Research article

Artificial Intelligence based accurately load forecasting system to forecast short and medium-term load demands

  • An efficient management and better scheduling by the power companies are of great significance for accurate electrical load forecasting. There exists a high level of uncertainties in the load time series, which is challenging to make the accurate short-term load forecast (STLF), medium-term load forecast (MTLF), and long-term load forecast (LTLF). To extract the local trends and to capture the same patterns of short, and medium forecasting time series, we proposed long short-term memory (LSTM), Multilayer perceptron, and convolutional neural network (CNN) to learn the relationship in the time series. These models are proposed to improve the forecasting accuracy. The models were tested based on the real-world case by conducting detailed experiments to validate their stability and practicality. The performance was measured in terms of squared error, Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE). To predict the next 24 hours ahead load forecasting, the lowest prediction error was obtained using LSTM with R2 (0.5160), MLP with MAPE (4.97), MAE (104.33) and RMSE (133.92). To predict the next 72 hours ahead of load forecasting, the lowest prediction error was obtained using LSTM with R2 (0.7153), MPL with MAPE (7.04), MAE (125.92), RMSE (188.33). Likewise, to predict the next one week ahead load forecasting, the lowest error was obtained using CNN with R2 (0.7616), MLP with MAPE (6.162), MAE (103.156), RMSE (150.81). Moreover, to predict the next one-month load forecasting, the lowest prediction error was obtained using CNN with R2 (0.820), MLP with MAPE (5.18), LSTM with MAE (75.12) and RMSE (109.197). The results reveal that proposed methods achieved better and stable performance for predicting the short, and medium-term load forecasting. The findings of the STLF indicate that the proposed model can be better implemented for local system planning and dispatch, while it will be more efficient for MTLF in better scheduling and maintenance operations.

    Citation: Faisal Mehmood Butt, Lal Hussain, Anzar Mahmood, Kashif Javed Lone. Artificial Intelligence based accurately load forecasting system to forecast short and medium-term load demands[J]. Mathematical Biosciences and Engineering, 2021, 18(1): 400-425. doi: 10.3934/mbe.2021022

    Related Papers:

    [1] Mostean Bahreinipour, Hajar Zarei, Fariba Dashtestani, Jamal Rashidiani, Khadijeh Eskandari, Seyed Ali Moussavi Zarandi, Susan Kabudanian Ardestani, Hiroshi Watabe . Radioprotective effect of nanoceria and magnetic flower-like iron oxide microparticles on gamma radiation-induced damage in BSA protein. AIMS Biophysics, 2021, 8(2): 124-142. doi: 10.3934/biophy.2021010
    [2] Thuy Hien Nguyen, Catherine C. Moore, Preston B. Moore, Zhiwei Liu . Molecular dynamics study of homo-oligomeric ion channels: Structures of the surrounding lipids and dynamics of water movement. AIMS Biophysics, 2018, 5(1): 50-76. doi: 10.3934/biophy.2018.1.50
    [3] Stephanie H. DeLuca, Samuel L. DeLuca, Andrew Leaver-Fay, Jens Meiler . RosettaTMH: a method for membrane protein structure elucidation combining EPR distance restraints with assembly of transmembrane helices. AIMS Biophysics, 2016, 3(1): 1-26. doi: 10.3934/biophy.2016.1.1
    [4] Michael B. Sherman, Juan Trujillo, Benjamin E. Bammes, Liang Jin, Matthias W. Stumpf, Scott C. Weaver . Decontamination of digital image sensors and assessment of electron microscope performance in a BSL-3 containment. AIMS Biophysics, 2015, 2(2): 153-162. doi: 10.3934/biophy.2015.2.153
    [5] Panisak Boonamnaj, Pornthep Sompornpisut, Piyarat Nimmanpipug, R.B. Pandey . Thermal denaturation of a coronavirus envelope (CoVE) protein by a coarse-grained Monte Carlo simulation. AIMS Biophysics, 2022, 9(4): 330-340. doi: 10.3934/biophy.2022027
    [6] Riyaz A. Mir, Jeff Lovelace, Nicholas P. Schafer, Peter D. Simone, Admir Kellezi, Carol Kolar, Gaelle Spagnol, Paul L. Sorgen, Hamid Band, Vimla Band, Gloria E. O. Borgstahl . Biophysical characterization and modeling of human Ecdysoneless (ECD) protein supports a scaffolding function. AIMS Biophysics, 2016, 3(1): 195-210. doi: 10.3934/biophy.2016.1.195
    [7] Ashwani Kumar Vashishtha, William H. Konigsberg . Effect of different divalent cations on the kinetics and fidelity of DNA polymerases. AIMS Biophysics, 2018, 5(4): 272-289. doi: 10.3934/biophy.2018.4.272
    [8] Vilas K. Mahajan, Sandip P. Patil, Shirish H. Sonawane, Gunvant H. Sonawane . Ultrasonic, photocatalytic and sonophotocatalytic degradation of Basic Red-2 by using Nb2O5 nano catalyst. AIMS Biophysics, 2016, 3(3): 415-430. doi: 10.3934/biophy.2016.3.415
    [9] Ganesh Prasad Tiwari, Santosh Adhikari, Hari Prasad Lamichhane, Dinesh Kumar Chaudhary . Natural bond orbital analysis of dication magnesium complexes [Mg(H2O)6]2+ and [[Mg(H2O)6](H2O)n]2+; n=1-4. AIMS Biophysics, 2023, 10(1): 121-131. doi: 10.3934/biophy.2023009
    [10] Roman Marks, Piotr H. Pawłowski . Rotational-electric principles of RNA/DNA and viability. AIMS Biophysics, 2023, 10(3): 385-400. doi: 10.3934/biophy.2023023
  • An efficient management and better scheduling by the power companies are of great significance for accurate electrical load forecasting. There exists a high level of uncertainties in the load time series, which is challenging to make the accurate short-term load forecast (STLF), medium-term load forecast (MTLF), and long-term load forecast (LTLF). To extract the local trends and to capture the same patterns of short, and medium forecasting time series, we proposed long short-term memory (LSTM), Multilayer perceptron, and convolutional neural network (CNN) to learn the relationship in the time series. These models are proposed to improve the forecasting accuracy. The models were tested based on the real-world case by conducting detailed experiments to validate their stability and practicality. The performance was measured in terms of squared error, Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE). To predict the next 24 hours ahead load forecasting, the lowest prediction error was obtained using LSTM with R2 (0.5160), MLP with MAPE (4.97), MAE (104.33) and RMSE (133.92). To predict the next 72 hours ahead of load forecasting, the lowest prediction error was obtained using LSTM with R2 (0.7153), MPL with MAPE (7.04), MAE (125.92), RMSE (188.33). Likewise, to predict the next one week ahead load forecasting, the lowest error was obtained using CNN with R2 (0.7616), MLP with MAPE (6.162), MAE (103.156), RMSE (150.81). Moreover, to predict the next one-month load forecasting, the lowest prediction error was obtained using CNN with R2 (0.820), MLP with MAPE (5.18), LSTM with MAE (75.12) and RMSE (109.197). The results reveal that proposed methods achieved better and stable performance for predicting the short, and medium-term load forecasting. The findings of the STLF indicate that the proposed model can be better implemented for local system planning and dispatch, while it will be more efficient for MTLF in better scheduling and maintenance operations.


