Input Feature Set Research Articles

A data driven approach to assess the petrophysical parametric sensitivity for lithology identification based on ensemble learning

Lithology identification is essential for geological analysis and reservoir characterization of hydrocarbon bearing formation. Conventional approaches of lithology identification are not only inconvenient but also time-consuming and expensive. Machine learning (ML) algorithms have been introduced as an effective, reliable, and accurate method of lithology identification using well-log data due to their ability to recognize the nonlinear relationship between log responses and lithology. In this study, we employed the XGBoost classifier model to assess the parametric sensitivity of the logging features in identifying lithology. Sequential Forward Feature Selection (SFFS) approach was applied to determine the most significant set of logging features that yields maximum classification accuracy. Nine logging parameters were used as the input of the XGBoost model to classify five lithology classes. Hyper-parameters were tuned within appropriate tuning search range to optimize the model performance along with Grid Search and 5-fold cross validation. The training and testing accuracy obtained from the model are 96% and 90.97%, respectively. The optimized XGBoost model showed the lowest precision, recall, and F1 score for the clay class. The shale class, on the other side, had the highest precision, recall, and F1 score of 95%, 94%, and 94%, respectively. The best set of logging features for identifying formation lithology is determined to be [PEF, SP, LLD, CALI, MSFL, NPHI, LLS], with a classification accuracy of 92.10% for the testing dataset. The least significant logging parameters for identifying lithology were found to be DT and GR, as their addition with the best input feature set, resulted in a negative deviation in classification accuracy. Finally, based on the classification accuracy, we ranked the top three sets of logging features, those contributed significantly.

An ultra-short-term forecasting method for multivariate loads of user-level integrated energy systems in a microscopic perspective: based on multi-energy spatio-temporal coupling and dual-attention mechanism

An ultra-short-term multivariate load forecasting method under a microscopic perspective is proposed to address the characteristics of user-level integrated energy systems (UIES), which are small in scale and have large load fluctuations. Firstly, the spatio-temporal correlation of users’ energy use behavior within the UIES is analyzed, and a multivariate load input feature set in the form of a class image is constructed based on the various types of load units. Secondly, in order to maintain the feature independence and temporal integrity of each load during the feature extraction process, a deep neural network architecture with spatio-temporal coupling characteristics is designed. Among them, the multi-channel parallel convolutional neural network (MCNN) performs independent spatial feature extraction of the 2D load component pixel images at each moment in time, and feature fusion of various types of load features in high dimensional space. A bidirectional long short-term memory network (BiLSTM) is used as a feature sharing layer to perform temporal feature extraction on the fused load sequences. In addition, a spatial attention layer and a temporal attention layer are designed in this paper for the original input load pixel images and the fused load sequences, respectively, so that the model can better capture the important information. Finally, a multi-task learning approach based on the hard sharing mechanism achieves joint prediction of each load. The measured load data of a UIES is analyzed as an example to verify the superiority of the method proposed in this paper.

Polynomial surface-fitting evaluation of new energy maximum power generation capacity based on random forest association analysis and support vector regression

Focusing on frequency problems caused by wind power integration in ultra-high-voltage DC systems, an accurate assessment of the maximum generation capacity of large-scale new energy sources can help determine the available frequency regulation capacity of new energy sources and improve the frequency stability control of power systems. First, a random forest model is constructed to analyze the key features and select the indexes significantly related to the generation capacity to form the input feature set. Second, by establishing an iterative construction model of the polynomial fitting surface, data are maximized by the upper envelope surface, and an effective sample set is constructed. Furthermore, a new energy maximum generation capacity assessment model adopts the support vector machine regression algorithm under the whale optimization algorithm to derive the correspondence between the input features and maximum generation capacity of new energy sources. Finally, we validate the applicability and effectiveness of the new maximum energy generation capacity evaluation model based on the results of an actual wind farm.

Early Identification of Cotton Fields Based on Gf-6 Images in Arid and Semiarid Regions (China)

Accurately grasping the distribution and area of cotton for agricultural irrigation scheduling, intensive and efficient management of water resources, and yield estimation in arid and semiarid regions is of great significance. In this paper, taking the Xinjiang Shihezi oasis agriculture region as the study area, extracting the spectroscopic characterization (R, G, B, panchromatic), texture feature (entropy, mean, variance, contrast, homogeneity, angular second moment, correlation, and dissimilarity) and characteristics of vegetation index (normalized difference vegetation index/NDVI, ratio vegetation index/DVI, difference vegetation index/RVI) in the cotton flowering period before and after based on GF-6 image data, four models such as the random forests (RF) and deep learning approach (U-Net, DeepLabV3+ network, Deeplabv3+ model based on attention mechanism) were used to identify cotton and to compare their accuracies. The results show that the deep learning model is better than that of the random forest model. In all the deep learning models with three kinds of feature sets, the recognition accuracy and credibility of the DeepLabV3+ model based on the attention mechanism are the highest, the overall recognition accuracy of cotton is 98.23%, and the kappa coefficient is 96.11. Using the same Deeplabv3+ model based on an attention mechanism with different input feature sets (all features and only spectroscopic characterization), the identification accuracy of the former is much higher than that of the latter. GF-6 satellite image data in the field of crop type recognition has great application potential and prospects.

