Modified Support Vector Regression Research Articles

Predicting long-term electricity prices using modified support vector regression method

Predicting long-term electricity prices using modified support vector regression method

PSO-SVR-based voltage stability online monitoring method for large-scale power grids

With the power load increasing, the issue of voltage stability and voltage collapse is getting greater attention. Online monitoring of grid voltage stability is therefore a necessary requirement. In this paper, a PSO-SVR-based voltage stability online monitoring method for large-scale power grid is proposed. Initially, the load impedance modulus margin (LIMM) method is used to identify weak nodes in the system and thereby determine the best installation locations of the phase measurement units (PMUs). Afterwards, a particle swarm optimization modified support vector regression (PSO-SVR) algorithm is trained by using the state variables of the weak nodes where the PMUs are installed as input features and the calculated LIMM values as output features. In this way, the PSO-SVR model learns the non-linear relationship between the operating state of the power system and the corresponding LIMM values. The LIMM values of the nodes are obtained from the respective state-variables. The method proposed in this paper not only accurately assesses the stable voltage level of the load node, but also avoids the problems of parameter drift and misjudgement that can occur in the original LIMM method. Therefore, it has significantly advantages for online applications. Lastly, it is tested on an IEEE 118 system to verify the effectiveness and accuracy of the proposed method in this paper.

Oil temperature prediction of power transformers based on modified support vector regression machine

Abstract Power transformer is an important part of the entire power grid and the normal operation of the power transformer can ensure the normal operation of the entire power grid. The oil in the transformer plays a non-negligible role in the transformer. There are a lot of machine learning methods to predict oil temperature of power transformer. The work of this paper is to predict the oil temperature based on support vector regression machine (SVM) with three-phase power load, while particle swarm optimization (PSO) is employed for the model parameter optimization. As there are many influential factors for oil temperature prediction, confidence intervals are introduced to determine the prediction results. The experimental results show that the prediction accuracy reaches 90 with 85% confidence level. For the sample points falling outside the prediction interval, they can be regarded as the abnormal transformer status in time. The experimental results verified that the proposed oil temperature prediction method for power transformers based on modified SVM is effective and feasible.

PM2.5 Estimation and Spatial-Temporal Pattern Analysis Based on the Modified Support Vector Regression Model and the 1 km Resolution MAIAC AOD in Hubei, China

The satellite-retrieved Aerosol Optical Depth (AOD) is widely used to estimate the concentrations and analyze the spatiotemporal pattern of Particulate Matter that is less than or equal to 2.5 microns (PM2.5), also providing a way for the related research of air pollution. Many studies generated PM2.5 concentration networks with resolutions of 3 km or 10 km. However, the relatively coarse resolution of the satellite AOD products make it difficult to determine the fine-scale characteristics of PM2.5 distributions that are important for urban air quality analysis. In addition, the composition and chemical properties of PM2.5 are relatively complex and might be affected by many factors, such as meteorological and land cover type factors. In this paper, an AOD product with a 1 km spatial resolution derived from the Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm, the PM2.5 measurements from ground sites and the meteorological data as the auxiliary variable, are integrated into the Modified Support Vector Regression (MSVR) model that proposed in this paper to estimate the PM2.5 concentrations and analyze the spatiotemporal pattern of PM2.5. Considering the relatively small dataset and the somewhat complex relationship between the variables, we propose a Modified Support Vector Regression (MSVR) model that based on SVR to fit and estimate the PM2.5 concentrations in Hubei province of China. In this paper, we obtained Cross Correlation Coefficient (R²) of 0.74 for the regression of independent and dependent variables, and the conventional SVR model obtained R² of 0.60 as comparison. We think our MSVR model obtained relatively good performance in spite of many complex factors that might impact the accuracy. We then utilized the optimal MSVR model to perform the PM2.5 estimating, analyze their spatiotemporal patterns, and try to explain the possible reasons for these patterns. The results showed that the PM2.5 estimations retrieved from 1 km MAIAC AOD could reflect more detailed spatial distribution characteristics of PM2.5 and have higher accuracy than that from 3 km MODIS AOD. Therefore, the proposed MSVR model can be a better method for PM2.5 estimating, especially when the dataset is relatively small.