    Abbreviations: IDP: Intrinsically disordered protein; IDR: Intrinsically disordered region; YPRA: Yeast proteinase A; TFE: 2,2,2-trifluoroethanol; CD: Circular dichroism; SDSL: Site-directed spin labeling; CW-EPR: Continuous wave electron paramagnetic resonance; IAP: 3-(2-Iodoacetamido)-PROXYL; MTSL: (1-oxyl-2,2,5,5-tetramethyl-Δ3-pyrroline-3-methyl) methanethiosulfonate; MSL: 4-maleimido-TEMPO; DTT: Dithiothreitol

    1. Introduction

    Proteins or protein regions of 50 or more residues, which do not have stable secondary or tertiary structure under physiological conditions, are characterized as intrinsically disordered proteins (IDPs) or intrinsically disordered regions (IDRs) [1,2,3,4,5,6,7]. Although the structure-function paradigm suggests that 3D structure is necessary for function, the discovered importance of intrinsic disorder within proteins or protein regions of higher eukaryotic systems shows how function could arise from various unstructured states [3,4,8,9]. IDPs have important roles in cellular signal transduction, translation, and transcription, influencing the study of disordered proteins, focusing on both the functions provided by IDPs, and the conformational changes associated with target binding [4,7,10].

    IA3, an IDP found in Sacchoromyeces cerevisiae, is composed of 68 amino acid residues and acts as an inhibitor of yeast proteinase A (YPRA). Previous studies have shown IA3 to adopt an α-helical conformation when bound in the active site of YPRA and when exposed to the secondary structure stabilizer 2,2,2-trifluoroethanol (TFE). These observed helical tendencies of IA3 differ between the N-terminal residues (2-34) and C-terminal residues (35-68) [11,12,13,14,15]. X-ray crystallographic models of IA3 bound to YPRA indicate α-helical structure in the N-terminal region (residues 2-34). In contrast, the C-terminal region (residues 35-68) has unresolved electron density, suggesting disorder (Figure 1) [11]. Biophysical methods of circular dichroism (CD) spectroscopy, 2D H1 N15 nuclear magnetic resonance spectroscopy, molecular dynamics simulations, laser temperature-jump fluorescence spectroscopy, fluorescence resonance energy transfer spectroscopy, and site directed spin labeling (SDSL), provide data supporting a two-state transition for IA3, where the C-terminus undergoes a helical transition but is less pronounced than the N-terminus [13,14,15,16,17,18].

    SDSL, in combination with continuous wave electron paramagnetic resonance (CW-EPR) spectroscopy is a biophysical tool for probing structure, dynamics, and conformational changes in macromolecules [19,20,21,22,23]. For proteins, typically a CW-EPR active chemical group, such as a nitroxide spin label, is incorporated at desired locations by chemical modification of a substituted cysteine. The introduction of an active chemical group provides a reactive side group for covalent bond spin labeling. Recent progress in spin labeling methodology has also demonstrated the ability to incorporate non-natural amino acids for subsequent chemical modification, as well as novel schemes for specifically labeling tyrosine residues; a potentially very useful strategy for spin labeling IDPs [24,25,26,27].

    Figure 1. Ribbon diagram of IA3 bound to yeast proteinase A (YPRA). Structurally resolved residues (3-31) are rendered as a light orange α-helix, with the unresolved C-terminal residues represented as a sequence chain of the single letter amino acids. Sites chosen for spin label scanning within the resolved α-helix are represented as red spheres located at the alpha carbon positions. Sites that were spin-labeled within the C-terminal region are annotated by blue letters (PDB: 1DPJ).

    The resultant CW-EPR spectral line shape reflects motional averaging of the nitroxide spin label [20,21,28]. Typically, the effects are described by three main types of motion, occurring in the 0.1-50 ns time scale [29,30]. The corresponding correlation times are referred to as the following: τR, the overall tumbling of the protein, τI, the movement of the spin label about the bonds connecting it to the protein, and τB, the motion of the spin labeled protein backbone. SDSL CW-EPR has been applied to study structure-to-unstructured transitions in proteins as well as to IDP systems [5,17,22,31,32]. For structured proteins ≥ 15 kDa, line shape parameters such as the second moment and scaled mobility readily reveal conformational changes without the need to increase solution viscosity [29,33]. Ordered to disordered transitions in structured proteins are readily characterized by these “traditional” line shape parameter analyses [34,35]. However, in IDPs, the degree of disorder and the dynamics of the system often times results in “isotropic-like” spectra that fall into the fast motional averaging limit. In these cases, for relatively fast motional averaging, the ratio of the intensities of the low field (h(+1)) and central field (h(0)) transitions provide an alternative line shape analysis parameter, h(1)/h(0), that is useful to describe conformational changes [5,17,22,36].

    Here, extending upon previous work, we performed a spin-labeled cysteine scanning profile of the N- and C-termini of IA3, where a series of fifteen singly spin-labeled IA3 constructs consisting of ten and five cysteine substitutions in the N- and C-terminus were generated, respectively (Figure 1) [17]. A comparison is provided between the termini on a site by site basis revealing variations in the degree of transition of the two termini, and sensitivity of the helical content of the N-terminus to the spin-labeled cysteine substitution and select amino acid substitutions. As these results show, the degree of site-specific helical propensity of the N-terminus is modulated by the introduction of a chemically modified cysteine residue. By further probing site V8 of IA3, the specific effects on secondary structure due to site-specific amino acid substitution and spin labeling are analyzed. The size of an introduced residue or chemical modification has marked effects on the helical propensity within the N-terminal region of IA3, whether found on the buried (concave) or solvent exposed (convex) side of the helix when bound to YPRA. By comparing SDSL and CD results, we are able to determine sites within IA3 that can tolerate spin-label incorporation and retain WT helical tendencies; thus, we label those our “wild-type mimics”.


    2. Materials and methods


    2.1. Materials

    BL21(DE3) pLysS cells were purchased from Invitrogen (Carlsbad, CA). E. coli codon-optimized DNA for the IA3 gene from S. cerevisiae, and DNA primers used for site-directed mutagenesis were purchased from DNA2.0 (Newark, CA). DNA taq polymerase and Dpn1 were purchased from New England Biolabs (Ipswich, MA). A 5-mL chelating column was purchased from GE Healthcare (Wauwatosa, WI). A HiPrep 26/10 desalting column was purchased from Amersham (Pittsburgh, PA). The 16.5% Tris-Tricene gels were purchased from Bio-Rad (Hercules, CA). 3-(2-Iodoacetamido)-PROXYL (IAP), (1-oxyl-2,2,5,5-tetramethyl-Δ3-pyrroline-3-methyl) methanethiosulfonate (MTSL), and 3-maleimido-TEMPO (MSL) spin label was purchased from Sigma Aldrich (St. Louis, MO). The 0.60 i.d. × 0.84 o.d. capillary tubes were purchased from Fiber Optic Center (New Bradford, MA). Unless otherwise stated, all other reagents and products were purchased from Fisher Scientific (Pittsburgh, PA) and used as received.


