknown material. Copyright © 2009 John Wiley & Sons, Ltd.

Pseudomonas chlororaphis 3732RN-L11 survival rates in soil and wheat rhizosphere were measured using intact soil core microcosms representing 23 sites across Canada. Linear multiple regression (LMR) models were developed to predict the survival rate of this genetically engineered microorganism (GEM) as a function of soil parameters measured at the time of microcosm inoculation. LMR models were tested by comparing their predicted survival rates with observed survival rates from environmental introductions of the GEM by Gagliardi et al. (2001) at five field sites across Canada over two years. No soil parameter (e.g., % clay) was highly correlated with GEM survival rates in soil or wheat rhizosphere. Total fungal colony-forming units (CFUs), % soil titanium (positive correlations), and % soil magnesium (negative correlation) were found to be the best LMR predictors of GEM survival rates in soil over two years. Total soil bacterial CFUs, nitrate, % soil potassium (positive correlations), and exchangeable magnesium (negative correlation) were found to be the best LMR predictors of GEM survival rate in wheat rhizosphere over two years. While LMR models were statistically significant, they were unable to reliably predict the survival rate of the GEM in field trial introductions. The results indicate that there can be considerable uncertainty associated with predicting GEM survival for multi-site environmental introductions.

The problem of reservoir inflow forecasting plays a critical role in reservoir management. However, reservoir inflow forecasting must be necessarily accurate and timely. This paper presents practical machine learning (ML) technique and the optimal lead time for reservoir inflow forecasting. The two well-knows ML: Support Vector Regression (SVR) and Random Forests (RF), were used to predict water inflow volume into the reservoir. Both methods will improve the efficiency of the reservoir inflow more accurately by using reservoir information data (amounts of rainfall and inflow) and climate indices (Sea Surface Temperatures (SSTs) and ocean indices). All historical monthly time series reservoir information for 18 years (between 1998 and 2015) is collected from the Hui Nam Sai Reservoir Project, located in Nakhon Si Thammarat Province, Thailand. Our experimental results showed Random Forest with climate indices gave the best performance method. We also found that the top three optimal lead time for reservoir inflow forecasting were the 10-month (t+10), 2-month (t+2) and 6-month (t+6) ahead with giving the least of RMSE of 4.41, 4.58 and 4.62, respectively. Similarly, when using ASE as a criterion, the top three optimal lead time for reservoir inflow forecasting were the 3-month (t+3), 10-month (t+10), and 5-month (t+5) with giving the least of ASE of 3.11, 3.14, and 3.16 respectively.KeywordsSupport vector regressionRandom forestsReservoir inflow forecastingTime seriesClimate indices

Estimating wet soil aggregate stability from easily available properties in a highly mountainous watershed

با برآورد ظرفیت تبادل کاتیونی با استفاده از ویژگی‏های پایه ای و زودیافت خاک می‏توان در وقت و هزینه صرفه‏جویی کرد. هدف این بررسی، مقایسه دو روش رگرسیون درختی و رگرسیون خطی چندگانه در برآورد ظرفیت تبادل کاتیونی (CEC) با استفاده از ویژگی‏هایخاک است. برای این منظور از داده های 106 نمونه خاک UNSODA استفاده شد. جهت برآورد CEC با استفاده از روش‏های رگرسیون درختی و رگرسیون خطی از ویژگی های اجزای بافت خاک،pH، ماده آلی و چگالی ظاهری استفاده شد. کارآیی روش رگرسیون درختی در برابر روش رگرسیون خطی چندگانه در برآورد CEC مقایسه شدند. نتایج نشان داد که در برآورد CEC با استفاده از روش های رگرسیونی تنها ضرایب ماده آلی (183/3) و درصد رس (274/0) که بیشترین همبستگی را با CEC دارند معنی دار شده و مدل رگرسیونی بر اساس این دو پارامتر توسعه یافت. همچنین از میان پارمترهای ورودی در روش رگرسیون درختی تنها پارامترهای ماده آلی و درصد رس در درخت رگرسیون ظاهر شد. روش رگرسیون درختی در دو مجموعه داده صحت سنجی و اعتبار سنجی بر اساس آماره های ارزیابی R2، RMSE، ME و GMER کارآیی بالاتری نسبت به روش های رگرسیونی خطی در برآورد CEC داشت. از میان روش های رگرسیونی خطی، مدل پیشنهادی کارآیی بالاتری نسبت به مدل های بل و ونکولن و بروسما و همکاران داشت.