Average Relative Root Mean Square Research Articles

Compactness prediction of asphalt concrete using Ground-Penetrating Radar: A comparative study

The compactness is an important indicator of quality control during the construction of the asphalt concrete pavement. The construction quality of asphalt concrete can be continuously, quickly, and non-destructively tested by establishing the correlation between dielectric constant and compactness through ground-penetrating radar. In this study, five compactness prediction models were established based on the relationship between volume parameters and the dielectric constant of each component of asphalt concrete. 18 samples of Sup20 asphalt concrete (Superior Performing Asphalt Pavement with the nominal aggregate size of 19 mm) were used to verify the established compactness prediction models. The results showed that the average relative error, maximum error, and root mean square error of the CP_Rayleigh model (compactness prediction model of Rayleigh.) were minimum, and thus the CP_Rayleigh model was selected for the following-up research in this study. The shapes of scatterers (aggregates and pores) were not spherical, and a shape factor u of 0.033 was introduced to modify the CP_Rayleigh model. The prediction accuracy of compactness can be effectively improved by the modified CP_Rayleigh model. Combined with field engineering verification, it was shown that the modified CP_Rayleigh model can be used to predict the compactness of asphalt concrete in the field.

Mapping Forest Stock Volume Based on Growth Characteristics of Crown Using Multi-Temporal Landsat 8 OLI and ZY-3 Stereo Images in Planted Eucalyptus Forest

Labeled as a fast-growing tree species, eucalyptus has outstanding carbon sequestration capacity. Forest stock volume (FSV) is regarded as an important parameter for evaluating the quality of planted eucalyptus forests. However, it is an intractable problem to map FSV of planted eucalyptus forests using optical images because of growth characteristics of the crown and low saturation levels. To improve the accuracy of FSV in planted eucalyptus forests, time series Landsat 8 OLI (LC8) images and ZY-3 stereo images were acquired in the study area. Additionally, then, three composite images were proposed using acquired Landsat 8 OLI images based on the size and shape of eucalyptus crowns, and several spectra variables were extracted from these composite images. Furthermore, corrected canopy height model (CCHM) was also extracted from ZY-3 stereo images. Meanwhile, four models (random forest (RF), support vector machine (SVM), K-nearest neighbor (KNN), and multiple linear regression (MLR)) were used to estimate the FSV with various variable sets using the importance of the alternative variables ranked by RF. The results show that the sensitivity between proposed spectral variables and FSV is significantly improved using proposed composed images based on the growth characteristics of the crown, especially for young eucalyptus forests. After adding CCHM and stand age to the optimal variable set, the average relative root mean square error (rRMSE) of estimated FSV decreased from 41.01% to 29.94% for single LC8 images and from 32.64% to 26.47% for proposed composite LC8 images, respectively. After using the variable set extracted from composite LC8 images, the number of samples with overestimated FSV was significantly decreased for the young forest. Furthermore, forest height plays an important role in improving the accuracy of mapping FSV, whether young or mature eucalyptus forest. It was also proved that composite images related to crown close and CCHM have great potential to delay the saturation phenomenon for mapping FSV in planted eucalyptus forest.

Retail Property Price Index Forecasting through Neural Networks

Rapid growth has been seen in the Chinese housing market during the past decade, and housing price forecasts have become more significant to investors and policy makers. In this study, we explore neural networks for retail property price index forecasts from ten major Chinese cities for July 2005–April 2021. Our goal is to build simple and accurate neural networks to contribute to purely technical forecasts of the Chinese retail property market. To facilitate the analysis, we examine different model settings based on algorithms, delays, hidden neurons, and data splitting ratios and construct a simple neural network with five delays and two hidden neurons. This leads to a rather stable performance of about 1.6% average relative root mean square error across the ten cities for the training, validation, and testing phases. Results here can be used on a stand-alone basis or combined with fundamental forecasts to form perspectives of retail property price trends and conduct policy analyses.

