Reduce Prediction Errors Research Articles

Research on Multi-Scale Signal Processing and Efficient Model Construction Strategies in Lithium-Ion Battery State Prediction

Abstract To tackle the question of limited generalization and inefficiency in predicting state of health (SOH) and state of charge (SOC) in lithium-ion batteries across diverse sequence lengths, a novel hybrid model is developed. This model integrates multivariate variational mode decomposition (MVMD), informer, and long short-term Memory (LSTM) networks. Initially, battery health features are extracted from the charge and discharge curves, which are then validated for their relevance to SOH and SOC via correlation analysis and random forest algorithms. These features undergo multi-scale decomposition using MVMD, thereby encapsulating the intricate dynamics of battery state changes across various time scales. This decomposition enhances the model's adaptability to different sequence lengths, bolstering its generalization capability. Subsequently, the informer model is utilized to identify temporal patterns within the decomposed features. Finally, LSTM exploits its capacity to capture temporal dependencies for further refinement of the predictions. This hybrid strategy yields substantial enhancements in both efficiency and accuracy. Compared to the transformer model, the proposed hybrid model demonstrates a 30% reduction in SOH prediction error and a 22% decrease in SOC prediction error, concurrently slashing training time significantly. Spanning diverse sequence lengths and battery types, demonstrates the model's strong generalization capabilities.

Early Quality Prediction of Complex Double-Walled Hollow Turbine Blades Based on Improved Whale Optimization Algorithm

Abstract The precision in forming complex double-walled hollow turbine blades significantly influences their cooling efficiency, making the selection of appropriate casting process parameters critical for achieving fine-casting blade formation. However, the high cost associated with real blade casting necessitates strategies to enhance product formation rates and mitigate cost losses stemming from the overshoot phenomenon. We propose a machine learning (ML) data-driven framework leveraging an enhanced whale optimization algorithm (WOA) to estimate product formation under diverse process conditions to address this challenge. Complex double-walled hollow turbine blades serve as a representative case within our proposed framework. We constructed a database using simulation data, employed feature engineering to identify crucial features and streamline inputs, and utilized a whale optimization algorithm-back-propagation neural network (WOA-BP) as the foundational ML model. To enhance WOA-BP’s performance, we introduce an optimization algorithm, the improved chaos whale optimization-back-propagation (ICWOA-BP), incorporating cubic chaotic mapping adaptation. Experimental evaluation of ICWOA-BP demonstrated an average mean absolute error of 0.001995 mm, reflecting a 36.21% reduction in prediction error compared to conventional models, as well as two well-known optimization algorithms (particle swarm optimization (PSO), quantum-based avian navigation optimizer algorithm (QANA)). Consequently, ICWOA-BP emerges as an effective tool for early prediction of dimensional quality in complex double-walled hollow turbine blades.

Smart Grids: Integrating AI for Efficient Renewable Energy Utilization

The urgent global shift from fossil fuels to renewable energy sources necessitates innovative solutions to address energy system management challenges. Smart grids, equipped with sophisticated infrastructures, play a crucial role in this transition. This study integrates Artificial Intelligence (AI) into smart grids to enhance their efficiency and reliability, directly supporting the United Nations Sustainable Development Goals (SDGs), particularly SDG 7 (Affordable and Clean Energy), SDG 9 (Industry, Innovation, and Infrastructure), and SDG 11 (Sustainable Cities and Communities). Employing a mixed-methods approach, the research utilizes historical and real-time data, applying machine learning algorithms such as Linear Regression, Support Vector Regression (SVR), Recurrent Neural Networks (RNN), and Long ShortxTerm Memory (LSTM) for predictive accuracy in energy management. Optimization techniques like Genetic Algorithms and Particle Swarm Optimization (PSO) are also implemented for resource scheduling and grid balancing. The results demonstrate significant improvements, with an 11.76% increase in energy efficiency and grid stability, a 66.67% reduction in prediction errors, and a 20% decrease in operational costs compared to conventional systems. These enhancements highlight the transformative potential of AI in smart grids, promoting more efficient and sustainable energy utilization. The study concludes that AI-driven smart grids are pivotal in achieving the SDGs by providing scalable and efficient solutions for renewable energy integration, thereby fostering sustainable development and reducing environmental impacts.