    2.2. Protein expression and purification of IA3 wild type and mutants

    The optimized IA3 gene was cloned into the pET-22b+ vector containing a C-terminal 6× His-tag, adding the sequence LEHHHHHH to the C-terminus as described previously [17]. Site directed mutagenesis via polymerase chain reaction was used to introduce non-native cysteines by designing mutagenic primers. The substituted sequence in each resultant plasmid was confirmed via Sanger DNA sequencing. WT and cysteine substituted constructs were purified as described previously [17]. Following affinity chromatography, residual nickel from the column was removed by addition of EDTA to a concentration of 100 mM and cysteine reduction was ensured by addition of dithiothreitol (DTT) to a concentration of 0.1 mM. Protein purity was assessed using a 16.5% tris-tricene SDS-PAGE gel.


    2.3. Spin labeling of IA3 mutants

    WT and cysteine substituted proteins were buffer exchanged into 50 mM sodium phosphate, 300 mM sodium chloride, pH 7.4 using a desalting column. This step also removed residual nickel, EDTA, and DTT that were added as described above. After desalting, IAP, MTSL, or MSL nitroxide spin label, dissolved in ethanol, was added in excess (5-10× molar ratio) and allowed to react with cysteine substituted constructs for 10-12 h at room temperature, in the dark. Spin labeling schemes are represented in Figure 2A. Excess spin label was removed by repeating the desalting buffer exchange process as described above.

    Figure 2. (A) Reaction schemes for chemical modification of a cysteine residue with various nitroxide probes. From top to bottom: IAP (3-(2-Iodoacetamido)-PROXYL) represented as P1, MTSL (2,2,5,5-tetramethyl-1-oxyl-3-methyl methanethiosulfonate) represented as R1, and MSL (3-maleimido-TEMPO). (B) A representative 100 G X-band CW-EPR spectrum with the three transition line pattern individually labeled: low field, h(+1), center field, h(0), and high field, h(−1). These three lines arise from the hyperfine interaction between the unpaired electron and the nitrogen nucleus of the spin label.


    2.4. Sample preparation for CW-EPR experiments

    For each spin labeled construct, nine separate samples were prepared with increasing 5% (v/v) increments of 2,2,2-trifluoroethanol (TFE) that spanned 0-40%. Each individual sample was prepared at a final volume of 500 µL: 300 µL of 50 mM sodium phosphate, 300 mM sodium chloride, pH 7.4 buffer, (200−x) µL of TFE, where x ranged from 0-200 µL, and x µL of appropriately prepared solution of 5× stock buffer diluted with water to keep all samples at equal ionic strength and pH. These larger 500 µL sample volumes were prepared to help mitigate pipetting errors of small volumes. Samples of ∼10 µL were drawn into 0.60 i.d. × 0.84 o.d. capillary tubes, which were then flame sealed on both ends ensuring minimal to no TFE evaporation.


    2.5. Sample preparation for CD experiments

    For each spin labeled construct, samples were prepared with varying percentages of increasing TFE concentrations. For IAP labeled cysteine scanning and the V8C construct with varying spin label attachments, three samples were prepared with increasing 15% (v/v) increments of TFE from 0-30%. For amino acid scanning at the 8th position of the IA3 sequence and cysteine or alanine substitution at the 11th position, six or nine samples were prepared with increasing 7% (v/v) increments of TFE from 0-35% or increasing 5% (v/v) increments of TFE from 0-40%, respectively. Samples were prepared with 80-120 μL of protein sample in 50 mM sodium phosphate, 300 mM sodium chloride, pH 7.4 to reach a final concentration of 10 μM protein in 1000 μL. The final volume was adjusted using 10 mM sodium phosphate, pH 7.0 buffer or TFE to reach appropriate % (v/v) TFE. For the V8C un-labeled construct, 0.1 mM DTT was present to prevent disulfide bonding of free cysteine residues.


    2.6. CW X-band EPR data collection and analysis

    Either a Bruker ER200 spectrometer with an ER023 M signal channel, an ER032 M field control unit or a Bruker E500 with a loop gap resonator (Medical Advances, Milwaukee, WI) was used to collect CW X-band EPR spectra. For all experiments the temperature was kept at a constant 27 ± 0.2 °C using a nitrogen gas passed through a copper coil submersed into a water bath, while monitored with a temperature probe and thermometer from OMEGA Engineering Inc (Norwalk, CT). All spectra are reported as an average of four scans, collected as 100 G sweep width, 0.6 modulation amplitude, 70 ms conversion time, 100 kHz modulation amplitude, and 2 mW incident microwave power. Spectra were normalized using labVIEW software allowing for baseline correction and double integral area normalization. A representative nitroxide EPR spectrum with the typical three-line transition pattern is shown and labeled in Figure 2B. Spectral line shapes were analyzed by calculating the ratio of the intensities of the low field and center field transitions, h(+1)/h(0) as a function of TFE percentage. Although the low field and center field transitions individually are not as sensitive to motion as the high field transition (h(−1)), this ratio has been previously shown to change considerably as the unstructured-to-structured transition occurs and is not sensitive to errors in baseline correction or double integration [17,36]. We have used this analysis previously to monitor the unstructured to helical transition in IA3 by % TFE. Plots of h(+1)/h(0) values versus the % TFE are sigmoidal in shape and are well fit by a two-state Boltzmann transition, given by Eq 1.

    y=(A1A2)1+e(xx0)/dx+A2 (1)

    A1 is the initial value of the curve, A2 is the final value of the curve and x0 is the midpoint of the curve—which corresponds well to the midpoint of folding transitions observed by CD and NMR [14]. Data were fit using Origin 8.5 software.


    2.7. CD data collection and analysis

    CD measurements were collected using an AVIV model 202 CD spectrometer set at 27 °C. To collect the spectra, 400 μL of each sample were loaded into a 1mm path length quartz cuvette, cleaned between runs with nanopure H2O and ethanol. Typically, four scans of each buffer blank and sample were collected between wavelengths of 200-250 nm with 1 nm increments and a four second averaging time. Each set of CD scans for each construct was averaged, and the averaged buffer baseline was subtracted from each individual construct averaged scan. Units were converted to Δε (molar circular dichroism, M−1 cm−1) from the instruments output unit of Θ (ellipticity, mdeg) using Eq 2.

    Δε (M1cm1)=Θ (mdeg)xCxL100xM (2)

    Θ is the output ellipticity value from the CD measurement, C is the concentration of sample in g/L, L is the path length in cm, and M is the average molecular weight in g/mol.