Acoustic Tomography Temperature Reconstruction Based on Virtual Observation and Residual Network

High-quality measurement of temperature distribution information is of great significance for industrial production. As a typical non-contact temperature measurement method, acoustic tomography (AT) can obtain temperature distribution through reconstruction algorithms. To improve the reconstruction quality, a two-stage algorithm based on virtual observation (VO) and residual network (ResNet) was proposed. This algorithm combines the advantages of VO and ResNet. Aiming at total least squares, the VO method was adopted to reconstruct the ultrasonic time-of-flight (TOF) to obtain the temperature distribution under a coarse grid. Then, ResNet was built to predict the temperature distribution under the refined grid, and the dual-input model was used to improve the network generalization ability. At the same time, the sub-pixel convolution layer was introduced to reduce the network computing dimension, which improves the computing efficiency and the reconstruction quality. The numerical simulation of typical temperature field models was carried out, and the reconstruction results show that the average relative error and root mean square error are less than 0.86% and 1.21%, respectively. The noise immunity and reconstruction spatial resolution are better than traditional algorithms and the previous two-stage algorithm.

Application of atomic electrostatic potential descriptors for predicting the eco-toxicity of ionic liquids towards leukemia rat cell line

The toxicity assessment of ionic liquids (ILs) towards the environment and living organisms has received great attention. Nevertheless, the huge number of ILs makes the toxicity data collection expensive and time-consuming, which motivates modeling development to fill data gaps of ILs toxicity. The group contribution (GC) method has been extensively applied for the estimation of various properties of ionic liquids. This study proposed a novel method, named the non-integer group contribution (NGC) method, which creatively utilizes the atomic electrostatic potential descriptors for modeling. Specifically, the average values of electrostatic potential (AVEP) and the electrostatic potential surface area (SEP) of atoms in the cations and anions of ILs were calculated and used to obtain the group descriptors in this work. Two NGC models were developed to predict the toxicity of ILs. Results show that both proposed models have satisfactory predictability. In contrast, the NGC-2 model based on SEP descriptors exhibits better predictability due to its higher coefficient of determination (R2 = 0.927), the lower average absolute relative deviation (AARD = 11.257 %) and root mean square error (RMSE = 0.261) for the entire dataset. The NGC-2 model also shows better performance than the traditional GC method, demonstrating its advanced superiority. Therefore, the proposed approach has high potential in terms of generalization and applicability for predicting the property of compounds.

Hyperparameter Tuning of Artificial Neural Networks for Well Production Estimation Considering the Uncertainty in Initialized Parameters.

A well production rate is an essential parameter in oil and gas field development. Traditional models have limitations for the well production rate estimation, e.g., numerical simulations are computation-expensive, and empirical models are based on oversimplified assumptions. An artificial neural network (ANN) is an artificial intelligence method commonly used in regression problems. This work aims to apply an ANN model to estimate the oil production rate (OPR), water oil ratio (WOR), and gas oil ratio (GOR). Specifically, data analysis was first performed to select the appropriate well operation parameters for OPR, WOR, and GOR. Different ANN hyperparameters (network, training function, and transfer function) were then evaluated to determine the optimal ANN setting. Transfer function groups were further analyzed to determine the best combination of transfer functions in the hidden layers. In addition, this study adopted the relative root mean square error with the statistical parameters from a stochastic point of view to select the optimal transfer functions. The optimal ANN model’s average relative root mean square error reached 6.8% for OPR, 18.0% for WOR, and 1.98% for GOR, which indicated the effectiveness of the optimized ANN model for well production estimation. Furthermore, comparison with the empirical model and the inputs effect through a Monte Carlo simulation illustrated the strength and limitation of the ANN model.