Locally linearized physics-informed neural networks for Riemann problems of hyperbolic conservation laws

In this work, we demonstrate that physics-informed neural networks (PINNs) tend to propagate predicted shock wave information bidirectionally in time, which does not align with the actual evolution direction of solutions to hyperbolic conservation laws. This mismatch results in instability and hinders the reduction of the loss of governing equations, as well as the initial condition loss by the deep neural network. In order to tackle this problem, we simplify the complexity of the problem by constructing equivalent linear transport equations in the region of shock wave generation. The speeds of these linearized waves are governed by the Rankine–Hugoniot relations of conservation laws. This approach is termed the Locally Linearized PINNs method. Specifically, an appropriate shock wave detector is initially designed to identify domains where shock waves occur. Near shock waves, the original nonlinear equations are transformed into their linearized forms, thereby modifying the residual terms of the partial differential equations. Additionally, an equilibrium factor is introduced in fluid compression regions to reduce prediction errors and stabilize the training of deep neural networks. Numerical examples illustrate that Locally Linearized PINNs effectively address the challenge of predicting global solutions in PINNs and significantly improve shock-capturing performance for hyperbolic conservation laws.

Outlier Detection Using Gaussian Mixture Model Clustering to Optimize XGBoost for Credit Approval Prediction

Credit approval prediction is one of the critical challenges in the financial industry, where the accuracy and efficiency of credit decision-making can significantly affect business risk. This study proposes an outlier detection method using the Gaussian Mixture Model (GMM) combined with Extreme Gradient Boosting (XGBoost) to improve prediction accuracy. GMM is used to detect outliers with a probabilistic approach, allowing for finer-grained anomaly identification compared to distance- or density-based methods. Furthermore, the data cleaned through GMM is processed using XGBoost, a decision tree-based boosting algorithm that efficiently handles complex datasets. This study compares the performance of XGBoost with various outlier detection methods, such as LOF, CBLOF, DBSCAN, IF, and K-Means, as well as various other classification algorithms based on machine learning and deep learning. Experimental results show that the combination of GMM and XGBoost provides the best performance with an accuracy of 95.493%, a recall of 91.650%, and an AUC of 95.145%, outperforming other models in the context of credit approval prediction on an imbalanced dataset. The proposed method has been proven to reduce prediction errors and improve the model's reliability in detecting eligible credit applications.

A Deep Learning Approach for Enhanced Real-Time Prediction of Winter Road Surface Temperatures in High-Altitude Mountain Areas

Low temperatures and icing in winter are significant factors that severely affect highway safety and traffic mobility. To enhance the precision and reliability of real-time winter road surface temperature (RST) prediction, a short-term prediction model is developed that harnesses both feature selection and deep learning. Leveraging meteorological data from a mountain highway in Yunnan, China, the key environmental variables affecting road surface temperature were first extracted using a random forest (RF) model for feature selection. These features were then combined with RST data to construct multiple groups of input variable combinations for the prediction model. A short-term prediction model with a 10-minute update frequency was built using a long short-term memory neural network (LSTM), namely RF-LSTM. The best input variable combination and preset parameters for the prediction model were determined through comparative testing with prevalent machine learning models, and the transferability of the prediction model was verified. The results showed that the best input variable combination for the RF-LSTM prediction model was road surface temperature and air temperature. The model recognised that the short-term RST was affected by long and short-term memory characteristics within a two-hour timeframe. When compared to the RF model, backpropagation (BP) neural network model and the standard LSTM model, the proposed model reduces prediction errors by 59.15%, 31.10% and 20.26%, respectively, while the prediction accuracy is 99.13% within an error margin of ±0.5℃. On the verification dataset, the proposed model maintains its time transferability with an average prediction absolute error of 0.0478. In all, the proposed model not only achieves a higher level of precision in real-time RST predictions but also ensures a more consistent and reliable performance under the challenging conditions of high altitude and mountainous terrain, offering enhanced support for traffic safety and road maintenance decision-making.

