Accuracy Of Forecasting Models Research Articles

Deep learning forecasting tool facilitating the participation of photovoltaic systems into day-ahead and intra-day electricity markets

The participation of photovoltaic (PV) power into the energy market requires the development of accurate forecasting models to alleviate the imbalance penalties and ensure the viability of the investments. Several forecasting models have been proposed, focusing on different forecasting horizons. However, their implementation does not consider the energy market rules. Here we present a Day-Ahead Market (DAM)/Intra-Day Market (IDM) forecasting tool which comprises of four individual Deep Learning (DL) forecasters to facilitate the participation of PV power into the DAM and the three auctions of IDM. Different combinations of historical and predicted data of cloud cover and temperature are examined on Convolutional Neural Network and Transformer models for the forecasters’ development. The proposed tool is tested on five PV systems, while the aggregated production is also examined considering their spatial characteristics. The results indicate that the proposed forecasting tool could decrease the Mean Absolute Error up to 31.96% and 32.45% for DAM and IDM respectively.

Improving of the Generation Accuracy Forecasting of Photovoltaic Plants Based on <i>k</i>-Means and <i>k</i>-Nearest Neighbors Algorithms

Renewable energy sources (RES) are seen as a means of the fuel and energy complex carbon footprint reduction but the stochastic nature of generation complicates RES integration with electric power systems. Therefore, it is necessary to develop and improve methods for forecasting of the power plants generation using the energy of the sun, wind and water flows. One of the ways to improve the accuracy of forecast models is a deep analysis of meteorological conditions as the main factor affecting the power generation. In this paper, a method for adapting of forecast models to the meteorological conditions of photovoltaic stations operation based on machine learning algorithms was proposed and studied. In this case, unsupervised learning is first performed using the k-means method to form clusters. For this, it is also proposed to use studied the feature space dimensionality reduction algorithm to visualize and estimate the clustering accuracy. Then, for each cluster, its own machine learning model was trained for generation forecasting and the k-nearest neighbours algorithm was built to attribute the current conditions at the model operation stage to one of the formed clusters. The study was conducted on hourly meteorological data for the period from 1985 to 2021. A feature of the approach is the clustering of weather conditions on hourly rather than daily intervals. As a result, the mean absolute percentage error of forecasting is reduced significantly, depending on the prediction model used. For the best case, the error in forecasting of a photovoltaic plant generation an hour ahead was 9 %.

An investigation of complex fuzzy sets for large-scale learning

Complex fuzzy sets are an extension of type-1 fuzzy sets with complex-valued membership functions. Over the last 20 years, time-series forecasting has emerged as the most important application of complex fuzzy sets, with neuro-fuzzy systems employing them shown to be accurate and compact forecasting models. In the complex fuzzy sets literature, two dominant approaches to designing forecasters can be observed: sinusoidal membership functions versus complex-valued Gaussian membership functions. To date, however, there has never been a systematic investigation that compares the performance of these two membership types (or their combination) within a common architecture.We propose a new neuro-fuzzy architecture using complex fuzzy sets that has been designed for large-scale learning problems. This architecture employs randomized learning to speed up network training. In designing this architecture, we empirically compared sinusoidal complex fuzzy sets and complex Gaussian fuzzy sets. Across multiple variations of the architecture, we find that the complex Gaussian fuzzy sets lead to significantly more accurate forecasts on moderate-to-large time series datasets, while still keeping the overall size of the network compact.

Implementation and application of ensemble optimal interpolation on an operational chemistry weather model for improving PM2.5 and visibility predictions