Equation Based New Methods for Residential Load Forecasting

This work proposes two non-linear and one linear equation-based system for residential load forecasting considering heating degree days, cooling degree days, occupancy, and day type, which are applicable to any residential building with small sets of smart meter data. The coefficients of the proposed nonlinear and linear equations are tuned by particle swarm optimization (PSO) and the multiple linear regression method, respectively. For the purpose of comparison, a subtractive clustering based adaptive neuro fuzzy inference system (ANFIS), random forests, gradient boosting trees, and long-term short memory neural network, conventional and modified support vector regression methods were considered. Simulations have been performed in MATLAB environment, and all the methods were tested with randomly chosen 30 days data of a residential building in Memphis City for energy consumption prediction. The absolute average error, root mean square error, and mean average percentage errors are tabulated and considered as performance indices. The efficacy of the proposed systems for residential load forecasting over the other systems have been validated by both simulation results and performance indices, which indicate that the proposed equation-based systems have the lowest absolute average errors, root mean square errors, and mean average percentage errors compared to the other methods. In addition, the proposed systems can be easily practically implemented.

Short-Term Load Forecasting of Microgrid via Hybrid Support Vector Regression and Long Short-Term Memory Algorithms

Short-Term Load Forecasting (STLF) is the most appropriate type of forecasting for both electricity consumers and generators. In this paper, STLF in a Microgrid (MG) is performed via the hybrid applications of machine learning. The proposed model is a modified Support Vector Regression (SVR) and Long Short-Term Memory (LSTM) called SVR-LSTM. In order to forecast the load, the proposed method is applied to the data related to a rural MG in Africa. Factors influencing the MG load, such as various household types and commercial entities, are selected as input variables and load profiles as target variables. Identifying the behavioral patterns of input variables as well as modeling their behavior in short-term periods of time are the major capabilities of the hybrid SVR-LSTM model. To present the efficiency of the suggested method, the conventional SVR and LSTM models are also applied to the used data. The results of the load forecasts by each network are evaluated using various statistical performance metrics. The obtained results show that the SVR-LSTM model with the highest correlation coefficient, i.e., 0.9901, is able to provide better results than SVR and LSTM, which have the values of 0.9770 and 0.9809, respectively. Finally, the results are compared with the results of other studies in this field, which continued to emphasize the superiority of the SVR-LSTM model.

A novel model modification method for support vector regression based on radial basis functions

There are some inherent limitations to the performance of support vector regression (SVR), such as (i) the loss function, penalty parameter, and kernel function parameter usually cannot be determined accurately; (ii) the training data sometimes cannot be fully utilized; and (iii) the local accuracy in the vicinity of training points still need to be improved. To further enhance the performance of SVR, this paper proposes a novel model modification method for SVR with the help of radial basis functions. The core idea of the method is to start with an initial SVR and modify it in a subsequent stage to extract as much information as possible from the existing training data; the second stage does not require new points. Four types of modified support vector regression (MSVR), including MSVg, MSVm, MSVi, and MSVc, are constructed by using four different forms of basis functions. This paper evaluates the performances of SVR, MSVg, MSVm, MSVi, and MSVc by using six popular benchmark problems and a practical engineering problem, which is designing a typical turbine disk for an air-breathing engine. The results show that all the four types of MSVR perform better than SVR. Notably, MSVc has the best performance.