    3. Results and discussion


    3.1. SDSL CW-EPR spectroscopy reveals sites where spin-label incorporation alters helicity

    It is well known that the labeling of a naturally occurring or substituted cysteine may impact the native function of a protein. To understand if spin-label and cysteine substitution within IA3 impacted folding or folding propensity, cysteine scanning was performed across the N- and C-termini of IA3. Sites K7, V8, S9, E10, I11, F12, Q13, S14, S15, and S27 were chosen within the N-terminus and span two helical turns containing both solvent exposed and buried residues in the α-helix when bound to YPRA. Sites Y57, N58, K59, L60, and K61, were selected in the C-terminus (Figure 1).

    CW X-band EPR spectra were collected for the fifteen IAP spin labeled-IA3 variants (referred to within as P1-IA3) as a function of TFE concentration that ranged from 0-40% (v/v) TFE. Figure 3 shows representative CW-EPR spectra for 0%, 15%, and 30% (v/v) TFE for each P1-IA3 variant. Spectra are plotted with normalized double integral area to allow for easy visualization of changes in mobility. As mobility (defined as both the rate of motion and order of motion) decreases, the line shape broadens. As a result, when plotted with normalized area, the intensity appears to be less. A decrease in mobility upon increasing TFE percentage is observed for nearly all sites and indicates a conformational change upon increasing TFE concentration (earlier work has ruled out the impact of increased solution viscosity) [17]. Inspection of the line shapes shows that the change in mobility differs among these 15 P1-IA3 variants, indicating that some spin-labeled cysteine substitutions affect the conformation of each state, altering the TFE-induced helical propensity of IA3.

    Figure 3. Stack plots of absorption area normalized 100 G X-band CW-EPR spectra of P1-IA3 variants in the presence of 0%, 15%, and 30% (v/v) TFE, at 27 ± 0.2 °C. Spectra are vertically offset for visual clarity.

    A more quantitative way to analyze the TFE-induced conformational changes for each variant is to plot h(+1)/h(0) values as a function of TFE % (Figure 4). For isotropic motion, the h(+1)/h(0) value is expected to be 1. As the mobility decreases as the protein becomes helical, lower values of h(+1)/h(0) are expected. In the unfolded state (absence of TFE), all P1-IA3 variants have similar CW-EPR spectra indicative of similar high nitroxide mobility (indicated by values of h(+1)/h(0) ∼ 0.9). As a function of TFE concentration, the shape of each h(+1)/h(0) plot can be well fit by Eq 1, indicative of a two-state transition.

    Several conclusions can be drawn from the data in Figure 4. Firstly, the sigmoidal behavior of the h(+1)/h(0) parameter for N-terminal P1-IA3 variants (gray shade, solid diamonds) varies both in sharpness and extent of transition, which is defined by the limiting h(+1)/h(0) value. In contrast C-terminal P1-IA3 variants (no shade, open circles) exhibit similar behavior, both in the sharpness and extent of transition. Secondly, there are site specific variations in the behavior of the h(+1)/h(0) parameter in the N-terminus that follow a helical trend that maps to the surface of IA3 in the YRPA bound site. Specifically, sites V8P1, I11P1, and S15P1 which are located on the concave buried face of the peptide when bound to YPRA, have small changes (<0.05 over the 0-40% (v/v) TFE range) in their h(+1)/h(0) values. In contrast, sites S9P1, Q13P1 and S27P1, which reside on the convex solvent exposed face when bound to YPRA, pose large changes in their h(+1)/h(0) values (>0.3 over the 0-40% TFE range). Figure 5 shows the location of these residues in the IA3: YPRA complex crystal structure. These results indicate a sensitivity of the N-terminal helicity to amino acid substitution.

    Figure 4. Plots of h(+1)/h(0) values as a function of % (v/v) TFE for N-terminus (gray shade, solid diamonds) and C-terminus (no shade, open circles) P1-IA3 variants. Solid lines through points are best fits of a two-state Boltzmann function (Eq 1), consistent with a two-state transition of IA3 from unstructured to α-helical; as described previously [17]. The size of each data point is larger than the standard deviation of triplicate measurements of the same sample.

    Figure 5. (Top) Ribbon diagram of IA3 (dark gray) bound to YPRA (light gray), with chosen P1-labeled variants of solvent exposed and YPRA facing, buried residues highlighted. Solvent exposed highlighted residues are shown in orange, and buried residues are rendered in green. (Bottom) Bound IA3 structure, with YPRA removed. Buried residues located on concave side of structure shown in green, and solvent exposed residues located on convex side of structure shown in orange (PDB: 1DPJ).

    To further explore these observations, CD spectroscopy was performed on each of the N-terminal P1-IA3 variants with 0%, 15%, and 30% (v/v) TFE. Data are shown in Figure 6, which also overlays results from WT IA3 (gray solid line) in 30% (v/v) TFE for comparison. For wild type IA3, CD data reveal a predominantly random coil conformation for 0% (v/v) TFE, whereas with 30% (v/v) TFE the spectra reflect a mostly helical conformation with features at 208 nm and 222 nm indicative of helix formation [37,38]. For all N-terminal P1-IA3 variants, CD data confirm that in the absence of TFE, all constructs adopt a random coil conformation and that the addition of TFE induces a helical conformation. As was observed with the CW-EPR data, the degree of the helical transition, as reflected in the values of Δε222nm in CD spectra, varies when compared to WT for the P1 variants. It can be seen that I11P1 and S14P1 possess CD spectra that very closely match that of WT (i.e., represent SL wild-type mimics). V8P1 is found to be less helical than WT, whereas all other variants have greater helicity than WT. To directly compare the SDSL CW-EPR characterized unstructured-to-structured transition within the N-terminus to the degree of α-helicity from CD spectroscopy, we plotted the values of Δh(+1)/h(0), or degree of transition (h(+1)/h(0) (40% (v/v) TFE)h(+1)/h(0) (0% (v/v) TFE)) against the Δε222nm (30% (v/v) TFE) (Eq 2) values from CD spectra for each P1-IA3 variant (Figure 7). The Δε222nm value for WT IA3 is displayed in Figure 7 as a gray line to represent how the helical conformation of each variant compares to WT behavior. As can be seen, there is remarkable agreement in overall trend between these two parameters, confirming that the helical propensity of the N-terminal region in IA3 is particularly sensitive to amino acid substitution/spin-labeling. It is striking that CW-EPR line shape analysis readily reflects this sensitivity to amino acid substitution/modification and in most cases the incorporation of the spin-label is found to enhance the helicity of IA3.

    Figure 6. Circular dichroism spectra for N-terminal P1-IA3 variants with WT spectra overlay at 30% (v/v) TFE concentration (solid gray line). P1-IA3 variant spectra were collected with increasing % TFE concentrations at 27 °C as an average of four scans; 0% (v/v) TFE (dotted black line), 15% (v/v) TFE (dashed black line), and 30% (v/v) TFE (solid black line).

    Figure 7. Comparison of helical propensity Δ(h(+1)/h(0)) (black squares) to the Δε (M−1 cm−1) at 222 nm (black circles) at 30% (v/v) TFE. Where error bars are not visible, the size of data point is larger than the standard deviation of triplicate measurement on the same sample.