Abstract. Data assimilation techniques are one of the most important ways to reduce the uncertainty in atmospheric chemistry model input and improve the model forecast accuracy. In this paper, an ensemble optimal interpolation assimilation (EnOI) system for a regional online chemical weather numerical forecasting system (GRAPES_Meso5.1/CUACE) is developed for operational use and efficient updating of the initial fields of chemical components. A heavy haze episode in eastern China was selected, and the key factors affecting EnOI, such as localization length scale, ensemble size, and assimilation moment, were calibrated by sensitivity experiments. The impacts of assimilating ground-based PM2.5 observations on the model chemical initial field PM2.5 and visibility forecasts were investigated. The results show that assimilation of PM2.5 reduces the uncertainty in the initial PM2.5 field considerably. Using only 50 % of observations in the assimilation, the root mean square error (RMSE) of initial PM2.5 for independent verification sites in mainland China decreases from 73.7 to 46.4 µg m−3, and the correlation coefficient increases from 0.58 to 0.84. An even larger improvement appears in northern China. For the forecast fields, assimilation of PM2.5 improves PM2.5 and visibility forecasts throughout the time window of 24 h. The PM2.5 RMSE can be reduced by 10 %–21 % within 24 h, and the assimilation effect is the most remarkable in the first 12 h. Within the same assimilation time, the assimilation efficiency varies with the discrepancy between model forecasts and observations at the moment of assimilation, and the larger the deviation, the higher the efficiency. The assimilation of PM2.5 further contributes to the improvement of the visibility forecast. When the PM2.5 increment is negative, it corresponds to an increase in visibility, and when the PM2.5 analysis increment is positive, visibility decreases. It is worth noting that the improvement of visibility forecasting by assimilating PM2.5 is more obvious in the light-pollution period than in the heavy-pollution period. The results of this study show that EnOI may provide a practical and cost-effective alternative to the ensemble Kalman filter (EnKF) for the applications where computational cost is the main limiting factor, especially for real-time operational forecast.

Study of foreign experience in dealing with ice and frost deposits on the wires of overhead power lines

The means and technologies for predicting and preventing frost and ice deposits on the wires of overhead power lines in the power grids of foreign countries are presented. The focus is on such systems as Meteo, dynamic thermal rating, online monitoring of China Southern Power Grid, Variable resistance cable de-icing system and mobile ice melting units. The advantages and disadvantages of these systems are analyzed and it is found that such approaches require energy consumption and are relevant only for the relevant region of the power grid, depending on weather conditions and the distance of electricity transmission. The mechanical, electromechanical, electrothermal, and physicochemical methods of predicting frost and ice deposits on the wires of overhead power lines are considered. It was found that the consideration of dynamic processes is essential and effective in the study of power grids in extreme weather conditions, and the data on frost and ice deposits on overhead line wires are characterized by high dimensionality, nonlinearity, multimodality, and heterogeneity, which makes it impossible to create an accurate forecasting model using traditional methods of evaluation and decision-making. The method of time series analysis and the method of Ensemble Empirical Mode Decomposition are proposed to solve this problem. The basis of these methods is to maximize the use of inherent regularities represented by frequency and time characteristics for effective data analysis and to create the basis for subsequent models and improve their forecasting accuracy. On the basis of the research, the author proposes ways to adapt and integrate the experience of foreign countries into the power grids of Ukraine. The necessity of creating integrated information systems for monitoring meteorological parameters and operating modes of power grids using specialized forecasting models and IT systems is substantiated, which will automate the process of assessing the current state of overhead power lines based on meteorological data in order to timely prevent emergencies in power grids provoked by frost and ice deposits.

Stochastic optimal scheduling strategy for a campus-isolated microgrid energy management system considering dependencies

Isolated microgrids have been widely used on campuses, becoming an important part of their power-supply infrastructure. In this study, a stochastic optimal scheduling strategy that considered dependencies was proposed for the energy management system (EMS) of an isolated microgrid on campus, which consists of three main components: an accurate forecasting model on both the demand and supply sides, an excellent core scheduling strategy, and an effective uncertainty analysis. First, the forecasting model involved data correction, feature selection, core forecasting, and error analyses to obtain consistently good forecasting results. A multilayer hybrid forecasting model was developed, including a stacked multilevel denoising autoencoder, mutual information, bidirectional long short-term memory network (LSTM), and LSTM, corresponding to the above four model elements. Based on the forecasting results, improved episodic deep reinforcement learning was used to build a stochastic optimal scheduling strategy for the EMS, which could realize rapid propagation of reward values and achieve stable updating of episodic memory, effectively improving sample efficiency. The objective of optimizing the scheduling was to obtain the lowest operational cost. Finally, the uncertainties in the forecasting process were analyzed using the Latin hypercube sampling method, and a regular vine copula was used to determine the dependence of the uncertainty analysis between buildings. To the best of our knowledge, this is the first study to consider dependence in the uncertainty analysis of a campus microgrid. The performance of the proposed method was evaluated by comparing it with other existing models, aiming to obtain more efficient and stable scheduling results.

