Convolutional Autoencoder Neural Network Research Articles

Microgrid energy optimization management considering consumer segmentation optimization and dynamic time intervals

ABSTRACT To prioritize critical loads and enhance microgrid energy management efficiency, this study introduces a method that combines consumer segmentation optimization and dynamic time intervals. A penalty-based approach is used to minimize unmet load demands, prioritizing energy for critical loads. A joint classification model for multi-type consumers is developed using a convolutional neural network autoencoder and hierarchical clustering. Dynamic time intervals are applied to reduce iterations and improve memory and processor efficiency in solving the energy management model. Simulation tests on a microgrid with distributed generation, battery storage, and consumption units confirm the method’s effectiveness. Results show that configuring dynamic time intervals and segmenting loads by priority yield solutions similar to fixed interval methods. Finally, the energy optimization management effects of the proposed method were compared with those of two latest methods. The comparison results demonstrate that if a microgrid underwent four different disconnection scenarios from the main distribution network, the proposed method saves 23.15%, 23.08%, 23.79%, and 34.61% time to achieve energy optimization management compared with that of the first latest method, and 24.20%, 23.87%, 25.11%, and 36.18% time than that of the second latest method.

Breast tumor segmentation in ultrasound using distance-adapted fuzzy connectedness, convolutional neural network, and active contour

This study addresses computer-aided breast cancer diagnosis through a hybrid framework for breast tumor segmentation in ultrasound images. The core of the three-stage method is based on the autoencoder convolutional neural network. In the first stage, we prepare a hybrid pseudo-color image through multiple instances of fuzzy connectedness analysis with a novel distance-adapted fuzzy affinity. We produce different weight combinations to determine connectivity maps driven by particular image specifics. After the hybrid image is processed by the deep network, we adjust the segmentation outcome with the Chan-Vese active contour model. We find the idea of incorporating fuzzy connectedness into the input data preparation for deep-learning image analysis our main contribution to the study. The method is trained and validated using a combined dataset of 993 breast ultrasound images from three public collections frequently used in recent studies on breast tumor segmentation. The experiments address essential settings and hyperparameters of the method, e.g., the network architecture, input image size, and active contour setup. The tumor segmentation reaches a median Dice index of 0.86 (mean at 0.79) over the combined database. We refer our results to the most recent state-of-the-art from 2022–2023 using the same datasets, finding our model comparable in segmentation performance.

A deep-learning-based surrogate modeling method with application to plasma processing

Rapid and accurate system evolution predictions are crucial in scientific and engineering research. However, the complexity of processing systems, involving multiple physical field couplings and slow convergence of iterative numerical algorithms, leads to low computational efficiency. Hence, this paper introduces a systematic deep-learning-based surrogate modeling methodology for multi-physics-coupled process systems with limited data and long-range time evolution, accurately predicting physics dynamics and considerably improving computational efficiency and generalization. The methodology comprises three main components: (1) generating datasets using a sequential sampling strategy, (2) modeling multi-physics spatio-temporal dynamics by designing a heterogeneous Convolutional Autoencoder and Recurrent Neural Network, and (3) training high-precision models with limited data and long-range time evolution via a dual-phase training strategy. A holistic evaluation using a 2D low-temperature plasma processing example demonstrates the methodology’s superior computational efficiency, accuracy, and generalization capabilities. It predicts plasma dynamics approximately 105 times faster than traditional numerical solvers, with a consistent 2% relative error across different generalization tasks. Furthermore, the potential for transferability across various geometries is explored, and the model’s transfer capability is demonstrated with two distinct geometric domain examples.

Motor Fault Diagnosis and Detection with Convolutional Autoencoder (CAE) Based on Analysis of Electrical Energy Data

This study develops a Convolutional Autoencoder (CAE) and deep neural network (DNN)-based model optimized for real-time signal processing and high accuracy in motor fault diagnosis. This model learns complex patterns from voltage and current data and precisely analyzes them in combination with DNN through latent space representation. Traditional diagnostic methods relied on vibration and current sensors, empirical knowledge, or harmonic and threshold-based monitoring, but they had limitations in recognizing complex patterns and providing accurate diagnoses. Our model significantly enhances the accuracy of power data analysis and fault diagnosis by mapping each phase (R, S, and T) of the electrical system to the red, green, and blue (RGB) channels of image processing and applying various signal processing techniques. Optimized for real-time data streaming, this model demonstrated high practicality and effectiveness in an actual industrial environment, achieving 99.9% accuracy, 99.8% recall, and 99.9% precision. Specifically, it was able to more accurately diagnose motor efficiency and fault risks by utilizing power system analysis indicators such as phase voltage, total harmonic distortion (THD), and voltage unbalance. This integrated approach significantly enhances the real-time applicability of electric motor fault diagnosis and is expected to provide a crucial foundation for various industrial applications in the future.