    3.2. Helical propensity at site V8 in IA3 is modulated by amino acid substitution and size

    Because the P1 substitution at site V8 revealed less helical character than WT, we proceeded to generate a series of substitutions at this site to probe how amino acid size and type impacted helical propensity. The following constructs were chosen: V8A and V8I (smaller and larger than V; respectively), V8P (typical helical disrupter), V8D (charged), and V8C (our cysteine mutation without spin-label). CD spectroscopy was performed on each of the V8 variants with 0-40% or 0-35% (v/v) TFE (Figures S1 and S2, respectively). CD spectra of each construct in 30% (v/v) TFE are given in Figure 8A. It is found that TFE does induce a helical transition in each of these constructs. We assessed the impact of each substitution on helical content by calculating the difference in Δε222nm of the variant to WT. Figure 8B plots values of Δ(Δε222nm), showing that, as expected, the proline substitution at site 8 is less helical than WT. Also as expected, the alanine substitution induces the greatest increase in helical content, whereas other substitutions resulted in near WT behavior.

    Figure 8. CD data and analysis of constructs compared to WT. (A) CD spectra of V8 variants (V8P, V8D, V8I, V8C, and V8A). (B) ΔΔε of V8 variants compared to WT. (C) CD spectra of V8C-spin label variants (V8P1, V8MSL, and V8MR1). (D) ΔΔε for V8C-spin labeled variants. (E) CD spectra for V8-I11 variants (V8P1, V8A-I11P1, and V8P1-I11A). (F) ΔΔε for V8-I11 variants. Red CD spectra of WT are added in each plot and where ΔΔε (M−1 cm−1) is calculated as Δε222nm (WT) − Δε222nm (variant) so that positive values represent more helical character and negative values indicate a decreased helicity.


    3.3. Choice of spin-label at site V8C alters helical behavior of IA3

    The result that V8C contained WT helicity was surprising, which prompted us to examine if and how spin-label structure influenced helicity when incorporated at this site. CD spectroscopy was performed on the V8C spin-labeled variants from 0-40% or 0-30% (v/v) TFE (Figures S1 and S3, respectively). Interestingly, we found that the P1 substitution disrupted helicity compared to WT, MSL-CYS behaved like WT, and R1 substitution slightly increased helical content (Figure 8C, D). Perhaps the structure of the P1 label, which contains a pseudo-peptide bond interferes with TFE induced folding. MSL is a bulkier label than P1, and R1 is oftentimes considered a compact label because of potential sulfur interactions with the protein backbone [39].

    Nevertheless, results appear to indicate a dependency of structural bulkiness for disruption or enhancement of helical propensity. Indeed, the structure of IA3 bound to YPRA shows a slight concave bend on the buried interface, making substitution to larger residues disrupt helical behavior and conversely, a smaller residue would enhance helicity. To test this idea, we made two additional constructs, V8P1-I11A and V8A-I11P1, which would place the smaller alanine residue next to the P1 spin-label. CD spectroscopy was performed on each additional construct from 0-40% or 0-35% (v/v) TFE (Figures S1 and S2, respectively). V8P1-I11A shows nearly WT helical behavior and although V8A-I11P1 is less helical than WT, it is more helical than V8P1 (Figure 8E, F).


    4. Conclusions

    It was determined through SDSL CW-EPR that the extent of helical order within the N-terminal region of IA3 is easily perturbed by amino acid substitution and IAP labeling. The C-terminal region did not have similar susceptibility, where changes in labeled amino acid substitution did not reveal drastic changes in the unstructured-to-structured transition, and as we showed previously, retained WT behavior [17]. CD results confirmed that the helical order within the N-terminus varied due to the location of the altered residue in relation to the buried or solvent-exposed, or the concave or convex face of the protein, respectively. The concave face of IA3, is negatively affected by cysteine mutation and labeling, whereas the convex surface increases in helical order. It is proposed that the change in helical propensity on either side of the protein is due to spatial availability when incorporating a cysteine residue and nitroxide probe. Residue V8, located on the concave side of the N-terminal region showed the greatest negative perturbation due to cysteine substitution, and spin labeling. Mutations to the V8 residue, including substitutions in residue, charge, and size did not drastically affect the helical propensity, but rather the addition of the IAP nitroxide probe had the greatest effect on helical formation.

    These experiments demonstrate the care that must be taken when spin-labeling IDPs. Proper checks-and-balances are needed to ensure that a labeled site retains WT helical propensity (in addition to any other functional assays). Overall, the results identify key sites for spin-label incorporation that mimic WT helical tendencies that can be used in future studies of IA3 conformational sampling studies.


    Acknowledgements

    This work was supported by NSF MCB 1329467 & MCB 1715384 and NIH 1S10-RR031603-01 to GEF. The authors would like to thank Stephen J. Hagen and Robert McKenna for access to the Aviv CD spectrometer, and Christian Altenbach for Labview software to analyze the EPR data.


    Conflict of interest

    All authors declare no conflict of interest in this paper.