Optimization study of diesel engine emission prediction based on neural network model cluster

This study aims to reduce the testing volume and cost of engine bench tests. By combining neural network models and ensemble learning algorithms, a model cluster prediction method is proposed to predict engine emissions. The core of this method is to use a large number of neural network models for prediction, taking the average of the prediction results as the final prediction result, thereby reducing prediction errors and improving accuracy. The findings reveal that this cluster-based approach significantly outperforms a single optimal model in forecasting steady-state emissions of NOx, HC, and CO in diesel engines. Notably, the performance of the model cluster is highly dependent on both the quality and the number of constituent sub-models, with enhanced predictive capabilities observed when higher-quality sub-models are included. Additionally, the study identifies a stabilization in predictive accuracy as the number of models increases, particularly within the range of 50 to 100 models, and achieving robust stability beyond 100 models. The research also underscores the importance of the training dataset size on the effectiveness of neural network modeling. A reduction in the training dataset from 28% to 18% leads to a decline in the coefficient of determination ( R2) for NOx emissions prediction in a single neural network model from 0.6954 to 0.6539, a 5.97% decrease. For the model cluster method, the R2 similarly drops from 0.8633 to 0.8154, a reduction of 5.55%. Notably, the neural network model suffers from distortion in predicting the peak position of NOx emissions. After supplementing specific operating condition training data in a targeted manner, the model training amount was increased from 18% to 25%. The model cluster method showed a good improvement in the prediction of NOx emissions, with the prediction coefficient increasing from 0.8154 to 0.8997, with an improvement rate of 10.34%. Through planning of training data, the study demonstrates that employing just 25% of experimental data for model clustering can effectively predict the engine emission trends for the remaining 75% of data in the target conditions. This approach offers a substantial reduction in experimental requirements while maintaining high prediction accuracy.

A deep learning approach to optimize remaining useful life prediction for Li-ion batteries

Accurately predicting the remaining useful life (RUL) of lithium-ion (Li-ion) batteries is vital for improving battery performance and safety in applications such as consumer electronics and electric vehicles. While the prediction of RUL for these batteries is a well-established field, the current research refines RUL prediction methodologies by leveraging deep learning techniques, advancing prediction accuracy. This study proposes AccuCell Prodigy, a deep learning model that integrates auto-encoders and long short-term memory (LSTM) layers to enhance RUL prediction accuracy and efficiency. The model’s name reflects its precision (“AccuCell”) and predictive strength (“Prodigy”). The proposed methodology involves preparing a dataset of battery operational features, split using an 80–20 ratio for training and testing. Leveraging 22 variations of current (critical parameter) across three Li-ion cells, AccuCell Prodigy significantly reduces prediction errors, achieving a mean square error of 0.1305%, mean absolute error of 2.484%, and root mean square error of 3.613%, with a high R-squared value of 0.9849. These results highlight its robustness and potential for advancing battery health management.

Diffusion-driven Incomplete Multimodal Learning for Air Quality Prediction

Predicting air quality using multimodal data is crucial to comprehensively capture the diverse factors influencing atmospheric conditions. Therefore, this study introduces a multimodal learning framework that integrates outdoor images with traditional ground-based observations to improve the accuracy and reliability of air quality predictions. However, aligning and fusing these heterogeneous data sources pose a formidable challenge, further exacerbated by pervasive data incompleteness issues in practice. In this paper, we propose a novel incomplete multimodal learning approach (iMMAir) to recovery missing data for robust air quality prediction. Specifically, we first design a shallow feature extractor to capture modal-specific features within the latent representation space. Then we develop a conditional diffusion-driven recovery module to mitigate the distribution gap between the recovered and true data. This module further incorporates two conditional constraints of temporal correlation and semantic consistency for effective modal completion. Finally, we reconstruct incomplete modalities and fuse available data using a multimodal transformer network to predict the air quality. To alleviate the modality imbalance problem, we employ an adaptive gradient modulation strategy to adjust the optimization of each modality. Experimental results demonstrate that iMMAir significantly reduces prediction errors, outperforming baseline models by an average of 5.6% and 2.5% in air quality regression and classification tasks. Our source code and data are available at https://github.com/pestasu/IMMAir.

Enhancing streamflow prediction in the Wujiang River basin: a two-stage decomposition approach with deep learning integration

ABSTRACT Global climate change and human activities have profoundly impacted the geological and hydrological processes in watersheds, increasing the challenges in streamflow prediction. In this study, we propose a streamflow prediction model based on deep learning and dual-mode decomposition specifically tailored for the Wujiang River basin, a significant tributary on the southern bank of the upper Yangtze River. This model effectively addresses the prediction error issue caused by high-frequency components through dual-mode decomposition of time-series data. The results demonstrate that employing the coupled complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN)–variational mode decomposition (VMD)–temporal convolutional network (TCN)–long short-term memory network (LSTM) model reduces prediction errors by at least 35% compared to single decomposition models. Furthermore, compared to individual TCN or LSTM models, the TCN–LSTM coupled model exhibits greater stability and higher prediction accuracy during training, with reductions in mean absolute error and root mean square error by 43.13 and 24.57%, respectively. This model holds promising prospects for application and can provide crucial insights for water resource management.