A Model Integrating Satellite‐Derived Shoreline Observations for Predicting Fine‐Scale Shoreline Response to Waves and Sea‐Level Rise Across Large Coastal Regions

AbstractSatellite‐derived shoreline observations combined with dynamic shoreline models enable fine‐scale predictions of coastal change across large spatiotemporal scales. Here, we present a satellite‐data‐assimilated, “littoral‐cell”‐based, ensemble Kalman‐filter shoreline model to predict coastal change and uncertainty due to waves, sea‐level rise (SLR), and other natural and anthropogenic processes. We apply the developed ensemble model to the entire California coastline (approximately 1,760 km), much of which is sparsely monitored with traditional survey methods (e.g., Lidar/GPS). Water‐level‐corrected, satellite‐derived shoreline observations (obtained from the CoastSat toolbox) offer a nearly unbiased representation of in situ surveyed shorelines (e.g., mean sea‐level elevation contours) at Ocean Beach, San Francisco. We demonstrate that model calibration with satellite observations during a 20‐year hindcast period (1995–2015) provides nearly equivalent model forecast accuracy during a validation period (2015–2020) compared to model calibration with monthly in situ observations at Ocean Beach. When comparing model‐predicted shoreline positions to satellite‐derived observations, the model achieves an accuracy of <10 m RMSE for nearly half of the entire California coastline for the validation period. The calibrated/validated model is then applied for multi‐decadal simulations of shoreline change due to projected wave and sea‐level conditions, while holding the model parameters fixed. By 2100, the model estimates that 24%–75% of California's beaches may become completely eroded due to SLR scenarios of 1.0–3.0 m, respectively. The satellite‐data‐assimilated modeling system presented here is generally applicable to a variety of coastal settings around the world owing to the global coverage of satellite imagery.

Forecasting hotel room prices when entering turbulent times: a game-theoretic artificial neural network model

PurposeThis study aims to propose a risk-induced game theoretic forecasting model to predict average daily rate (ADR) under COVID-19, using an advanced recurrent neural network.Design/methodology/approachUsing three data sets from upper-midscale hotels in three locations (i.e. urban, interstate and suburb), from January 1, 2018, to August 31, 2020, three long-term, short-term memory (LSTM) models were evaluated against five traditional forecasting models.FindingsThe models proposed in this study outperform traditional methods, such that the simplest LSTM model is more accurate than most of the benchmark models in two of the three tested hotels. In particular, the results show that traditional methods are inefficient in hotels with rapid fluctuations of demand and ADR, as observed during the pandemic. In contrast, LSTM models perform more accurately for these hotels.Research limitations/implicationsThis study is limited by its use of American data and data from midscale hotels as well as only predicting ADR.Practical implicationsThis study produced a reliable, accurate forecasting model considering risk and competitor behavior.Theoretical implicationsThis paper extends the application of game theory principles to ADR forecasting and combines it with the concept of risk for forecasting during uncertain times.Originality/valueThis study is the first study, to the best of the authors’ knowledge, to use actual hotel data from the COVID-19 pandemic to determine an appropriate neural network forecasting method for times of uncertainty. The application of Shapley value and operational risk obtained a game-theoretic property-level model, which fits best.

Forecasting global solar radiation using a robust regularization approach with mixture kernels

AbstractAccurately forecasting global solar radiation plays a key role in photovoltaic evaluations. To quantify and control the uncertainties in global solar radiation forecasting, this study developed a robust and accurate forecasting model. This was constructed in the reproducing kernel Hilbert space with a novel regularization. Global solar radiation datasets were collected from the autonomous region of Tibet in China. Experimental results demonstrate that the proposed model can quantify uncertainties and obtain more accurate forecasting compared with machine learning models.

Predictive evaluation of solar energy variables for a large-scale solar power plant based on triple deep learning forecast models

The advanced development of large-scale solar power plants (LSSPs) has made it necessary to improve accurate forecasting models for the output of solar energy. Solar energy is still hampered by the lack of predictability in its output, which remains a major hurdle in the solar industry. This paper focuses on triple deep learning (DL) techniques such as Artificial Neural Network (ANN), Recurrent Neural Network (RNN) and Convolutional Neural Network- Long-Short Term Memory CNN-LSTM to address this problem. These techniques are utilized in solar energy variables (SEVs) such as power generation (MWh), soiling loss (%), and performance ratio (PR %) to determine the optimal forecast model. The novelty of this research is that it is the first time that important solar system parameters such as PR and soiling loss have been studied to predict a feasible forecast model using a different DL scheme. The SEVs real-time dataset is procured from the largest solar plant in Pakistan, titled “Quaid-e-Azam Solar Park” (QASP). The main significance of the study is that ANN, RNN, and CNN-LSTM-based models were developed in the DL process through feature generation, data scaling, training, and testing steps to predict the optimal model. The prediction values were compared with the solar plant's actual values over the last 7 years, and then a comparison was made to predict the future forecast trend over the next 20 years. The aim and goal is to develop three models to investigate the accurate results of time-series forecasting on the SEVs dataset, as well as to evaluate the performance measure errors to determine the appropriate model. Based on the forecasting/prediction graphic and error results, it was demonstrated that the CNN-LSTM hybrid model is a more capable forecasting model for the output of power generation and PR values and a more accurate predictor of the future trend over the ANN and RNN models. However, the ANN model is slightly better performed in predicting of soling loss value than CNN-SLTM and RNN. Thus, the CNN-LSTM hybrid model is an optimal model for SEVs, which can guarantee a variety of LSSPs of similar nature in the following time-series forecasting investigations which shows key findings and its importance for the solar industrial forecast issue.