    [1] Y. Al-Rashid, L. D. Paarmann, Short-term electric load forecasting using neural network models, in: Proc. 39th Midwest Symp. Circuits Syst., IEEE, (1996), 1436–1439.
    [2] T. Hong, P. Pinson, S. Fan, H. Zareipour, A. Troccoli, R. J. Hyndman, Probabilistic energy forecasting: Global Energy Forecasting Competition 2014 and beyond, Int. J. Forecast., 32 (2016), 896–913. doi: 10.1016/j.ijforecast.2016.02.001
    [3] T. Hong, P. Pinson, S. Fan, Global Energy Forecasting Competition 2012, Int. J. Forecast., 30 (2014), 357–363. doi: 10.1016/j.ijforecast.2013.07.001
    [4] L. Hussain, M.S. Nadeem, S. A. A.Shah, Short term load forecasting system based on Support Vector Kernel methods, Int. J. Comput. Sci. Inf. Technol., 6 (2014), 93–102.
    [5] G. Gross, F. D. Galiana, Short-term load forecasting, Proc. IEEE., 75 (1987), 1558–1573. doi: 10.1109/PROC.1987.13927
    [6] P. J. Santos, A. G. Martins, A. J. Pires, Short-term load forecasting based on ANN applied to electrical distribution substations, in: 39th Int. Univ. Power Eng. Conf. UPEC 2004 - Conf. Proc., IEEE, 1(2004), 427–432.
    [7] E. Hossain, I. Khan, F. Un-Noor, S. S. Sikander, M. S. H. Sunny, Application of big data and machine learning in smart grid, and associated security concerns: A Review, IEEE Access, 7 (2019), 13960–13988. doi: 10.1109/ACCESS.2019.2894819
    [8] H. J. Sadaei, P. C. D.L. e Silva, F. G. Guimarães, M. H. Lee, Short-term load forecasting by using a combined method of convolutional neural networks and fuzzy time series, Energy, 175 (2019), 365–377. doi: 10.1016/j.energy.2019.03.081
    [9] H. M. Al-Hamadi, S. A. Soliman, Short-term electric load forecasting based on Kalman filtering algorithm with moving window weather and load model, Electr. Power Syst. Res., 68 (2004), 47–59. doi: 10.1016/S0378-7796(03)00150-0
    [10] K.-B. Song, Y.-S. Baek, D. H. Hong, G. Jang, Short-term load forecasting for the holidays using fuzzy linear regression method, IEEE Trans. Power Syst., 20 (2005), 96–101. doi: 10.1109/TPWRS.2004.835632
    [11] D. K. Ranaweera, N. F. Hubele, G. G. Karady, Fuzzy logic for short term load forecasting, Int. J. Electr. Power Energy Syst., 18 (1996), 215–222. doi: 10.1016/0142-0615(95)00060-7
    [12] Y. He, Q. Xu, J. Wan, S. Yang, Short-term power load probability density forecasting based on quantile regression neural network and triangle kernel function, Energy, 114 (2016), 498–512. doi: 10.1016/j.energy.2016.08.023
    [13] M. Q. Raza, A. Khosravi, A review on artificial intelligence based load demand forecasting techniques for smart grid and buildings, Renew. Sustain. Energy Rev., 50 (2015), 1352–1372. doi: 10.1016/j.rser.2015.04.065
    [14] L. C. P. Velasco, C. R. Villezas, P. N. C. Palahang, J. A. A. Dagaang, Next day electric load forecasting using Artificial Neural Networks, in: 2015 Int. Conf. Humanoid, Nanotechnology, Inf. Technol. Control. Environ. Manag., IEEE, 2015, 1–6.
    [15] L. Hernández, C. Baladrón, J. M. Aguiar, L. Calavia, B. Carro, A. Sánchez-Esguevillas, et al., Artificial Neural Network for short-term load forecasting in distribution systems, Energies, 7 (2014), 1576–1598. doi: 10.3390/en7031576
    [16] J. Buitrago, S. Asfour, Short-term forecasting of electric loads using nonlinear autoregressive Artificial Neural Networks with exogenous vector inputs, Energies, 10 (2017), 40. doi: 10.3390/en10010040
    [17] L. Suganthi, S. Iniyan, A. A. Samuel, Applications of fuzzy logic in renewable energy systems- A review, Renew. Sustain. Energy Rev., 48 (2015), 585–607. doi: 10.1016/j.rser.2015.04.037
    [18] K-H. Kim, J-K. Park, K-J. Hwang, S-H. Kim, Implementation of hybrid short-term load forecasting system using artificial neural networks and fuzzy expert systems, IEEE Trans. Power Syst., 10 (1995), 1534–1539. doi: 10.1109/59.466492
    [19] D. X. Niu, H. F. Shi, D. D. Wu, Short-term load forecasting using bayesian neural networks learned by Hybrid Monte Carlo algorithm, Appl. Soft Comput., 12 (2012), 1822–1827. doi: 10.1016/j.asoc.2011.07.001
    [20] F. Kaytez, M. C. Taplamacioglu, E. Cam, F. Hardalac, Forecasting electricity consumption: A comparison of regression analysis, neural networks and least squares support vector machines, Int. J. Electr. Power Energy Syst., 67 (2015), 431–438. doi: 10.1016/j.ijepes.2014.12.036
    [21] X. Li, X. Wu, Constructing long short-term memory based deep recurrent neural networks for large vocabulary speech recognition, in: 2015 IEEE Int. Conf. Acoust. Speech Signal Process, IEEE, 2015, 4520–4524.
    [22] R. J. Williams, D. Zipser, A learning algorithm for continually running fully recurrent Neural Networks, Neural Comput., 1 (1989), 270–280. doi: 10.1162/neco.1989.1.2.270
    [23] M. Ghofrani, M. Ghayekhloo, A. Arabali, A. Ghayekhloo, A hybrid short-term load forecasting with a new input selection framework, Energy, 81 (2015), 777–786. doi: 10.1016/j.energy.2015.01.028
    [24] G.-F. Fan, L.-L. Peng, W.-C. Hong, Short term load forecasting based on phase space reconstruction algorithm and bi-square kernel regression model, Appl. Energy, 224 (2018), 13–33. doi: 10.1016/j.apenergy.2018.04.075
    [25] R. Mamlook, O. Badran, E. Abdulhadi, A fuzzy inference model for short-term load forecasting, Energy Policy, 37 (2009), 1239–1248. doi: 10.1016/j.enpol.2008.10.051
    [26] K. Metaxiotis, A. Kagiannas, D. Askounis, J. Psarras, Artificial intelligence in short term electric load forecasting: a state-of-the-art survey for the researcher, Energy Convers. Manag., 44 (2003), 1525–1534. doi: 10.1016/S0196-8904(02)00148-6
    [27] K. Fukushima, S. Miyake, A self-organizing neural network model for a mechanism of visual pattern recognition, in: Comp. cooperation Neural Net., Springer Berlin Heidelb., 1982,267–85.
    [28] Y. LeCun, P. Haffner, L. Bottou, Y. Bengio, Object recognition with gradient-based learning, in: Shape, Cont. grop. comp. vision., Springer Berlin Heidelb., 1999,319–345.
    [29] Y. Guo, Y. Liu, A. Oerlemans, S. Lao, S. Wu, Y. Guo, et al., Deep learning for visual understanding: A review, Neurocomputing, 187 (2016), 27–48. doi: 10.1016/j.neucom.2015.09.116
    [30] H.-K. Yu, Weighted fuzzy time series models for TAIEX forecasting, Phys. A Stat. Mech. Appl., 349 (2005), 609–624. doi: 10.1016/j.physa.2004.11.006
    [31] N.-Y. Wang, S.-M. Chen, Temperature prediction and TAIFEX forecasting based on automatic clustering techniques and two-factors high-order fuzzy time series, Expert Syst. Appl., 36 (2009), 2143–2154. doi: 10.1016/j.eswa.2007.12.013