Investigation on solubility behavior of 2-chloronicotinamide in ten mono-solvents: Measurement, correlation, molecular simulation and thermodynamic analysis

Solubility behavior as a significant molecular property should be known in the field of crystal form screening, dosage form along with separation and purification in the pharmaceutical industry. In this research work, solubility data of 2-chloronicotinamide (2-CNA) in ten pure solvents including methanol (MeOH), ethanol (EtOH), ethylene glycol (EG), 2-methoxyethanol (EGME), 2-ethoxyethanol (EGEE), methyl acetate (MeAc), water, 1,4-dioxane (Diox), N,N-dimethylformamide (DMF) and N,N-dimethylacetamide (DMAC) within the range of 278.15/288.15 K to 323.15 K was experimentally determined by laser monitoring method at 0.1 MPa. Solubility of 2-CNA presented a positive correlation with temperature, and increased with incremental temperature. The result was indicative of a common tendency: DMAC > DMF > EGME > EGEE > EG > MeOH > Diox > EtOH > MeAc > water. The solvation free energy, molecular electrostatic potential surface (MEPS) and Hirshfeld surface were used to study the intermolecular interactions. Moreover, six mathematical models namely “Margules, UNIQUAC, NRTL, Wilson, NRTL-SAC and Jouyban model” were utilized to correlate solubility data of 2-CNA in mono-solvents. UNIQUAC model could provide maximum regression accuracy in correlation with solubility data among six models in terms of average relative deviation (ARD) and root mean square deviation (RMSD). Finally, thermodynamic parameters obtained from UNIQUAC equation indicated that all dissolution and mixing process were always entropy-increasing, endothermic and spontaneous.

Evaluation of Crop Health Status With UAS Multispectral Imagery

This study presents the results of a field experiment conducted for assessing the crop health status of several barley and oat crop fields in Prince Edward Island, Canada. The crop fields were mapped with an Unmanned Aircraft System (UAS) and the crop health status was assessed through the Green Area Index (GAI) and vegetation indices (VIs). GAI maps were produced from the UAS imagery and VIs using machine learning pipelines with several regression algorithms (Multiple Linear Models, Support Vector Machines, Random Forests, and Artificial Neural Networks) along with a feature selection strategy. The Random Forests algorithm was shown to be the best algorithm for GAI prediction with an average relative Root Mean Square Error of 10.86% and a Mean Absolute Error of 0.67. The resulting GAI maps and the regression feature space were classified with Random Forests to discriminate between vigorous and stressed crop areas. We achieved a mean overall accuracy of 94%. The limits of the study are also presented.

A Classification-Based, Semianalytical Approach for Estimating Water Clarity From a Hyperspectral Sensor Onboard the ZY1-02D Satellite

Water clarity (Zsd) is a widely used quality indicator that can be estimated from remote sensing imagery. China’s newest generation Advanced HyperSpectral Imager (AHSI) onboard the ZY1-02D satellite is expected to enable accurate water clarity retrieval for inland waters, since AHSI can provide abundant band choices while its 30-m spatial resolution is advantageous for monitoring small inland water bodies. In this study, to retrieve Zsd from the ZY1-02D imagery for inland waters with varying turbidities, we propose a classification-based, semi-analytical method in which the red/blue band ratio is employed to distinguish clear to moderately turbid water and highly turbid waters. Two Quasi Analytical Approaches (QAAs), QAAv5 and QAAm14, are used to estimate the total absorption coefficient (a(λ)) and the backscattering coefficient (bb(λ)) for clear to moderately turbid water and highly turbid waters, respectively. The estimated a(λ) and bb (λ) are utilized to obtain the diffuse attenuation coefficient Kd, followed by the Zsd calculations. Compared with 70 matchups of in situ measured Zsd values (0–6.5 m), the ZY1-02D image-derived Zsd achieved an R2 of 0.98, with an average unbiased relative error and root mean square error of 29.1% and 0.52 m, respectively. In addition, the proposed method can yield Zsd with higher accuracies than that of optimized empirical models. Therefore, the ZY1-02D AHSI imagery can retrieve reliable Zsd for both clear (> 3 m) and turbid waters (0–3.0 m), thereby serving as a useful satellite data source for monitoring the water clarity of large-scale inland water bodies.