Optimizing solar power generation forecasting in smart grids: a hybrid convolutional neural network -autoencoder long short-term memory approach

Incorporating zero-carbon emission sources of energy into the electric grid is essential to meet the growing energy needs in public and industrial sectors. Smart grids, with their cutting-edge sensing and communication technologies, provide an effective approach to integrating renewable energy resources and managing power systems efficiently. Improving solar energy efficiency remains a challenge within smart grid infrastructures. Nonetheless, recent progress in artificial intelligence (AI) techniques presents promising opportunities to improve energy production control and management. In this study, initially, we employed two different Machine learning (ML) models: Recurrent Neural Network (RNN) and Long Short Term Memory (LSTM), to forecast solar power plant parameters. The analysis revealed that the LSTM model performed better than RNN in terms of Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and Mean Squared Error (MSE). Following a review of the LSTM model’s graphical results, it was further enhanced by combining Autoencoder with LSTM, creating the Autoencoder LSTM (AELSTM) model. Next, a new hybrid model was introduced: Convolutional Neural Network-Autoencoder Long Short-Term Memory (HCAELSTM), designed to boost prediction accuracy. These models were trained on a one-year real-time solar power plant dataset for training and performance assessment. Ultimately, the hybrid HCAELSTM model surpassed the AELSTM model in terms of MAPE, MAE, and MSE. It excelled in MAPE scores for Daily Power Production, Peak Grid Power Production, and Solar Radiance, achieving low scores of 1.175, 2.116, and 1.592 respectively, demonstrating superior accuracy. The study underscores the importance of AI and ML, in particular, the hybrid model HCAELSTM, in enhancing the smart grid’s ability to integrate renewable energy sources. The hybrid model excels at accurately forecasting key measurements, improving solar power generation efficiency within the smart grid system which also plays a key role in the broader shift toward the fourth energy revolution.

Reduced-Order Modeling of Steady and Unsteady Flows with Deep Neural Networks

Large-eddy and direct numerical simulations generate vast data sets that are challenging to interpret, even for simple geometries at low Reynolds numbers. This has increased the importance of automatic methods for extracting significant features to understand physical phenomena. Traditional techniques like the proper orthogonal decomposition (POD) have been widely used for this purpose. However, recent advancements in computational power have allowed for the development of data-driven modal reduction approaches. This paper discusses four applications of deep neural networks for aerodynamic applications, including a convolutional neural network autoencoder, to analyze unsteady flow fields around a circular cylinder at Re = 100 and a supersonic boundary layer with Tollmien–Schlichting waves. The autoencoder results are comparable to those obtained with POD and spectral POD. Additionally, it is demonstrated that the autoencoder can compress steady hypersonic boundary-layer profiles into a low-dimensional vector space that is spanned by the pressure gradient and wall-temperature ratio. This paper also proposes a convolutional neural network model to estimate velocity and temperature profiles across different hypersonic flow conditions.

Integration of pre-treatment computational radiomics, deep radiomics, and transcriptomics enhances soft-tissue sarcoma patient prognosis