    [32] W.-J. Lee, J. Hong, A hybrid dynamic and fuzzy time series model for mid-term power load forecasting, Int. J. Electr. Power Energy Syst., 64 (2015), 1057–1062. doi: 10.1016/j.ijepes.2014.08.006
    [33] R. Efendi, Z. Ismail, M. M. Deris, A new linguistic out-sample approach of fuzzy time series for daily forecasting of Malaysian electricity load demand, Appl. Soft Comput., 28 (2015), 422–430. doi: 10.1016/j.asoc.2014.11.043
    [34] R. Enayatifar, H. J. Sadaei, A. H. Abdullah, A. Gani, Imperialist competitive algorithm combined with refined high-order weighted fuzzy time series (RHWFTS–ICA) for short term load forecasting, Energy Convers. Manag., 76 (2013), 1104–1116. doi: 10.1016/j.enconman.2013.08.039
    [35] L. Hussain, W. Aziz, J. S. Alowibdi, N. Habib, M. Rafique, S. Saeed, et al., Symbolic time series analysis of electroencephalographic (EEG) epileptic seizure and brain dynamics with eye-open and eye-closed subjects during resting states, J. Physiol. Anthropol., 36 (2017), 21. doi: 10.1186/s40101-017-0136-8
    [36] L. Hussain, Detecting epileptic seizure with different feature extracting strategies using robust machine learning classification techniques by applying advance parameter optimization approach, Cogn. Neurodyn., 12 (2018), 271–294. doi: 10.1007/s11571-018-9477-1
    [37] L. Hussain, I. A. Awan, W. Aziz, S. Saeed, A. Ali, F. Zeeshan, et al., Detecting congestive heart failure by extracting multimodal features and employing machine learning techniques, Biomed. Res. Int., 2020 (2020), 1–19.
    [38] Y. Asim, B. Raza, A. Kamran, M. Saima, A. K. Malik, S. Rathore, et al., A multi-modal, multi-atlas-based approach for Alzheimer detection via machine learning, Int. J. Imag. Syst. Tech., 28 (2018), 113–123. doi: 10.1002/ima.22263
    [39] L. Hussain, S. Saeed, I. A. Awan, A. Idris, M. S. A. Nadeem, Q.-A. Chaudhry, et al., Detecting brain tumor using machines learning techniques based on different features extracting strategies, Curr. Med. Imag. Rev., 15 (2019), 595–606. doi: 10.2174/1573405614666180718123533
    [40] L. Hussain, W. Aziz, S. Saeed, I. A. Awan, A. A. Abbasi, N. Maroof, Arrhythmia detection by extracting hybrid features based on refined Fuzzy entropy (FuzEn) approach and employing machine learning techniques, Waves Rand. Complex Media, 30 (2020), 656–686. doi: 10.1080/17455030.2018.1554926
    [41] L. Hussain, W. Aziz, A. A. Alshdadi, M. S. A. Nadeem, I. R. Khan, Q.-A. Chaudhry, Analyzing the dynamics of lung cancer imaging data using refined fuzzy entropy methods by extracting different features, IEEE Access, 7 (2019), 64704–64721. doi: 10.1109/ACCESS.2019.2917303
    [42] A. A. Abbasi, L. Hussain, I. A. Awan, I. Abbasi, A. Majid, M. S. A. Nadeem, Q. -A. Chaudhary, Detecting prostate cancer using deep learning convolution neural network with transfer learning approach, Cogn. Neurodyn., 5 (2020), 1–11.
    [43] L. Hussain, A. Ahmed, S. Saeed, S. Rathore, I. A. Awan, S. A. Shah, et al., Prostate cancer detection using machine learning techniques by employing combination of features extracting strategies, Cancer Biomark., 21 (2018), 393–413. doi: 10.3233/CBM-170643
    [44] C. Lang, F. Steinborn, O. Steffens, E. W. Lang, Applying a 1D-CNN network to electricity load forecasting, in: Int. conf. time series forecas., Springer, Cham, 2020,205–218.
    [45] N. Javaid, NADEEM: A Novel Reliable Data Delivery Routing Protocol for Underwater WSNs, in: Works. Int. Conf. Adv. Inf. Netw. App., Springer, Cham, 2019,103–115.
    [46] D. Hussain, T. Hussain, A. A. Khan, S. A. A. Naqvi, A. Jamil, A deep learning approach for hydrological time-series prediction: A case study of Gilgit river basin, Earth Sci. Inform., 13 (2020), 915–927. doi: 10.1007/s12145-020-00477-2
    [47] A.M. Tudose, D.O. Sidea, I. I. Picioroaga, V. A. Boicea, C. Bulac, A CNN Based Model for Short-Term Load Forecasting: A Real Case Study on the Romanian Power System, in: 2020 55th Int. Univ. Power Eng. Conf., IEEE, 2020, 1–6.
    [48] M. Solas, N. Cepeda, J. L. Viegas, Convolutional Neural Network for Short-term Wind Power Forecasting, in: 2019 IEEE PES Innov. Smart Grid Technol. Eur., IEEE, 2019, 1–5.
    [49] A. E. Khantach, M. Hamlich, N. E. Belbounaguia, Short-term load forecasting using machine learning and periodicity decomposition, AIMS Energy, 7 (2019), 382–394. doi: 10.3934/energy.2019.3.382
    [50] G. Dudek, Forecasting time series with multiple seasonal cycles using neural networks with local learning, in: Int. Conf. Artif. Intell. Soft Comput., Springer, Berlin, Heidelberg., 2013, 52–63.
    [51] M. A. Ghorbani, R. C. Deo, Z. M. Yaseen, M. H. Kashani, B. Mohammadi, Pan evaporation prediction using a hybrid multilayer perceptron-firefly algorithm (MLP-FFA) model: Case study in North Iran, Theoret. App. Climat., 133 (2018), 1119–1131. doi: 10.1007/s00704-017-2244-0
    [52] M. E. Günay, Forecasting annual gross electricity demand by artificial neural networks using predicted values of socio-economic indicators and climatic conditions: Case of Turkey, Energy Policy, 90 (2016), 92–101.
    [53] G. Dudek, Multilayer perceptron for short-term load forecasting : From global to local approach, Neural Comput. Appl., 32 (2020), 3695–3707. doi: 10.1007/s00521-019-04130-y
    [54] X. Glorot, Y. Bengio, Understanding the difficulty of training deep feedforward neural networks, in: Proc. thirteenth int. conf. artif. intell. stat., 2010,249–256.
    [55] S. Hochreiter, J Schmidhuber, Long short-term memory, Neural Comput., 9 (1997), 1735–1780. doi: 10.1162/neco.1997.9.8.1735
    [56] H. Zheng, J. Yuan, L. Chen, Short-term load forecasting using EMD-LSTM neural networks with a Xgboost algorithm for feature importance evaluation, Energies, 10 (2017), 1168. doi: 10.3390/en10081168
    [57] Y. Zhao, Y. Shen, Y. Zhu, J. Yai, Forecasting wavelet transformed time series with attentive neural networks, in: 2018 IEEE Int. Conf. Data Min., IEEE, 2018, 1452–1457.
    [58] K. Greff, R. K. Srivastava, J. Koutnik, B. R. Steunebrink, J. Schmidhuber, LSTM: A Search Space Odyssey, IEEE Trans. Neural Networks Learn. Syst., 28 (2016), 2222–2232.
    [59] P. Nagabushanam, S. T. George, S. Radha, EEG signal classification using LSTM and improved neural network algorithms, Soft Comput., 24 (2020), 9981–10003. doi: 10.1007/s00500-019-04515-0