Integrative forecasting and analysis of stock price using neural network and ARIMA model

The volatilities of stock prices have a crucial effect on financial decision-making worldwide. With a reliable and accurate forecast model, investors could gain insights into stock price fluctuations and market trends, thus maximizing the opportunity to make profits. In this work, two models were proposed for stock price forecasting. A neural network based on exploiting the abilities of convolutional neural network and bi-directional long short-term memory is proposed and implemented for forecasting the Nasdaq-100 daily closing price. For long-term stock price forecast, we proposed a hybrid model that combined the autoregressive integrated moving average procedure and a neural network layer for modeling the linear and nonlinear features of the Nasdaq Composite monthly closing price. The proposed models produced promising experiment results, indicating the models capability of making practical forecasts and analyses based on different data scales and volumes. This work also proposed further considerations of indicators related to the stock price in financial time-series forecasting.

Multivariate time series short term forecasting using cumulative data of coronavirus.

Coronavirus emerged as a highly contagious, pathogenic virus that severely affects the respiratory system of humans. The epidemic-related data is collected regularly, which machine learning algorithms can employ to comprehend and estimate valuable information. The analysis of the gathered data through time series approaches may assist in developing more accurate forecasting models and strategies to combat the disease. This paper focuses on short-term forecasting of cumulative reported incidences and mortality. Forecasting is conducted utilizing state-of-the-art mathematical and deep learning models for multivariate time series forecasting, including extended susceptible-exposed-infected-recovered (SEIR), long-short-term memory (LSTM), and vector autoregression (VAR). The SEIR model has been extended by integrating additional information such as hospitalization, mortality, vaccination, and quarantine incidences. Extensive experiments have been conducted to compare deep learning and mathematical models that enable us to estimate fatalities and incidences more precisely based on mortality in the eight most affected nations during the time of this research. The metrics like mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are employed to gauge the model's effectiveness. The deep learning model LSTM outperformed all others in terms of forecasting accuracy. Additionally, the study explores the impact of vaccination on reported epidemics and deaths worldwide. Furthermore, the detrimental effects of ambient temperature and relative humidity on pathogenic virus dissemination have been analyzed.

Electricity Demand Forecasting using LSTMs

Electricity demand forecasting is an essential task in the energy industry, enabling utilities and energy suppliers to optimize the generation, transmission, and distribution of electricity. In recent years, deep learning techniques such as Long Short -Term Memory (LSTM) neural networks have shown great potential in improving the accuracy and efficiency of time-series forecasting tasks, including electricity demand forecasting. This research proposes an LSTM-based neural network architecture for short-term electricity demand forecasting. The proposed model is evaluated on real-world electricity demand data, and the results demonstrate its effectiveness in predicting future demand patterns. The model's performance is evaluated using the Mean Squared Error loss function and the Root Mean Squared Error metric. The proposed model shows promising results compared to traditional time-series forecasting models. The results suggest that LSTM-based neural networks can be a powerful tool for electricity demand forecasting, providing more accurate and efficient forecasting models that can help improve energy system planning and decision making.

Forecasting Pendapatan UMKM: Strategi Tiga UMKM Unggulan di Kalurahan Hargorejo Dalam Kontribusi Terhadap Kalurahan Hargorejo, Kapanewon Kokap Kulon Progo Menggunakan Metode Arima

This research aims to forecast the income of MSMEs (Micro, Small and Medium Enterprises), namely Batik Ngudi Rejeki, Geblek Mbak Yuni, and Better Snack in Hargorejo Village, Kapanewon Kokap, Kulon Progo using the ARIMA (AutoRegressive Integrated Moving Average) method. MSME income data from 2019 to 2022 is used in this analysis. The ARIMA method is used to identify patterns and trends in revenue data, as well as predict future earnings. The results of the ARIMA analysis show forecasting the income of MSMEs Batik Ngudi Rejeki, Geblek Mbak Yuni, and Better Snack in the contribution of the Hargorejo, Kokap, Kulon Progo Villages for the 2023 period. The results of this forecasting can provide insight into the development of MSME income in the future and become a reference for business owners in more effective business planning, decision making, and strategy development. The data used is historical data on MSME income in a time series, where the process is analyzed using statistical software Eviews 12 to produce an accurate forecasting model.