Gas path parameter prediction of aero-engine based on an autoregressive discrete convolution sum process neural network

In order to improve the approximation ability of neural network to functional and improve the prediction accuracy of time series data, this paper proposes an autoregressive discrete convolution sum process neural network prediction model and applies it to the prediction of aero-engine gas path performance parameters. Coupling threshold wavelet de-noising method is applied to the preprocessing of engine gas path parameters, which can effectively remove the noise in the time series data. Process neurons are applied to artificial neural networks to expand the computational functions of artificial neurons. The process neuron can not only realize the spatial aggregation operation of discrete input data and the connection weight, but also realize the time aggregation operation of the product integral of the continuous input function and the connection weight function. The output of discrete time series is affected by the current input/output, and also by the historical input/output value. Therefore, an autoregressive feedback link is added to the network topology, and the value of the network's output node is used as the input to adjust the network connection weight. The convolution sum operation of discrete time series data can realize the time accumulation effect, so the discrete convolution sum operator is used to replace the integral operator to realize the time aggregation function of the process neuron. Since the integral operation of the continuous function is avoided, the weight training process of the process neural network is effectively simplified, and the accuracy loss in the continuous-time function fitting process of discrete input samples is effectively avoided. Bayesian regularization network weight learning algorithm is used to solve the problems of slow convergence speed and easy to fall into local optimum of back propagation learning algorithm, and improve the generalization ability of neural network. It can be seen from the prediction simulation results of the two engines that the average relative error and root mean square error of the engine gas path parameter prediction based on the autoregressive discrete convolution sum process neural network model can be reduced to 0.67% and 0.35 respectively. The stable and high-precision prediction results show that the network model and weight learning algorithm proposed in this paper have better robustness in functional approximation, and have higher accuracy in time series data prediction.

Rent index forecasting through neural networks

PurposeChinese housing market has been growing fast during the past decade, and price-related forecasting has turned to be an important issue to various market participants, including the people, investors and policy makers. Here, the authors approach this issue by researching neural networks for rent index forecasting from 10 major cities for March 2012 to May 2020. The authors aim at building simple and accurate neural networks to contribute to pure technical forecasting of the Chinese rental housing market.Design/methodology/approachTo facilitate the analysis, the authors examine different model settings over the algorithm, delay, hidden neuron and data spitting ratio.FindingsThe authors reach a rather simple neural network with six delays and two hidden neurons, which leads to stable performance of 1.4% average relative root mean square error across the ten cities for the training, validation and testing phases.Originality/valueThe results might be used on a standalone basis or combined with fundamental forecasting to form perspectives of rent price trends and conduct policy analysis.

House price forecasting with neural networks

• We develop neural network models for house price forecasting in China. • We explore the forecasting issue for 100 major Chinese cities. • We construct simple, accurate, and stable models. • The models are useful as technical tools and for policy analysis. The house market has been rapidly growing for the past decade in China, making price forecasting an important issue to the people and policy makers. We approach this problem by exploring neural networks for forecasting of house prices from one hundred major cities for the period of June 2010–May 2019, serving as the first study with such wide coverage for the emerging Chinese market through a machine learning technique. We aim at constructing simple and accurate neural networks as a contribution to pure technical forecasting of house prices. To facilitate the analysis, we investigate different model settings over the algorithm (the Levenberg-Marquardt, scaled conjugate gradient, and Bayesian regularization), delay (from two to six), hidden neuron (two, three, five, and eight), and data spitting ratio (70%–15%–15%, 60%–20%–20%, and 80%–10%–10% for trainingvalidationtesting), and arrive at a rather simple neural network with only four delays and three hidden neurons that leads to stable performance of 1% average relative root mean square error across the one hundred cities for the training, validation, and testing phases. We demonstrate the usefulness of the machine learning approach to the house price forecasting problem in the Chinese market. Our results could be used on a standalone basis or combined with fundamental forecasting in forming perspectives of house price trends and conducting policy analysis. Our empirical framework should not be difficult to deploy, which is an important consideration to many decision makers, and has potential to be generalized for house price forecasting of other cities in China.

Remaining capacity prediction of lithium-ion battery based on the feature transformation process neural network

In order to improve the prediction accuracy of discrete time series data, a lithium-ion battery remaining capacity prediction model based on feature transformation process neural network is proposed. According to the time series characteristics of lithium-ion battery performance degradation data, the remaining capacity prediction of lithium-ion battery is converted into a functional approximation method. The integral operation of continuous function is used to realize the time accumulation effect of network input data. In order to simplify the integral operation of continuous function, a discrete Walsh transform is performed on the input data, and the integral operation of continuous function is transformed into the inner product operation of the discrete Walsh transform pair. This method simplifies the integral operation of the continuous function and eliminates the loss of precision caused by the continuity of discrete time series data. A Levenberg-Marquardt network weight learning algorithm based on the discrete Walsh transform is developed. The model and algorithm are applied to predict the remaining capacity of lithium-ion batteries. The experimental results show that the model can reduce the average absolute error, average relative error and root mean square error of lithium-ion battery remaining capacity prediction to 0.0231AH, 1.73% and 0.0299 respectively.

Evaluation of bio-inspired optimization algorithms hybrid with artificial neural network for reference crop evapotranspiration estimation

Reference crop evapotranspiration (ETo) is a determinant factor in agricultural water resource management. Therefore, accurate ETo information is critical to quantify crop water requirements for precision agriculture management. This study coupled bio-inspired optimization algorithms with artificial neural network (ANN), i.e., ANN with bat algorithm (BA-ANN), ANN with cuckoo search algorithm (CSA-ANN), and ANN with whale optimization algorithm (WOA-ANN), and developed three hybrid ANN models for daily ETo modeling with limited inputs. The models were trained and evaluated using a k-fold test approach and long-term daily climatic data from 2001 to 2018 at six climatic stations in the Loess Plateau of north China. Three input scenarios were used, including temperature-based inputs, radiation-based inputs, and mass transfer-based inputs. The statistical comparison showed that the hybrid WOA-ANN offered better estimates than BA-ANN and CSA-ANN in all three input scenarios. In general, the radiation-based WOA-ANN provided the most accurate ETo estimations, with regional average relative root mean square error and Nash-Sutcliffe efficiency coefficient of 13.3% and 0.959, respectively. The temperature-based WOA-ANN offered acceptable and reasonable ETo estimates. Thus, it is a reliable tool for ETo modeling, given that air temperature is available in many regions. Overall, the bio-inspired optimization algorithms are robust tools for enhancing ANN performance in ETo simulation, and thus they are highly recommended to estimate ETo in the study region. Our study proposed powerful models for accurately estimating ETo with limited inputs, offering practical implications for the development of precision agriculture.

Calculation of equivalent charge weight per delay and vibration velocity prediction for millisecond delay blasting