DDEL-15. EXPLORING THE USE OF MACHINE LEARNING TO PREDICT BLOOD BRAIN BARRIER PERMEABILITY OF DRUG MOLECULES

Abstract The blood-brain barrier (BBB) is a selective boundary of the central nervous system (CNS) that plays a critical role in protecting the brain microenvironment by allowing passage of certain molecules. Clinical experiments accurately determine which chemotherapy drugs effectively cross the BBB to reach the tumor site but are also time consuming and labor intensive. Machine learning offers the ability to rapidly screen large datasets of drug molecules and assess their potential for CNS therapeutic effect. The goal of this study is to compare the performance across machine learning, deep learning, and transfer learning methods in predicting BBB permeability on public drug datasets. The dataset used for training and validation was composed of 7,807 compounds compiled from 50 published resources. The machine and deep learning methods used in this study included support vector machines (SVMs), deep neural networks (DNNs), and graph convolutional neural networks (GCNNs). For transfer learning, we first trained DNN models to a single quantum chemical property before appending neural network layers and retraining to the new task of BBB permeability. The prediction accuracies on the validation set for SVM, DNN, GCNN, transfer learning of polarizability, and transfer learning of dipole moment were 82.33%, 83.09%, 87.14%, 76.89%, and 70.23%, respectively. Overall, the results indicate that GCNN was the best performing model. This highlights GCNNs ability to learn the molecular features most relevant to the predictive task in the first stage of the algorithm. Future work entails expanding the set of input features to include key chemical and structural information such as the presence or absence of certain functional groups as well as training more transfer learning models to other quantum chemical properties. This study further motivates the predictive capability of machine learning methods in identifying drug compounds with potential CNS-activity.

Sequential backward feature selection for optimizing permanent strain model of unbound aggregates

This study proposes a novel framework for identifying the optimal feature set required to predict the permanent strain of unbound aggregates. An experimental database consisting of 16 input features is preprocessed and the performance of 10 machine learning models is evaluated. The best-performing model is then paired with a sequential backward selection algorithm to determine the optimal feature set for predicting the permanent strain. Finally, the selected features are used to predict the permanent strain, and the performance is compared with those obtained from the principal components analysis. Six features are selected as the optimal feature set. Furthermore, the selected features accurately predict permanent strain with a root mean square error value of 0.014, which is smaller than those obtained from principal components analysis. Thus, the feature selection approach for machine learning models effectively predicts the permanent strain of unbound aggregates using a limited set of input features.

Machine learning-based prediction of longitudinal cognitive decline in early Parkinson’s disease using multimodal features

Patients with Parkinson’s Disease (PD) often suffer from cognitive decline. Accurate prediction of cognitive decline is essential for early treatment of at-risk patients. The aim of this study was to develop and evaluate a multimodal machine learning model for the prediction of continuous cognitive decline in patients with early PD. We included 213 PD patients from the Parkinson’s Progression Markers Initiative (PPMI) database. Machine learning was used to predict change in Montreal Cognitive Assessment (MoCA) score using the difference between baseline and 4-years follow-up data as outcome. Input features were categorized into four sets: clinical test scores, cerebrospinal fluid (CSF) biomarkers, brain volumes, and genetic variants. All combinations of input feature sets were added to a basic model, which consisted of demographics and baseline cognition. An iterative scheme using RReliefF-based feature ranking and support vector regression in combination with tenfold cross validation was used to determine the optimal number of predictive features and to evaluate model performance for each combination of input feature sets. Our best performing model consisted of a combination of the basic model, clinical test scores and CSF-based biomarkers. This model had 12 features, which included baseline cognition, CSF phosphorylated tau, CSF total tau, CSF amyloid-beta1-42, geriatric depression scale (GDS) scores, and anxiety scores. Interestingly, many of the predictive features in our model have previously been associated with Alzheimer’s disease, showing the importance of assessing Alzheimer’s disease pathology in patients with Parkinson’s disease.

Autoencoder Feature Residuals for Network Intrusion Detection: One-Class Pretraining for Improved Performance

The proliferation of novel attacks and growing amounts of data has caused practitioners in the field of network intrusion detection to constantly work towards keeping up with this evolving adversarial landscape. Researchers have been seeking to harness deep learning techniques in efforts to detect zero-day attacks and allow network intrusion detection systems to more efficiently alert network operators. The technique outlined in this work uses a one-class training process to shape autoencoder feature residuals for the effective detection of network attacks. Compared to an original set of input features, we show that autoencoder feature residuals are a suitable replacement, and often perform at least as well as the original feature set. This quality allows autoencoder feature residuals to prevent the need for extensive feature engineering without reducing classification performance. Additionally, it is found that without generating new data compared to an original feature set, using autoencoder feature residuals often improves classifier performance. Practical side effects from using autoencoder feature residuals emerge by analyzing the potential data compression benefits they provide.