An Intelligent IoT-Cloud-Based Air Pollution Forecasting Model Using Univariate Time-Series Analysis.

Air pollution is a significant environmental issue affecting public health and ecosystems worldwide, resulting from various sources such as industrial activities, vehicle emissions, and fossil fuel burning. Air pollution contributes to climate change and can cause several health problems, such as respiratory illnesses, cardiovascular disease, and cancer. A potential solution to this problem has been proposed by using different artificial intelligence (AI) and time-series models. These models are implemented in the cloud environment to forecast the Air Quality Index (AQI)utilizing Internet of things (IoT) devices. The recent influx of IoT-enabled time-series air pollution data poses challenges for traditional models. Various approaches have been explored to forecast AQI in the cloud environment using IoT devices. The primary objective of this study is to assess the efficacy of an IoT-Cloud-based model for forecasting the AQI under different meteorological conditions. To achieve this, we proposed a novel BO-HyTS approach that combines seasonal autoregressive integrated moving average (SARIMA) and long short-term memory (LSTM) and fine-tuned it byusing Bayesian optimization to predict air pollution levels. The proposed BO-HyTS model can capture both linear and nonlinear characteristics of the time-series data, thus augmenting the accuracy of the forecasting process. Additionally, several AQI forecasting models, including classical time-series, machine learning, and deep learning, are employed to forecast air quality from time-series data. Five statistical evaluation metrics are incorporated to evaluate the effectiveness ofmodels. While comparing the various algorithms among themselves becomes difficult, anon-parametric statistical significance test (Friedman test) is applied to assess the performance of the different machine learning, time-series, and deep learning models. The findings reveal that the proposed BO-HyTS model produced significantly better results than their competitor's, providing the most accurate and efficient forecasting model, with an MSE of 632.200, RMSE of 25.14, Med AE of 19.11, Max Error of 51.52, and MAE of 20.49. The results of this study provide insights into the future patterns of AQI in various Indian states and set a standard for these states as governments develop their healthcare policies accordingly. The proposed BO-HyTS model has the potential to inform policy decisions and enable governments and organizations to protect better and manage the environment beforehand.

An Overview of Pulses Production in India: Retrospect and Prospects of the Future Food with an Application of Hybrid Models.

Forecasts are valuable to countries to make informed business decisions and develop data-driven strategies. The production of pulses is an integral part of agricultural diversification initiatives because it offers promising economic opportunities to reduce rural poverty and unemployment in developing countries. Pulses are the cheapest source of protein needed for human health. India's pulses production guidelines must be based on accurate and best forecast models. Comparing classical statistical and machine learning models based on different scientific data series is the subject of high-level research today. This study focused on the forecasting behaviour of pulses production for India, Karnataka, Madhya Pradesh, Maharashtra, Rajasthan and Uttar Pradesh. The data series was split into a training dataset (1950-2014) and a testing dataset (2015-2019) for model building and validation purposes, respectively. ARIMA, NNAR and hybrid models were used and compared on training and validation datasets based on goodness of fit (RMSE, MAE and MASE). This research demonstrates that due to the diverse agricultural conditions across different provinces in India, there is no single model that can accurately predict pulse production in all regions. This study's highest accuracy model is ARIMA. ARIMA outperforms NNAR, a machine learning model. Pulse production in India, Rajasthan, and Madhya Pradesh will expand by 26.11%, 12.62%, and 0.51% from 2020 to 2030, whereas it would decline by -6.5%, -6.21%, and -6.76 per cent in Karnataka, Maharashtra, and Uttar Pradesh, respectively. The current forecast results could allow policymakers to develop more aggressive food security and sustainability plans and better Indian pulses production policies in the future.