An optical soft-sensor based shape sensing using a bio-inspired pattern recognition technique to realise fly-by-feel capability for intelligent aircraft operation

ABSTRACTInformation regarding deformations in large and complex systems is necessary in the prediction of structural failures caused by un-natural flexural occurrences. Sensing systems which are used to predict shapes, in order to develop a global surface picture require high precision and lower time lag. In this work, a unique bio-inspired training mechanism for support vector regression is presented for shape sensing in structures mounted with Fiber Bragg Gratings. Experimental validation was carried out on a simply supported beam, loaded at different positions and an aircraft wing model for different types of bending. The resulting deflections at specified locations along the length of the beam and on both surfaces of the wing were interpreted from the wavelength shifts of the corresponding Fiber Bragg Gratings through the specially modified Support Vector Regression. The method has shown high accuracy, low computational requirements and enhanced prediction times. The proposed bio-inspired training method has also been compared with two conventional training methodologies.

Research and application of a combined model based on variable weight for short term wind speed forecasting

Abstract Wind speed forecasting plays a prominent part in the operation of wind power plants and power systems. However, it is often difficult to obtain satisfactory prediction results because wind speed data comprise random nonlinear series. Current some statistical models are not proficient in predicting nonlinear time series, whereas artificial intelligence models often fall into local optima. For these reasons, a novel combined forecasting model, which combines hybrid models based on decomposed methods and optimization algorithms, is successfully developed with variable weighting combination theory for multi-step wind speed forecasting. In this model, three different hybrid models are proposed and to further improve the forecasting performance, a modified support vector regression is used to integrate all the results obtained by each hybrid model and obtain the final forecasting results. To verify the forecasting effectiveness of the proposed forecasting model, 10-min wind speed series from Penglai, China, are used as case studies. The experimental results indicate that the developed combined model not only outperforms other benchmark models but also can be satisfactorily used for planning for smart grids.

The strategy of investment in the stock market using modified support vector regression model

Stock indices forecasting has become a popular research issue in recent years. Although many statistical time series models have been applied in stock indices forecasting, they are limited to certain assumptions. Accordingly, the traditional statistical time series models might not be suitable for forecasting real-life stock indices data. Hence, this paper proposes a novel forecasting model to assist investors in determining a strategy for investments in the stock market. The proposed model is called the modified support vector regression model, which is composed of the correlation coefficient method, the sliding window algorithm and the support vector regression model. The results show that the forecasting accuracy of the proposed model is more stable than the existing models in terms of average and standard deviation of the root mean square error (RMSE) and the mean absolute percentage error (MAPE). Accordingly, the proposed model would be used to assist investors in determining a strategy for investing in stocks.

A modified support vector regression: Integrated selection of training subset and model

In recent years, support vector regression (SVR) has become an emerging and popular forecasting technique in the field of machine learning. However, it is subjected to the model selection and learning complexity O(K*N3), especially for a massive data set (N is the size of training dataset, and K is the number of search). How to simultaneously reduce K and N can give us insight and inspiration on designing an effective and accurate selection algorithm. To this end, this paper tries to integrate the selection of training subset and model for SVR, and proposes a nested particle swarm optimization (NPSO) by inheriting the model selection of the existing training subset based SVR (TS-SVR). This nested algorithm is achieved by adaptively and periodically estimating the search region of the optimal parameter setting for TS-SVR. Complex SVR, involving large-scale training data, can be seen as extensions of TS-SVRs, yielding a nested sequence of TS-SVRs with increasing sample size. The uniform design idea is transplanted to the above modeling process, and the convergence for the proposed model is proofed. By using two artificial regression problems, Boston housing and electric load in New South Wales as empirical data, the proposed approach is compared with the standard ones, the APSO-OTS-SVR, and other existing approaches. Empirical results show that the proposed approach not only can select proper training subset and parameter, but also has better generalization performance and fewer processing time.

Non-sparse ϵ-insensitive support vector regression for outlier detection

To estimate the approximate relationship between the dependent variable and its independent variables, it is necessary to diagnose outliers commonly present in numerous real applications before constructing the model. Nevertheless, the techniques of the standard support vector regression (-SVR) and modified support vector regression () achieved good performance for outliers' detection for nonlinear functions with high-dimensional inputs. However, they still suffer from the costs of time and the setting of parameters. In this study, we propose a practical method for detecting outliers, using non-sparse -SVR, which minimizes time cost and introduces fixed parameters. We apply this approach for real and simulation data sets to test its effectiveness.