Our objective was to capture subgroups of soft-tissue sarcoma (STS) using handcraft and deep radiomics approaches to understand their relationship with histopathology, gene-expression profiles, and metastatic relapse-free survival (MFS). We included all consecutive adults with newly diagnosed locally advanced STS (N = 225, 120 men, median age: 62 years) managed at our sarcoma reference center between 2008 and 2020, with contrast-enhanced baseline MRI. After MRI postprocessing, segmentation, and reproducibility assessment, 175 handcrafted radiomics features (h-RFs) were calculated. Convolutional autoencoder neural network (CAE) and half-supervised CAE (HSCAE) were trained in repeated cross-validation on representative contrast-enhanced slices to extract 1024 deep radiomics features (d-RFs). Gene-expression levels were calculated following RNA sequencing (RNAseq) of 110 untreated samples from the same cohort. Unsupervised classifications based on h-RFs, CAE, HSCAE, and RNAseq were built. The h-RFs, CAE, and HSCAE grouping were not associated with the transcriptomics groups but with prognostic radiological features known to correlate with lower survivals and higher grade and SARCULATOR groups (a validated prognostic clinical-histological nomogram). HSCAE and h-RF groups were also associated with MFS in multivariable Cox regressions. Combining HSCAE and transcriptomics groups significantly improved the prognostic performances compared to each group alone, according to the concordance index. The combined radiomic-transcriptomic group with worse MFS was characterized by the up-regulation of 707 genes and 292 genesets related to inflammation, hypoxia, apoptosis, and cell differentiation. Overall, subgroups of STS identified on pre-treatment MRI using handcrafted and deep radiomics were associated with meaningful clinical, histological, and radiological characteristics, and could strengthen the prognostic value of transcriptomics signatures.

Toward ultra-efficient high-fidelity predictions of wind turbine wakes: Augmenting the accuracy of engineering models with machine learning

This study proposes a novel machine learning (ML) methodology for the efficient and cost-effective prediction of high-fidelity three-dimensional velocity fields in the wake of utility-scale turbines. The model consists of an autoencoder convolutional neural network with U-Net skipped connections, fine-tuned using high-fidelity data from large-eddy simulations (LES). The trained model takes the low-fidelity velocity field cost-effectively generated from the analytical engineering wake model as input and produces the high-fidelity velocity fields. The accuracy of the proposed ML model is demonstrated in a utility-scale wind farm for which datasets of wake flow fields were previously generated using LES under various wind speeds, wind directions, and yaw angles. Comparing the ML model results with those of LES, the ML model was shown to reduce the error in the prediction from 20% obtained from the Gauss Curl hybrid (GCH) model to less than 5%. In addition, the ML model captured the non-symmetric wake deflection observed for opposing yaw angles for wake steering cases, demonstrating a greater accuracy than the GCH model. The computational cost of the ML model is on par with that of the analytical wake model while generating numerical outcomes nearly as accurate as those of the high-fidelity LES.

Performance comparison analysis of classification methodologies for effective detection of intrusions

Intrusion detection systems (IDS) are critical in many applications, including cloud environments. The intrusion poses a security threat and extracts privacy data and information from the cloud. Additionally, intrusions can cause damage to system hardware, resulting in significant financial losses and exposing critical IT infrastructure to risk. To overcome these issues, this study employs the performance comparison analysis for IDS, which has been performed with different models like Autoencoder Convolutional neural network (AE+CNN), Random forest K-means clustering assisted deep neural network (RF+K-means+DNN), Autoencoder K-means clustering assisted long short term memory (AE+K-means+LSTM), Alexnet+Bi-GRU, AE+Alexnet+Bi-GRU and Wild horse AlexNet assisted Bi-directional Gated Recurrent Unit (WABi-GRU) models to choose the best methodology for effective detection of intrusions. The data needed for the analysis is collected from CICIDS2018, UNSW-NB15, NSL-KDD and ToN-IoT datasets. The collected data are pre-processed using data normalization and data cleaning. Finally, the best model has been chosen for effective intrusion detection, which is used for further processes. Various performances, such as accuracy, precision, recall, and f1-score, are analyzed for various existing and proposed models. From this performance comparison of six models such as AE+CNN, RF+K-means+DNN, AE+K-means+LSTM, Alexnet+Bi-GRU, AE+Alexnet+Bi-GRU and WABi-GRU. WABi-GRU can attain an accuracy of 99.890 % for multi-class classification in the CICIDS 2018 dataset, 99.7 % in the NSL-KDD dataset, 99.53 % in the UNSW-NB 15 dataset and 99.988 % for the ToN-IoT dataset. In this analysis, the models containing AlexNet Bi-GRU-based models can obtain better performances than other existing models. The WABi-GRU model can obtain better results than other models.

Multitask-Guided Deep Clustering With Boundary Adaptation.