A comparison of machine learning methods for ozone pollution prediction

Precise and efficient ozone (hbox {O}_{3}) concentration prediction is crucial for weather monitoring and environmental policymaking due to the harmful effects of high hbox {O}_{3} pollution levels on human health and ecosystems. However, the complexity of hbox {O}_{3} formation mechanisms in the troposphere presents a significant challenge in modeling hbox {O}_{3} accurately and quickly, especially in the absence of a process model. Data-driven machine-learning techniques have demonstrated promising performance in modeling air pollution, mainly when a process model is unavailable. This study evaluates the predictive performance of nineteen machine learning models for ozone pollution prediction. Specifically, we assess how incorporating features using Random Forest affects hbox {O}_{3} concentration prediction and investigate using time-lagged measurements to improve prediction accuracy. Air pollution and meteorological data collected at King Abdullah University of Science and Technology are used. Results show that dynamic models using time-lagged data outperform static and reduced machine learning models. Incorporating time-lagged data improves the accuracy of machine learning models by 300% and 200%, respectively, compared to static and reduced models, under RMSE metrics. And importantly, the best dynamic model with time-lagged information only requires 0.01 s, indicating its practical use. The Diebold-Mariano Test, a statistical test used to compare the forecasting accuracy of models, is also conducted.

Conflict forecasting using remote sensing data: An application to the Syrian civil war

Conflict research is increasingly influenced by modern computational and statistical techniques. Combined with recent advances in the collection and public availability of new data sources, this allows for more accurate forecasting models in ever more fine-grained spatial areas. This paper demonstrates the utilization of remote sensing data as a potential solution to the lack of official data sources for conflict forecasting in crisis-ridden countries. We evaluate and quantify remote sensing data’s differentiated impact on forecasting accuracy across fine-grained spatial grid cells using the Syrian civil war as a use case. It can be shown that conflict, particularly its onset, can be forecasted more accurately by employing publicly available remote sensing datasets. These results are consistent across a range of established statistical and machine learning models, which raises the hope to get closer to reliable early-warning systems for conflict prediction.

Forecasting CPI of restaurants and hotels in Korea using the seasonal autoregressive integrated moving average (SARIMA) model

Economic development is influenced by many factors, one of which is the inflation rate. The consumer price index (CPI) is a commonly used indicator for measuring the inflation rate. In recent years, CPI prediction has attracted the attention of many scholars due to its excellent measure of macroeconomic performance. It is very essential to develop an accurate and precise forecasting model. This research fitted and forecasted the CPI of restaurants and hotels in South Korea using the seasonal autoregressive integrated moving average (SARIMA) model. In contrast to other time series models, the SARIMA model incorporates the seasonal component of a time series to improve the accuracy of its forecasts. The time series data were obtained from the website of Statistics Korea (KOSTAT) for the monthly CPI of restaurants and hotels in South Korea from January 2010 to December 2022. Data were analyzed using R-Statistics software and EViews. In this paper, the best model was based on the results of mean absolute percentage error (MAPE) and R2. A detailed description of model selection and predictive accuracy is provided. The findings suggest that the proposed research approach achieves good prediction accuracy from a range of different SARIMA models. Therefore, the developed SARIMA model can be considered for forecasting of monthly CPI of restaurants and hotels in South Korea.

A novel SGD-DLSTM-based efficient model for solar power generation forecasting system

Abstract Globally, Solar Power (SP) is generated by employing Photovoltaic (PV) systems. Accurate forecasting of PV power is a critical issue in ensuring secure operation along with economic incorporation of PV in smart grids. For providing an accurate forecasting model, various prevailing methodologies have been developed even then, there requires a huge enhancement. Thus, for Solar Power Generation (SPG) forecasting with deviation analysis, a novel Strengthen Gaussian Distribution-centric Deep Long Short Term Memory (SGD-DLSTM) methodology has been proposed here. Firstly, the PV modelling is formulated. After that, as of the PV, the data is gathered; likewise, for the deviation analysis, the historical data is gathered. Next, the pre-processing is performed; this stage undergoes two steps namely the Missing Value (MV) imputation and the scaling process. Afterwards, the features pertinent to the weather condition along with SP are extracted. After that, by utilizing the Intensive Exploitation-centric Shell Game Optimizer (IESGO) algorithm, the significant features are selected as of the features extracted. Then, the SPG is predicted by inputting the selected features into the SGD-DLSTM classifier. Next, by computing the Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE) measures, the predicted outcome’s deviation is assessed. In the experimental evaluation, by means of these measures, the proposed system’s performance is contrasted with the conventional techniques. Therefore, from the experimental assessment, it was established that the proposed model exhibits better performance than the prevailing research works. When analogized to the prevailing methodologies, a better accuracy of 97.25% was attained by the proposed system.