    [60] V. Y. Senyurek, M. H. Imtiaz, P. Belsare, S. Tiffany, E. Sazonov, A CNN-LSTM neural network for recognition of puffing in smoking episodes using wearable sensors, Biomed. Eng. Lett., 10 (2020), 195–203. doi: 10.1007/s13534-020-00147-8
    [61] A. Graves, J. Schmidhuber, Framewise phoneme classification with bidirectional LSTM and other neural network architectures, Neural Networks, 18 (2005), 602–610. doi: 10.1016/j.neunet.2005.06.042
    [62] A. Graves, Long Short-Term Memory, in: Sup. Seq. Label.Recurr. Neural Networks. Stud. Comput. Intell., Springer Berlin Heidelberg, 2012, 37–45.
    [63] X. Ma, Z. Tao, Z. Wang, H. Yu, Y. Wang, Long short-term memory neural network for traffic speed prediction using remote microwave sensor data, Transp. Res. Part C Emerg. Technol., 54 (2015), 187–197. doi: 10.1016/j.trc.2015.03.014
    [64] E. Shelhamer, J. Long, T. Darrell, Fully convolutional networks for semantic segmentation, IEEE Trans. Pattern Anal. Mach. Intell., 39 (2017), 640–651. doi: 10.1109/TPAMI.2016.2572683
    [65] Y. Taigman, M. Yang, M. Ranzato, L. Wolf, DeepFace: Closing the gap to human-level performance in face verification, in: Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., 2014, 1701–1708.
    [66] A. Karpathy, G. Toderici, S. Shetty, T. Leung, R. Sukthankar, L. Fei-Fei, Large-scale video classification with convolutional neural networks, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. IEEE, 2014, 1725–1732.
    [67] T. Young, D. Hazarika, S. Poria, E. Cambria, Recent trends in deep learning based natural language processing, IEEE Comput. Intell. Mag., 13 (2018), 55–75. doi: 10.1109/MCI.2018.2840738
    [68] M. Cai, M. Pipattanasomporn, S. Rahman, Day-ahead building-level load forecasts using deep learning vs. traditional time-series techniques, Appl. Energy, 236 (2019), 1078–1088.
    [69] M. Canizo, I. Triguero, A. Conde, E. Onieva, Multi-head CNN–RNN for multi-time series anomaly detection: An industrial case study, Neurocomputing, 363 (2019), 246–260. doi: 10.1016/j.neucom.2019.07.034
    [70] Z. Zeng, H. Xiao, X. Zhang, Self CNN-based time series stream forecasting, Electron. Lett., 52 (2016), 1857–1858. doi: 10.1049/el.2016.2626
    [71] Y. Lecun, L. Bottou, Y. Bengio, P. Haffner, Gradient-based learning applied to document recognition, Proc. IEEE, 86 (1998), 2278–2324. doi: 10.1109/5.726791
    [72] L. Atlas, T. Homma, R. Marks, An artificial neural network for spatio-temporal bipolar patterns: Application to phoneme classification, in: Neural Inf. Process, 1988, 31–40.
    [73] J. Günther, P. M. Pilarski, G. Helfrich, H. Shen, K. Diepold, First steps towards an intelligent laser welding architecture using deep neural networks and reinforcement learning, Procedia Technol., 15 (2014), 474–483. doi: 10.1016/j.protcy.2014.09.007
    [74] N. Amjady, F. Keynia, Short-term load forecasting of power systems by combination of wavelet transform and neuro-evolutionary algorithm, Energy, 34 (2009), 46–57. doi: 10.1016/j.energy.2008.09.020
    [75] L. Hussain, S. Saeed, A. Idris, I. A. Awan, S. A. Shah, A. Majid, et al., Regression analysis for detecting epileptic seizure with different feature extracting strategies, Biomed. Eng. Biomed. Tech., 64 (2019), 619–642.
    [76] S. Du, T. Li, X. Gong, Y. Yang, S.J. Horng, Traffic flow forecasting based on hybrid deep learning framework, in: 2017 12th Int. Conf. Intell. Syst. Knowl. Eng., IEEE, 2017, 1–6.
    [77] P.-H. Kuo, C.-J. Huang, A high precision artificial neural networks model for short-term energy load forecasting, Energies, 11 (2018), 213. doi: 10.3390/en11010213
    [78] G. M. U. Din, A. K. Marnerides, Short term power load forecasting using Deep Neural Networks, in: 2017 Int. Conf. Comput. Netw. Commun., IEEE, (2017), 594–598.
    [79] S. Fan, L. Chen, Short-term load forecasting based on an adaptive hybrid method, IEEE Trans. Power Syst., 21 (2006), 392–401. doi: 10.1109/TPWRS.2005.860944
    [80] T. Ouyang, Y. He, H. Li, Z. Sun, S. Baek, Modeling and forecasting short-term power load with copula model and deep belief network, IEEE Trans. Emerg. Top. Comput. Intell., 3 (2019), 127–136. doi: 10.1109/TETCI.2018.2880511
    [81] X. Wang, F. Fang, X. Zhang, Y. Liu, L. Wei, Y. Shi, LSTM-based Short-term Load Forecasting for Building Electricity Consumption, in: 2019 IEEE 28th Int. Symp. Ind. Electron., IEEE, 2019, 1418–1423.
    [82] F. Rodrigues, C. Cardeira, J. M. F. Calado, The daily and hourly energy consumption and load forecasting using artificial neural network method: A case study using a set of 93 households in Portugal, Energy Proced., 62 (2014), 220–229. doi: 10.1016/j.egypro.2014.12.383
    [83] I. Shah, H. Iftikhar, S. Ali, Modeling and forecasting medium-term electricity consumption using component estimation technique, Forecasting, 2 (2020), 163–179. doi: 10.3390/forecast2020009
    [84] S. H. Rafi, N. Al-Masood, Short Term Electric Load Forecasting Using High Precision Convolutional Neural Network, in: 2020 Int. Conf. Comput. Electr. Commun. Eng., IEEE, 2020, 1–7.
    [85] S. H. Rafi, N. Al-Masood, Highly Efficient Short Term Load Forecasting Scheme Using Long Short Term Memory Network, in: 2020 8th Int. Electr. Eng. Congr., IEEE, (2020), 1–4.
    [86] R. A. Abbasi, N. Javaid, M. N. J. Ghuman, Z. A. Khan, S. Ur Rehman, Amanullah, Short Term Load Forecasting Using XGBoost, in: Work. Int. Conf. Adv. Info. Netw. App., (2019), 1120–1131.
  • This article has been cited by:

    1. Mohit Jaiswal, Trang T. Tran, Qingjiang Li, Xin Yan, Mingwei Zhou, Krishnendu Kundu, Gail E. Fanucci, Zhongwu Guo, A metabolically engineered spin-labeling approach for studying glycans on cells, 2020, 11, 2041-6520, 12522, 10.1039/D0SC03874A
    2. Katie M. Dunleavy, Collin Oi, Tianyan Li, Andrew Secunda, Afnan M. Jaufer, Yinlu Zhu, Lee Friedman, Alexander Kim, Gail E. Fanucci, Hydrogen Bonding Compensation on the Convex Solvent-Exposed Helical Face of IA3, an Intrinsically Disordered Protein, 2023, 0006-2960, 10.1021/acs.biochem.3c00169
    3. Katie M. Dunleavy, Tianyan Li, Eugene Milshteyn, Afnan M. Jaufer, Shamon A. Walker, Gail E. Fanucci, Charge Distribution Patterns of IA3 Impact Conformational Expansion and Hydration Diffusivity of the Disordered Ensemble, 2023, 1520-6106, 10.1021/acs.jpcb.3c06170
  • Reader Comments
  • © 2021 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(11273) PDF downloads(1424) Cited by(46)

Article outline

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog