Photovoltaic (PV) power prediction has a constantly evolving solutions landscape with a myriad of data-driven techniques. Each technique leverages a self-adaptive algorithm that must retrain in intervals, be it each day, week, or season, to avoid the model generalizing poorly because of overfitting, underfitting, or concept drift. This paper aims to improve the generalization capability of PV power predictors such as autoencoders used widely in the industry by introducing feature-enhanced ensemble learning (FEEL) after the feature selection step. This framework uses a combination of nonparametric regression and generalized additive models, and an ensemble of weak regularized multilayer perceptron models. Once trained, the framework can reliably generalize on test data across long time periods without any significant degradation in performance. The proposed framework was validated against the baseline autoencoder-based feature enhancement model on a real PV system from a smart neighborhood in Alabama for September 2019. The FEEL framework performed three times better than the baseline, but when applied to the baseline, its performance improved by two times on average. Furthermore, the framework generalized consistently better than five other feature enhancement strategies. Despite fluctuations in weather, the FEEL framework’s R-square score had a range of 8.1%, whereas that of the baseline was 48.3%. The mutual information and Minkowski distance scores attempted to quantify concept and model drift, respectively. These scores show that the FEEL framework generalized the ensemble learning models at least two times better than the baseline across the different test days. These results form the first step toward decentralized intelligence for smart grid applications that could free up resources for other expensive analytics in the field.

/pdf/regression-and-generalized-additive-model-to-enhance-the-12uys0mntu.pdf

Regression and Generalized Additive Model to Enhance the Performance of Photovoltaic Power Ensemble Predictors

Probabilistic forecasting of PV generation is crucial in uncertainty management to reinforce PV‐integrated power systems for long‐term planning. In this context, developing a reliable probabilistic forecast model is challenging due to weather conditions' stochastic nature and varying daily PV production patterns at multiple time instants. Due to varying probability distribution patterns, a nonparametric approach, such as quantile regression, is challenging to approximate the forecast error distribution. A forecast combination concept that yields impressive probabilistic forecasting results has rarely been utilized for PV generation forecasting. The use of complementary and sensible point forecasters characterizing different aspects of intricate PV production patterns in a k‐nearest neighbor quantile regression averaging framework is the proposed probabilistic forecasting model. The proposed forecasting model's accuracy is assessed with real‐world PV generation data from the USA for multi‐time instants. The probabilistic forecast performance of the proposed quantile k‐nearest neighbors regression averaging model is evaluated against the traditional quantile regression averaging, quantile regression forests, quantile k‐nearest neighbors, basic quantile regression, and regression bootstrapping in terms of the forecast (quantiles and prediction intervals) accuracy, using relevant error measures and scores. The proposed model has superior performance compared to the other models, as indicated by the lowest forecast errors and scores.

A k‐nearest neighbor‐based averaging model for probabilistic PV generation forecasting

A short-term probabilistic photovoltaic power prediction method based on feature selection and improved LSTM neural network

A smart grid is a modern electricity system enabling a bidirectional flow of communication that works on the notion of demand response. The stability prediction of the smart grid becomes necessary to make it more reliable and improve the efficiency and consistency of the electrical supply. Due to sensor or system failures, missing input data can often occur. It is worth noting that there has been no work conducted to predict the missing input variables in the past. Thus, this paper aims to develop an enhanced forecasting model to predict smart grid stability using neural networks to handle the missing data. Four case studies with missing input data are conducted. The missing data is predicted for each case, and then a model is prepared to predict the stability. The Levenberg–Marquardt algorithm is used to train all the models and the transfer functions used are tansig and purelin in the hidden and output layers, respectively. The model’s performance is evaluated on a four-node star network and is measured in terms of the MSE and R2 values. The four stability prediction models demonstrate good performances and depict the best training and prediction ability.

/pdf/smart-grid-stability-prediction-model-using-neural-networks-3p5pazyf.pdf

Smart Grid Stability Prediction Model Using Neural Networks to Handle Missing Inputs

Chaos plays a prominent role in nonlinear systems like energy, finance, and weather. In these systems, a changing parameter to time yields a chaotic time series that contains a lot of system information. This information helps to analyze and understand the behavior of the chaotic system. However, traditional methods of investigating these chaotic time series have low prediction accuracy. Recently, various researchers have got interested in applying machine learning models for these time series. Accordingly, this study has developed two machine learning models using LSTM and NAR neural networks to forecast monthly smoothed sunspot number time series. Further, Adam and Levenberg-Marquardt algorithms are utilized for LSTM and NAR neural networks during the training process. Besides, mean square error and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{R}^{2}$</tex> are used to assess the proposed models' performance. The simulation results demonstrated that the developed prediction models have successfully predicted the sunspot number. 