Multitask learning uses external knowledge to improve internal clustering and single-task learning. Existing multitask learning algorithms mostly use shallow-level correlation to aid judgment, and the boundary factors on high-dimensional datasets often lead algorithms to poor performance. The initial parameters of these algorithms cause the border samples to fall into a local optimal solution. In this study, a multitask-guided deep clustering (DC) with boundary adaptation (MTDC-BA) based on a convolutional neural network autoencoder (CNN-AE) is proposed. In the first stage, dubbed multitask pretraining (M-train), we construct an autoencoder (AE) named CNN-AE using the DenseNet-like structure, which performs deep feature extraction and stores captured multitask knowledge into model parameters. In the second phase, the parameters of the M-train are shared for CNN-AE, and clustering results are obtained by deep features, which is termed as single-task fitting (S-fit). To eliminate the boundary effect, we use data augmentation and improved self-paced learning to construct the boundary adaptation. We integrate boundary adaptors into the M-train and S-fit stages appropriately. The interpretability of MTDC-BA is accomplished by data transformation. The model relies on the principle that features become important as the reconfiguration loss decreases. Experiments on a series of typical datasets confirm the performance of the proposed MTDC-BA. Compared with other traditional clustering methods, including single-task DC algorithms and the latest multitask clustering algorithms, our MTDC-BA achieves better clustering performance with higher computational efficiency. Deep features clustering results demonstrate the stability of MTDC-BA by visualization and convergence verification. Through the visualization experiment, we explain and analyze the whole model data input and the middle characteristic layer. Further understanding of the principle of MTDC-BA. Through additional experiments, we know that the proposed MTDC-BA is efficient in the use of multitask knowledge. Finally, we carry out sensitivity experiments on the hyper-parameters to verify their optimal performance.

A component diagnostic and prognostic framework for pump bearings based on deep learning with data augmentation

To support the mission of providing safe electricity generation with a high capacity factor, a Predictive Maintenance (PdM) framework using Machine Learning Models (MLM) to optimize component maintenance operations is developed. Using sensor measurements to better predict the true component’s Remaining Useful Life (RUL), the PdM framework has the potential to optimize maintenance costs by performing maintenance only when necessary. The PdM framework to pump bearings, the framework consists of a Convolutional Neural Network Autoencoder (CNN-AE) to detect component deviations from normality, a CNN to characterize component fault modes, and a Bayesian Neural Network (BNN) to estimate the component RUL with uncertainty. To increase the number of training samples, a synthetic data generation procedure was developed and includes procedures to recreate the fault-specific characteristic frequencies for diagnostics and the degradation trends for prognostics. The MLMs trained on the synthetic data are tested on the Center for Intelligent Maintenance Systems (IMS) dataset to showcase how well the synthetic data replicates measurement data. Utilizing this framework, the PdM was found to delay maintenance on average by a total of 8.92 years over 40 years and decrease the unexpected component failure rate from 10% to 0% when compared to traditional maintenance philosophies.

Fast fluid–structure interaction simulation method based on deep learning flow field modeling

The rapid acquisition of high-fidelity flow field information is of great significance for engineering applications such as multi-field coupling. Current research in flow field modeling predominantly focuses on low Reynolds numbers and periodic flows, exhibiting weak generalization capabilities and notable issues with temporal inferring error accumulation. Therefore, we establish a reduced order model (ROM) based on Convolutional Auto-Encoder (CAE) and Long Short-Term Memory (LSTM) neural network and propose an unsteady flow field modeling method for the airfoil with a high Reynolds number and strong nonlinear characteristics. The attention mechanism and weak physical constraints are integrated into the model architecture to improve the modeling accuracy. A broadband excitation training strategy is proposed to overcome the error accumulation problem of long-term inferring. With only a small amount of latent codes, the relative error of the flow field reconstructed by CAE can be less than 5‰. By training LSTM with broadband excitation signals, stable dynamic evolution can be achieved in the time dimension. CAE-LSTM can accurately predict the forced response and complex limit cycle behavior of the airfoil in a wide range of amplitude and frequency under subsonic/transonic conditions. The relative errors of predicted variables and integral force are less than 1%. The fluid–structure interaction framework is built by coupling the ROM and motion equations of the structure. CAE-LSTM predicts the time series response of pitch displacement and moment coefficient at different reduced frequencies, which is in good agreement with computational fluid dynamics, and the simulation time savings exceed one order of magnitude.