Machine Learning for Optimising Renewable Energy and Grid Efficiency

This research investigates the application of machine learning models to optimise renewable energy systems and contribute to achieving Net Zero emissions targets. The primary objective is to evaluate how machine learning can improve energy forecasting, grid management, and storage optimisation, thereby enhancing the reliability and efficiency of renewable energy sources. The methodology involved the application of various machine learning models, including Long Short-Term Memory (LSTM), Random Forest, Support Vector Machines (SVMs), and ARIMA, to predict energy generation and demand patterns. These models were evaluated using metrics such as Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). Key findings include a 15% improvement in grid efficiency after optimisation and a 10–20% increase in battery storage efficiency. Random Forest achieved the lowest MAE, reducing prediction error by approximately 8.5%. The study quantified CO2 emission reductions by energy source, with wind power accounting for a 15,000-ton annual reduction, followed by hydropower and solar reducing emissions by 10,000 and 7500 tons, respectively. The research concludes that machine learning can significantly enhance renewable energy system performance, with measurable reductions in errors and emissions. These improvements could help close the “ambition gap” by 20%, supporting global efforts to meet the 1.5 °C Paris Agreement targets.

PREDICTION INDONESIA COMPOSITE INDEX USING INTEGRATION DECOMPOSITION- NEURAL NETWORK ENSEMBLE DURING VUCA ERA

The Volatility, Uncertainty, Complexity, and Ambiguity (VUCA) era causes turmoil in the capital markets, stocks, commodities, etc. The impact is a decline in the Composite Stock Price Index (IHSG) in 2020. Therefore, future data is needed to inform investors and business people when making portfolio decisions. This paper develops a decomposition and Neural Network (NN) integration model to predict ICI during the VUCA era. The results are presented empirically to show the model's effectiveness in reducing prediction errors. First, the actual data is converted into three components; second, with the Neural Network Ensemble (NNE) approach where the initial step of decomposition results is trained using artificial NN with architecture, training data, and topology to produce individual networks; The output is selected using Principal Component Analysis (PCA) and becomes input to the ensemble model, then combined using a simple average and weighted average. The empirical results from ICI predictions illustrate: (1) decomposition has the potential to overcome data that is characterized by high volatility; (2) NNE is able to reduce errors (MSE≤0.100e-4, MAE≤0.01) compared to individual networks (MSE=0.0024 MAE=0.0376); (3) ensemble combinations using weighted averages (MSE≤3.00e-5,MAE≤0.002) are superior to simple averages (MSE≤5.00e-5,MAE≤0.01); (4) the integration carried out shows effectiveness in predicting ICI and provides better prediction results.

Probabilistic Tide Calculation for Multi-Wind Farm Distribution Network Based on CATTSTS Model

Traditional probabilistic tide calculation methods often fall short in capturing the dynamic nature of these systems, leading to discrepancies between calculated and actual data. To address this, a dynamic probabilistic tide calculation model is proposed in this study, incorporating a wind farm probabilistic model enhanced by the CATTSTS model. This integration aims to optimize prediction accuracy by improving feature focusing and data fitting, thereby reducing prediction errors and providing more reliable data for power system adjustments and optimizations. Experimental results demonstrate that the model's predictions for wind speed series in multi-wind farm distribution networks closely match actual conditions, underscoring its practical applicability and reliability in real-world settings. Furthermore, the study highlights the impact of spatial and temporal correlations on model performance. This research contributes to advancing the management of renewable energy integration in power systems, offering insights for more informed decision-making and operational efficiency.

Protecting Infinite Data Streams from Wearable Devices with Local Differential Privacy Techniques

The real-time data collected by wearable devices enables personalized health management and supports public health monitoring. However, sharing these data with third-party organizations introduces significant privacy risks. As a result, protecting and securely sharing wearable device data has become a critical concern. This paper proposes a local differential privacy-preserving algorithm designed for continuous data streams generated by wearable devices. Initially, the data stream is sampled at key points to avoid prematurely exhausting the privacy budget. Then, an adaptive allocation of the privacy budget at these points enhances privacy protection for sensitive data. Additionally, the optimized square wave (SW) mechanism introduces perturbations to the sampled points. Afterward, the Kalman filter algorithm is applied to maintain data flow patterns and reduce prediction errors. Experimental validation using two real datasets demonstrates that, under comparable conditions, this approach provides higher data availability than existing privacy protection methods for continuous data streams.