Time Series Forecasting Model for Sunspot Number

This paper focuses on developing a prediction model for chaotic behavior in fractional-order Duffing's oscillator using neural networks. The model predicts the change in state variables' values of the oscillator using its past observations obtained by numerically solving the governing equations using the famous Grunwald-Letnikov's approach. Further, a comparison of hold-out and k-fold techniques is made using the Levenberg-Marquardt training algorithm. The results show the best-proposed model's prediction performance with mean square errors (MSE) and R2 values close to zero and one, respectively. In all the cases, the k-fold cross-validation has performed better than hold-out. However, the k-fold method has taken more computational time for training the model as it is trained k-times compared to one time using the hold-out method.

Chaotic Time Series Prediction Model for Fractional-Order Duffing's Oscillator

The demand for renewable energy to sustain today’s vulnerability towards depleting fossil fuels is a crucial agenda for research. Various inverter topologies have been proposed to convert renewable sources into a usable form. But output THD, additional filtering components at line frequency (leading to bulky circuitry), lower efficiency, etc., are some of the limitations faced in all those topologies. This paper aims to change a voltage source inverter’s traditional behavior, which generates lesser output voltage with higher THD. The paper proposes a closed-loop non-ideal differential boost inverter (DBI) employing a PI controller. The optimization techniques such as, genetic algorithm (GA) and bacterial foraging optimization algorithm (BFOA) are incorporated to accentuate the PI controller’s performance to produce a better response during line and load disturbance conditions with reduced THD. DBI performance is evaluated on a laboratory prototype with different loading conditions. A comparison between the algorithms and the previous topologies from the literature survey has also been provided to validate this research’s claims. This paper’s required simulation study is carried out using MATLAB, and real-time validation is carried out using dSPACE 1104 with sampling time of one $\mu \text{s}$ .

/pdf/implementation-of-optimization-based-pi-controller-tuning-tmhajm1709.pdf

Implementation of Optimization-Based PI Controller Tuning for Non-Ideal Differential Boost Inverter

Probabilistic PV generation forecasts are necessary for the uncertainty management in the long-term planning of power systems with PV integrations. The weather-dependent PV generation makes it a challenging task necessitating a nonparametric approach, such as quantile regression, for obtaining probabilistic forecasts. Here, a quantile regression averaging approach is used to combine the selective point forecasts of autoregressive conditional heteroscedastic model, random forests model, and multiple linear regression model to further enhance the forecast accuracy. The selection of sensible regressors with physical relevance is necessary for the proposed framework. Hence, such theoretically formulated regressors capable of modeling real-world PV generation data collected from the USA are utilized for assessing the efficacy of the proposed quantile regression averaging model. The reliabilities of the prediction intervals of the proposed model is compared with the popular quantile regression forests, the quantile k-nearest neighbors, and the basic quantile regression approaches via widely used performance indices.

Reliable Prediction Intervals of PV Generation Using Quantile Regression Averaging Approach

Probabilistic PV generation forecasting plays a significant role in the uncertainty management of power systems with higher penetration of PV generation PV generation forecasting is more challenging due to the stochastic nature of weather conditions Various outmoded probability models have been espoused for PV generation uncertainty; the most popular ones rely on specific parametric density functions to fit forecasting error However, PV generation uncertainty has varying probability distribution patterns, and a parametric distribution for forecast error may not always be applicable at different time instants and places Non-parametric approaches, eg, quantile regression, on the other hand, estimate the predictive densities directly from the data without any constraints on the distribution shape On this note, the benefit of the association of a few potential and sensible regressors set with the intricate PV generation pattern is envisioned for effective probabilistic forecasting The regressors for the proposed quantile regression model are chosen based on the physics of the underlying phenomenon The effectiveness of the proposed probabilistic forecasting is tested using real-world multi-time instant PV generation data collected from the USA Out-of-sample quantile forecasts are generated for the PV generation, which is found to be accurate with a minimum deviation of estimated quantiles from the theoretical quantiles Probability densities are found from these estimated quantiles, and their goodness-of-fit is tested using the famous KS test

Probabilistic Forecasting of Daily PV Generation Using Quantile Regression Method

https://sciendo.com/pdf/10.34768/amcs-2021-0026

B Rajanarayan Prusty

Papers

Chaotic Time Series Prediction Model for Fractional-Order Duffing's Oscillator

Implementation of Optimization-Based PI Controller Tuning for Non-Ideal Differential Boost Inverter

Reliable Prediction Intervals of PV Generation Using Quantile Regression Averaging Approach

Probabilistic Forecasting of Daily PV Generation Using Quantile Regression Method

Forecasting Models for Chaotic Fractional–Order Oscillators Using Neural Networks