An intelligent fault detection and diagnosis monitoring system for reactor operational resilience: Unknown fault detection

Multiple advanced reactor designs envision deployment scenarios that feature reactor operations with significantly reduced operating staff compared to present reactors, or even feature fully autonomous operations. Since many of these new reactor designs, like small modular reactors (SMRs) and microreactors, also include extended fuel cycles which limit inspection intervals, load-following capabilities, and reduced access to critical components, safe and reliable semi- or fully autonomous operations must be enabled through an on-line monitoring (OLM) system that effectively detects and diagnoses any faults in the reactor. This paper presents the development of the Fault Detection Module of the previously proposed intelligent OLM system: the Fault Detection and Diagnosis Monitoring System (FDDMS). The Fault Detection Module monitors various plant parameters from 86 sensors throughout the nuclear power plant and accurately defines operations as nominal or abnormal. The methodology leverages the advantages of unsupervised learning and semi-supervised optimization data-driven techniques to successfully detect any fault in the system in real-time without fault information a priori. This allows the Fault Detection Module to identify any unknown fault that may affect the system. When integrated with the rest of the FDDMS architecture, the Fault Detection Module provides power transient dependent fault detection by using a separate model for steady state, ramping up in power, and ramping down in power operations. This makes the FDDMS especially applicable for load-following reactor systems. The Fault Detection Module architecture features a dimensionality reduction algorithm followed by an anomaly detection method. Three dimensionality reduction techniques: 1) Principal Component Analysis (PCA), 2) Deep Neural Network Autoencoder (DNNAE), and 3) Convolutional Neural Network Autoencoder (CNNAE); and three anomaly detection methods: 1) One-Class Support Vector Machine (OC-SVM), 2) Clustering (DBSCAN), and 3) Reconstruction Error (RecErr) thresholding were compared to find the best model for each power transient dataset. Precision rates, recall rates, F1-scores, and accuracies were used to compare model performance. The best models produced total accuracies of 94.50%, 93.63%, and 91.76% for the steady state, ramp up, and ramp down datasets, respectively, or 99.19%, 98.97%, and 97.09% when omitting 1%-severe fault cases. Specifically, the best performing models for the steady state, ramp up, and ramp down cases were the CNNAE-RecErr, DNNAE-DBSCAN, and CNNAE-RecErr, respectively. Recall rates were calculated for various fault types at multiple severity levels. The results show that the Fault Detection Module can identify any type of fault in the system with a delay of a few seconds on average. Additionally, robustness against noisy sensor data was tested, with models maintaining 100% accuracy at various levels of noise, and an illustrative real-time application of the methodology is provided.

Application of machine learning in determining and resolving state estimation anomalies in power systems

The state estimation (SE) process is one of the most important and efficient tools in achieving this goal. However, the occurrence of anomalies in the power grid, such as false data injection (FDI), can significantly impact the accuracy of the SE results. FDI results in reduced accuracy of SE results, potentially putting the power system in a critical situation. This paper addresses the limitations of simple residual-based algorithms in detecting FDI and proposes a scenario that utilizes machine learning (ML) models such as auto-encoder (AE), long short-term memory auto-encoder (LSTM AE), and 1-D convolutional neural network auto-encoder (1-D CNN AE) algorithms. The proposed method aims to offer a more effective FDI detection approach in the power grid, exhibiting higher detection accuracy. Also, the paper introduces the LSTM variational auto-encoder (LSTM VAE) algorithm for reconstructing anomalous data. By utilizing the LSTM VAE, anomalous data can be transformed to closely resemble the original data with an acceptable level of accuracy. Moreover, in critical situations involving FDI, the power system can maintain normal operation by employing the proposed method to reconstruct anomalous data. Finally, the performance of the presented methods is evaluated on the IEEE 14-bus, 30-bus, and 118-bus test networks. The results are then presented and discussed to demonstrate the effectiveness of the proposed method.

Design and implementation of an Automatic Deep Stacked Sparsely Connected Convolutional Autoencoder (ADSSCCA) neural network for remote sensing lithological mapping using calculated dropout

Design and implementation of an Automatic Deep Stacked Sparsely Connected Convolutional Autoencoder (ADSSCCA) neural network for remote sensing lithological mapping using calculated dropout

Direct Exoplanet Detection using Convolutional Image Reconstruction (ConStruct): A New Algorithm for Post-processing High-contrast Images