ID-Score: A New Empirical Scoring Function Based on a Comprehensive Set of Descriptors Related to Protein–Ligand Interactions

Scoring functions have been widely used to assess protein-ligand binding affinity in structure-based drug discovery. However, currently commonly used scoring functions face some challenges including poor correlation between calculated scores and experimental binding affinities, target-dependent performance, and low sensitivity to analogues. In this account, we propose a new empirical scoring function termed ID-Score. ID-Score was established based on a comprehensive set of descriptors related to protein-ligand interactions; these descriptors cover nine categories: van der Waals interaction, hydrogen-bonding interaction, electrostatic interaction, π-system interaction, metal-ligand bonding interaction, desolvation effect, entropic loss effect, shape matching, and surface property matching. A total of 2278 complexes were used as the training set, and a modified support vector regression (SVR) algorithm was used to fit the experimental binding affinities. Evaluation results showed that ID-Score outperformed other selected commonly used scoring functions on a benchmark test set and showed considerable performance on a large independent test set. ID-Score also showed a consistent higher performance across different biological targets. Besides, it could correctly differentiate structurally similar ligands, indicating higher sensitivity to analogues. Collectively, the better performance of ID-Score enables it as a useful tool in assessing protein-ligand binding affinity in structure-based drug discovery as well as in lead optimization.

On-Line Prediction of Motor Shaft Misalignment Using Fast Fourier Transform Generated Spectra Data and Support Vector Regression

Shaft alignment prediction is essential for the development of effective coupling and rotating equipment maintenance systems. In this paper, we present a modified support vector regression (SVR) approach for shaft alignment predictions based on fast Fourier transform generated spectra data. The modified SVR approach uses data-dependent parameters in order to reduce computation time and achieve better predictions. The spectra data used is characterized by a large number of descriptors and very few data points. The strengths of SVR for shaft misalignment prediction include its ability to represent data in high-dimensional space through kernel functions. We reduce the dimension of the data using a multivariate AIC criterion in order to guarantee that the selected spectra are response dependent. We compare the performance of SVR with two of the most popular techniques used in condition monitoring, partial least squares, and principal components regression. Our results show that we can improve the performance of shaft misalignment prediction using SVR and the approach compares very favorably with partial least squares and principal components regression approaches. Also, we present a quantitative measure, shaft misalignment monitoring index, which can be used to facilitate easy identification of the alignment condition and as input to maintenance systems design.

Support Vector Machines for 3D Shape Processing

We propose statistical learning methods for approximating implicit surfaces and computing dense 3D deformation fields. Our approach is based on Support Vector (SV) Machines, which are state of the art in machine learning. It is straightforward to implement and computationally competitive; its parameters can be automatically set using standard machine learning methods. The surface approximation is based on a modified Support Vector regression. We present applications to 3D head reconstruction, including automatic removal of outliers and hole filling. In a second step, we build on our SV representation to compute dense 3D deformation fields between two objects. The fields are computed using a generalized SV Machine enforcing correspondence between the previously learned implicit SV object representations, as well as correspondences between feature points if such points are available. We apply the method to the morphing of 3D heads and other objects.

A Nonlinear Soft Sensor Based on Modified SVR for Quality Estimation in Polymerization

In this paper, a modified Support Vector Regression (SVR) have been proposed as an empirical model for developing a soft sensor to estimate production quality of polymerization process which is highly nonlinear and that which has high dimension. This method is derived through the modification of the risk function of standard Support Vector Machine using the concept of Locally Weighted Regression. Case studies have shown that the proposed method shows a better performance over standard SVR that had been shown to be superior to traditional statistical learning machine in case of high dimension, sparse and nonlinear data.