Drilling and blasting is the most economical rock fracture technology in water conservancy, transportation, mining and tunnel engineering. And the application of nonel detonator network in rock blasting is still the most widely used initiation method in engineering blasting practice. Due to the detonator delay error, there is a deviation between the actual initiation time and designed initiation time in the Nonel detonation network, which will cause the change of blasting time sequence and the overlapping of blast-holes. There are detonator dispersion phenomenon with the same delay time and superposition effect for seismic waves with different delay time, which brings great trouble to the value of charge weight per delay and the prediction of particle peak vibration velocity. In order to the predict particle peak vibration velocity more accurately and efficiently, the millisecond delay blasting test was conducted, and the calculation model of vibration velocity for group blast-hole simultaneous blasting with dispersed charge was established. The influence of the blast-hole number on the equivalent charge weight for simultaneous blasting and its value selection method were studied and constructed. The modified particle peak vibration velocity scaled distance formula and the particle peak vibration velocity prediction method were proposed based on the results of regression analysis of single-hole blasting. The results show that the equivalent charge weight of group blast-hole simultaneous blasting is smaller than the nominal charge weight per delay, and the equivalent charge weight of simultaneous blasting can be calculated by converting through the reduction coefficient, which decreases exponentially with the increase of the blast-hole number. The superposition effect of seismic waves with different delay time can be considered by introducing vibration wave superposition factor into the modified particle peak vibration velocity scaled distance formula. The average absolute error, average relative error and root mean square error between the actual and the predicted particle peak vibration velocity values are 0.05 cm/s, 9.52% and 0.059 cm/s, respectively. It is feasible to apply the modified particle peak vibration velocity proportional distance regression analysis method to the prediction and control of blasting vibration velocity in the field.

Support Vector Machine Model for Predicting Gas Saturated and Undersaturated Crude Oil Viscosity of Niger Delta Oil Reservoir

Oil viscosity is one of the most important physical and thermodynamic property used when considering reservoir simulation, production forecasting and enhanced oil recovery. Traditional experimental procedure is expensive and time consuming while correlations are replete however they are limited in precision, hence need for a new Machine Learning (ML) models to accurately quantify oil viscosity of Niger Delta crude oil.
 This work presents use of ML model to predict gas-saturated and undersaturated oil viscosities. The ML used is the Support Vector Machine (SVM), it is applicable for linear and non-linear problems, the algorithm creates a hyperplane that separates data into two classes. The model was developed using data sets collected from the Niger Delta oil field. The data set was used to train, cross-validate, and test the models for reliability and accuracy. Correlation of Coefficient, Average Absolute Relative Error (AARE) and Root Mean Square Error (RMSE) were used to evaluate the developed model and compared with other correlations.
 Result indicated that SVM model outperformed other empirical models revealing the accuracy and advantage SVM a ML technique over expensive empirical correlations.

Assessment of river water quality indices based on various fuzzy models and arithmetic indexing method

River water is a major source of natural freshwater for both rural and urban areas. The severe threat in the form of pollutants is the major cause of deterioration of surface water quality. Water Quality Index (WQI) is a single measure of overall water quality in a specific location with a special emphasis on the time-based readings of water quality parameters. WQI can be used as a good tool to assess the intensity of water pollution. In this study, two water quality index models using fuzzy logic in MATLAB R2015a by trapezoidal and sigmoid membership functions are proposed. Fuzzy water quality index models were developed for various seasons, using eight experimentally estimated water quality parameters, such as Temperature (T), Chlorides (Cal−), Nitrates (NO3 −), Sulphates (SO4 −), Total Coli forms (TC), Total Dissolved Solids(TDS), Electrical conductivity (EC) and Total Hardness (TH) of the water samples at 8 locations stretching55 km of Chalakudy River from January 2018 to December 2018. The models are validated by comparing with the WQI values obtained by the application of arithmetic index method (AIM) based on Indian Standards, by finding the absolute average relative error (AARE) and root mean square error (RMSE). The fuzzy logic model using trapezoidal membership function (FTWQI) was found to be more reliable with the least error (AARE 0.0214 and RMSE 0.318) compared with the fFuzzy sigmoid water quality index (FSWQI) (AARE 0.573 and RMSE 0.86). The models enable the easy prediction of the risk of water consumption.