Short-Term Prediction of Origin–Destination Passenger Flow in Urban Rail Transit Systems with Multi-Source Data: A Deep Learning Method Fusing High-Dimensional Features

Short-term origin–destination (OD) passenger flow forecasting is crucial for urban rail transit enterprises aiming to optimise transportation products and increase operating income. As there are large-scale OD pairs in an urban rail transit system, OD passenger flow cannot be obtained in real time (temporal hysteresis). Additionally, the distribution characteristics are also complex. Previous studies mainly focus on passenger flow prediction at metro stations, while few methods solve the OD passenger flow prediction problems of an urban rail transit system. In view of this, we propose a novel deep learning method fusing high-dimensional features (HDF-DL) with multi-source data. The HDF-DL method is combined with three modules. The temporal module incorporates the time-varying, trend, and cyclic characteristics of OD passenger flow, while the latest OD passenger flow time sequence (within 1 h) is excluded from the time-varying characteristics. In the spatial module, the K-means and K-shape algorithms are used to classify OD pairs from multiple perspectives and capture the spatial features, reducing the difficulty of OD passenger flow predictions with large-scale and complex characteristics. Weather factors are considered in the external feature module. The HDF-DL method is tested on a large-scale metro system in China, in which eight baseline models are designed. The results show that the HDF-DL method achieves high prediction accuracy across multiple time granularities, with a mean absolute percentage error of about 10%. OD passenger flow in every departure time interval can be predicted with high and stable accuracy, effectively capturing temporal characteristics. The modular design of HDF-DL, which fuses high-dimensional features and employs appropriate neural networks for different data types, significantly reduces prediction errors and outperforms baseline models.

The role of news sentiment in salmon price prediction using deep learning

This paper employs deep learning models and sentiment analysis to predict salmon spot prices. Our data includes historical price data and sentiment scores from 2018 to 2022. We extract sentiment scores from salmon-related news headlines by using FinBERT and TextBlob. We begin with price prediction using only historical price data and then introduce sentiment scores to improve the prediction accuracy of deep learning models. We find that the prediction performance of deep learning models outperforms traditional prediction methods in the salmon market. Our primary hybrid CNN-LSTM model outperforms other deep learning models and traditional models. Additionally, deep learning models incorporating sentiment scores exhibit reduced prediction errors. Our findings confirm the value of sentiment information in improving forecasting performance. These findings highlight the effectiveness and robustness of our CNN-LSTM model combined with sentiment analysis for price prediction in the salmon market.

PB-Trajectron: Physics bounded neural network for generalized trajectory prediction

Vehicle trajectory prediction is a critical technology in autonomous driving systems, as the quality of prediction directly affects downstream path planning and vehicle control. Although recent studies have shown that models combining kinematic rules with data-driven methods exhibit better interpretability in trajectory prediction, these models often adopt fixed constraint strengths, limiting their generalization ability in diverse traffic scenarios. This fixed constraint strength design restricts the model’s adaptability to different traffic environments.To address this issue, we propose PB-Trajectron, a dynamic integration mechanism that enhances the model’s prediction performance and generalization capability. PB-Trajectron is a dynamic integration mechanism that combines vehicle kinematic rules with deep learning prediction models for generalized vehicle trajectory prediction. The model integrates physics-based motion models and Deep Neural Networks (DNNs) to provide reasonable physical explanations for predicting vehicle trajectories at different speeds. First, we establish two vehicle kinematic constraints applicable to different prediction scenarios, enabling the agent to switch between these constraints based on the causal relationships between vehicle motion states to avoid generating non-physical trajectory results. Second, we propose a kinematic constraint decision framework based on velocity thresholds, which demonstrates adaptive adjustment, allowing the agent to switch tasks according to real-time state conditions and adaptively adjust the contribution of different physical constraints based on actual vehicle speeds. Finally, we investigate the advantages of PB-Trajectron in Out-of-Domain(OOD) generalization.Cross-scenario experiments on the nuScenes dataset show that, compared to previous methods, PB-Trajectron reduces the final displacement error by 7.69% when the prediction step length is 4. Furthermore, out-of-domain generalization tests on the INTERACTION dataset demonstrate that PB-Trajectron achieves a 12.23% reduction in average prediction error on datasets from different countries and scenarios compared to previous methods. The proposed mechanism can better adapt to complex and diverse traffic scenarios, laying the foundation for explainable and robust autonomous driving systems.