We present a novel machine-learning approach for detecting faint point sources in high-contrast adaptive optics (AO) imaging data sets. The most widely used algorithms for primary subtraction aim to decouple bright stellar speckle noise from planetary signatures by subtracting an approximation of the temporally evolving stellar noise from each frame in an imaging sequence. Our approach aims to improve the stellar noise approximation and increase the planet detection sensitivity by leveraging deep learning in a novel direct imaging post-processing algorithm. We show that a convolutional autoencoder neural network, trained on an extensive reference library of real imaging sequences, accurately reconstructs the stellar speckle noise at the location of a potential planet signal. This tool is used in a post-processing algorithm we call Direct Exoplanet Detection with Convolutional Image Reconstruction, or ConStruct. The reliability and sensitivity of ConStruct are assessed using real Keck/NIRC2 angular differential imaging data sets. Of the 30 unique point sources we examine, ConStruct yields a higher signal-to-noise ratio than traditional principal component analysis-based processing for 67% of the cases and improves the relative contrast by up to a factor of 2.6. This work demonstrates the value and potential of deep learning to take advantage of a diverse reference library of point-spread function realizations to improve direct imaging post-processing. ConStruct and its future improvements may be particularly useful as tools for post-processing high-contrast images from JWST and extreme AO instruments, both for the current generation and those being designed for the upcoming 30 m class telescopes.

Deep convolutional autoencoders for the time–space reconstruction of liquid rocket engine flames

The integration of high-fidelity numerical simulations into the rocket engine design-cycle promises to cut costs in an ever more competitive launch market. Although still being intractable in the optimization of most of combustion devices, their potential can be harnessed via appropriate reduced order models (ROMs). In recent years, significant progress has been made in the ROM literature through the adoption of non-linear methods, coming from the realm of deep learning. These methods promise to deliver where conventional linear techniques, such as Proper Orthogonal Decomposition (POD), fall short due to their inherent qualities.In the current work, we carry out a first attempt to develop purely data-driven emulators for a shear-coaxial liquid rocket engine injector. In particular, a Convolutional Neural Network autoencoder (CNN-AE) with a Multi-Layer Perceptron (MLP) latent dynamics encoder architecture is considered. The main objective is to assess the capacity of autoencoders based models to reconstruct, in time and space, the instantaneous temperature field of a non-canonical, complex, reactive, flow configuration. The data used is made of Large Eddy Simulations (LES) of three distinct, real-life inspired, injector configurations.The autoencoder based models show superior reconstruction capabilities compared to POD with similar latent-space dimension. Moreover, the time–space reconstructed fields preserve the spectral content of the snapshots, with only a small loss in gain at high frequencies. These results serve as proof-of-concept for further developments on surrogate models of liquid rocket engine injectors.

Analysis of Damage in a Warren Truss Bridge Using CAE and DANN Neural Networks

Bridges require constant monitoring to detect damages. This study analyzes the Japanese Warren truss bridge using neural networks: Convolutional Autoencoder (CAE) and Domain-Adversarial Neural Network (DANN). The methodology focuses on two aspects: reconstruction of bridge acceleration data with CAE and damage analysis with DANN using CAE-processed data. CAE is trained to reconstruct acceleration data by recovering missing data and generating new data to improve dataset quality. Then, DANN uses this data to identify and evaluate anomalies in the bridge structure. The results obtained were 84% accuracy with respect to the synthetic data generated with the CAE network and 95% accuracy and an F1-score of 92% in the damage analysis of the bridge with the DANN network.

A 3d convolutional neural network autoencoder for predicting solvent configuration changes in condensed phase biomass reactions

Self-supervised 3d CNN autoencoder combined with a distance-based classifier is trained on first principles simulations to predict the extent of solvent configuration changes during reactive chemical transformations.

SERS combined with the SAE-CNN model for estimating apple rootstocks under heavy metal copper stress

Copper-containing chemicals are commonly used in orchard management, and their use over a long time causes excessive copper content in orchard soil. Surface-enhanced Raman spectroscopy (SERS) technique combined with deep learning (DL) methods were used to characterize the response of apple rootstocks under heavy metal Cu stress. The hydroponic method prepared Apple rootstocks under five concentration gradients of heavy metal Cu stress as the experimental samples. The Raman spectral data were collected and subjected to spectral preprocessing. A stacked auto-encoder convolutional neural network (SAE-CNN) model was used to develop a rapid discriminative model for root, stem, and leaf stress levels of heavy metal Cu on apple rootstocks. The SAE-CNN models all outperformed the traditional models, with performance metrics above 99%. The results showed that the proposed SAE-CNN model could rapidly discriminate heavy metal Cu in apple rootstocks.