Ocean Chlorophyll-a retrieval using GF1-WFV data-a case study of the central Bohai Sea

Chlorophyll-a (Chl-a) concentration is an important indicator of phytoplankton bloom and water eutrophication. It is also a basic parameter of water quality. Remote sensing technology has been widely used for water quality monitoring in the past. In this study, the Chinese GF1 (Gaofen-1) WFV (Wide Field of View) images acquired on May 19th, 2016 were used to retrieve Chl-a concentration of central Bohai sea in order to assess the recovery condition of marine ecosystem which has been seriously polluted by the Penglai 19-3 platform oil spill accident in 2011. The remote sensing model of Chl-a concentration was developed using the linear regression method. The leave-one-out cross-validation (LOOCV) method showed that the average relative error and root mean square error (RMSE) between retrieved Chl-a concentration and in situ measurement was 11.67% and 0.1808, respectively. It indicated that the remote sensing model could correctly reflect the Chl-a concentration and its spatial distribution of the study area. The average Chl-a concentration in the buffer regions with the distance of 2km, 4km, 6km to the Penglai 19-3 platform was 0.32μg/L, 0.37μg/L and 0.40μg/L, respectively. They were in normal range and lower than that of the coastal waters. The spatial distribution of the result was similar to previous studies of Bohai sea in spring. These results indicated that Chl-a concentration of central Bohai sea was returned to normal after oil spill accident occurred for 5 years. The growth of phytoplankton in surface seawater has been returned to normal. The major causes for Chl-a concentration in the Bohai Sea was the land-sourced pollutant during the study period.

An Effective PSO-LSSVM-Based Approach for Surface Roughness Prediction in High-Speed Precision Milling

Surface roughness is one of the important indicators to measure the surface quality of parts processed, in addition to the cutting parameters affecting the surface roughness, the inevitable tool wear during the cutting process also makes the surface roughness constantly changing. In order to achieve high-precision prediction of machined surface roughness, a high-speed precision milling surface roughness prediction method based on particle swarm optimization least squares support vector machine (PSO-LSSVM) is proposed in this paper. The prediction method uses standardized cutting parameters and tool wear as the input variables, and uses the LSSVM algorithm to model the relationship between the input variables and the surface roughness, the improved PSO algorithm is used to optimize the hyperparameters of LSSVM so as to improve the generalization ability. In order to verify the effectiveness and superiority of PSO-LSSVM predictive performance, two surface roughness prediction models have been developed based on support vector machine (SVM) and response surface method (RSM), respectively. With the same sample conditions, average relative error and root mean square error of PSO-LSSVM prediction model are the smallest, and the correlation coefficient of PSO-LSSVM method is the largest. It is indicate that the surface roughness prediction accuracy and generalization ability based on PSO-LSSVM are the best under different cutting parameters.

The Ultra-Short-Term Forecasting of Global Horizonal Irradiance Based on Total Sky Images

Solar photovoltaics (PV) has advanced at an unprecedented rate and the global cumulative installed PV capacity is growing exponentially. However, the ability to forecast PV power remains a key technical challenge due to the variability and uncertainty of solar irradiance resulting from the changes of clouds. Ground-based remote sensing with high temporal and spatial resolution may have potential for solar irradiation forecasting, especially under cloudy conditions. To this end, we established two ultra-short-term forecasting models of global horizonal irradiance (GHI) using Ternary Linear Regression (TLR) and Back Propagation Neural Network (BPN), respectively, based on the observation of a ground-based sky imager (TSI-880, Total Sky Imager) and a radiometer at a PV plant in Dunhuang, China. Sky images taken every 1 min (minute) were processed to determine the distribution of clouds with different optical depths (thick, thin) for generating a two-dimensional cloud map. To obtain the forecasted cloud map, the Particle Image Velocity (PIV) method was applied to the two consecutive images and the cloud map was advected to the future. Further, different types of cloud fraction combined with clear sky index derived from the GHI of clear sky conditions were used as the inputs of the two forecasting models. Limited validation on 4 partly cloudy days showed that the average relative root mean square error (rRMSE) of the 4 days ranged from 5% to 36% based on the TLR model and ranged from 12% to 32% based on the BPN model. The forecasting performance of the BPN model was better than the TLR model and the forecasting errors increased with the increase in lead time.