Uncertainty and sensitivity analysis of non-acoustic parameters on acoustic characteristics prediction models for porous foam materials

Porous foam materials are widely used in noise control applications. Their acoustic performance can be predicted by acoustic models composed of non-acoustic parameters such as flow resistivity, porosity, tortuosity, viscosity, and characteristic length. However, the influence of these parameters on model outputs varies across different frequency ranges. Therefore, when using a single model for predictions across the entire frequency range, parameters may not fully act within the frequency range where their influence is greatest. To reduce prediction errors and improve the correlation between predicted numerical results and experimental data, uncertainty and sensitivity analyses can be conducted to determine the extent of each parameter's influence at different frequency ranges. In this study, polyurethane foam materials are investigated, and uncertainty and sensitivity analyses are performed on two acoustic models. Firstly, the uncertainty of parameters is assessed using the Monte Carlo method. Secondly, Sobol index method is employed to reveal the influence of parameters across various frequency ranges. Finally, by comparing the results of both analyses, the optimal frequency ranges for each parameter can be summarized. This provides valuable reference for enhancing the accuracy of acoustic model predictions.

Investigation on transition characteristics of hydrofoil boundary layer based on algebraic local-correlation-based transition modeling model

Hydrofoil shapes are used for the marine turbine blades to capture kinetic energy from water currents effectively. Predicting transitions is a critical concern when studying the hydrofoil boundary layer. This paper analyzed the transitional behavior of the boundary layer in the National Advisory Committee of Aeronautics (NACA) hydrofoil, NACA0009, with a blunt trailing edge using the Algebraic Local-Correlation-based Transition Modeling (Algebraic LCTM) model. First, through sensitivity analysis, the effects of the maximum y+ (the dimensionless distance y to the wall), grid expansion ratio, number of normal and streamlined grids, and timescale on transition prediction were studied. The results indicate that finer y+ value and appropriate grid expansion ratios can improve the accuracy of transition prediction, while the influence of timescale on the prediction results is relatively small within the range of Courant number theory values. Second, further analysis was conducted on the transition prediction performance under different Reynolds numbers. It was found that the model predictions were consistent with experimental values at low Reynolds numbers, but the predicted transition position was advanced at high Reynolds numbers, mainly because of the significant disparity in eddy viscosity coefficients within the free flow field. In the study of leading-edge roughness bands' impact on boundary layer transition for hydrofoil, the introduction of roughness significantly expedited the transition process. The Algebraic LCTM model outperformed the gamma (γ) transition model, reducing prediction errors by 5–40% for boundary layer parameters and maintaining errors between 0.005 and 4% for wake vortex shedding frequency, as opposed to the γ model's 0–23%. The results of this study provide a theoretical basis for hydrofoil design.

Enhancing photovoltaic power prediction using a CNN-LSTM-attention hybrid model with Bayesian hyperparameter optimization

Improving the accuracy of solar power forecasting is crucial to ensure grid stability, optimize solar power plant operations, and enhance grid dispatch efficiency. Although hybrid neural network models can effectively address the complexities of environmental data and power prediction uncertainties, challenges such as labor-intensive parameter adjustments and complex optimization processes persist. Thus, this study proposed a novel approach for solar power prediction using a hybrid model (CNN-LSTM-attention) that combines a convolutional neural network (CNN), long short- term memory (LSTM), and attention mechanisms. The model incorporates Bayesian optimization to refine the parameters and enhance the prediction accuracy. To prepare high-quality training data, the solar power data were first preprocessed, including feature selection, data cleaning, imputation, and smoothing. The processed data were then used to train a hybrid model based on the CNN-LSTM-attention architecture, followed by hyperparameter optimization employing Bayesian methods. The experimental results indicated that within acceptable model training times, the CNN-LSTM-attention model outperformed the LSTM, GRU, CNN-LSTM, CNN-LSTM with autoencoders, and parallel CNN-LSTM attention models. Furthermore, following Bayesian optimization, the optimized model demonstrated significantly reduced prediction errors during periods of data volatility compared to the original model, as evidenced by MRE evaluations. This highlights the clear advantage of the optimized model in forecasting fluctuating data.