Adaboost Model-Based Approach for Effectively Detecting Spam in IoT Devices

Abstract: The number of Internet of Things (IoT) devices in smart homes is rising swiftly, producing enormous amounts of data that are mostly transmitted over wireless communication channels. IoT devices can be at risk from a variety of threats, such as hacker attacks, cyberattacks, erratic network connectivity, data leakage, etc. By evaluating vast amounts of data using complex algorithms, machine learning may assist uncover spam in IoT data. It can also help to raise the security level of the IoT system in smart homes by utilising statistical analysis and machine learning to find anomalies in the data. In this work, two machine learning models—the Bagged model and the Adaboost model—are evaluated using a wide range of criteria employing a vast number of input feature sets. Each model generates a spam score using the improved input attributes. The suggested algorithm is used to determine the network's linked IoT devices' spamicity score. The REFIT Smart Home dataset is used to test the suggested method. The outcomes show that the suggested strategy is beneficial when compared to other current plans.

An Evolutionary Neural Network Approach for Slopes Stability Assessment

A current big challenge for developed or developing countries is how to keep large-scale transportation infrastructure networks operational under all conditions. Network extensions and budgetary constraints for maintenance purposes are among the main factors that make transportation network management a non-trivial task. On the other hand, the high number of parameters affecting the stability condition of engineered slopes makes their assessment even more complex and difficult to accomplish. Aiming to help achieve the more efficient management of such an important element of modern society, a first attempt at the development of a classification system for rock and soil cuttings, as well as embankments based on visual features, was made in this paper using soft computing algorithms. The achieved results, although interesting, nevertheless have some important limitations to their successful use as auxiliary tools for transportation network management tasks. Accordingly, we carried out new experiments through the combination of modern optimization and soft computing algorithms. Thus, one of the main challenges to overcome is related to the selection of the best set of input features for a feedforward neural network for earthwork hazard category (EHC) identification. We applied a genetic algorithm (GA) for this purpose. Another challenging task is related to the asymmetric distribution of the data (since typically good conditions are much more common than bad ones). To address this question, three training sampling approaches were explored: no resampling, the synthetic minority oversampling technique (SMOTE), and oversampling. Some relevant observations were taken from the optimization process, namely, the identification of which variables are more frequently selected for EHC identification. After finding the most efficient models, a detailed sensitivity analysis was applied over the selected models, allowing us to measure the relative importance of each attribute in EHC identification.

Day-Ahead Electricity Price Probabilistic Forecasting Based on SHAP Feature Selection and LSTNet Quantile Regression

Electricity prices are a central element of the electricity market, and accurate electricity price forecasting is critical for market participants. However, in the context of increasingly integrated economic markets, the complexity of the electricity system has increased. As a result, the number of factors required to consider in electricity price forecasting is growing. In addition, the high percentage of renewable energy penetration has increased the volatility of electricity generation, making it more challenging to predict prices accurately. In this paper, we propose a probabilistic forecasting method based on SHAP (SHapley Additive exPlanation) feature selection and LSTNet (long- and short-term time-series network) quantile regression. First, to reduce feature redundancy and overfitting, we use the SHAP method to perform feature selection in a high-dimensional input feature set, and specifically analyze the magnitude and manner in which features affect electricity prices. Second, we apply the LSTNet quantile regression model to predict the electricity value under different quantiles. Finally, the probability density function and the prediction interval of the predicted electricity prices are obtained by kernel density estimation. The case of the Danish electricity market validates the effectiveness and accuracy of our proposed method. The accuracy of the proposed method is better than that of other methods, and we assess the importance and direction of the impact of features on electricity prices.

Multi-view radiomics and deep learning modeling for prostate cancer detection based on multi-parametric MRI.

This study aims to develop an imaging model based on multi-parametric MR images for distinguishing between prostate cancer (PCa) and prostate hyperplasia. A total of 236 subjects were enrolled and divided into training and test sets for model construction. Firstly, a multi-view radiomics modeling strategy was designed in which different combinations of radiomics feature categories (original, LoG, and wavelet) were compared to obtain the optimal input feature sets. Minimum-redundancy maximum-relevance (mRMR) selection and least absolute shrinkage selection operator (LASSO) were used for feature reduction, and the next logistic regression method was used for model construction. Then, a Swin Transformer architecture was designed and trained using transfer learning techniques to construct the deep learning models (DL). Finally, the constructed multi-view radiomics and DL models were combined and compared for model selection and nomogram construction. The prediction accuracy, consistency, and clinical benefit were comprehensively evaluated in the model comparison. The optimal input feature set was found when LoG and wavelet features were combined, while 22 and 17 radiomic features in this set were selected to construct the ADC and T2 multi-view radiomic models, respectively. ADC and T2 DL models were built by transferring learning from a large number of natural images to a relatively small sample of prostate images. All individual and combined models showed good predictive accuracy, consistency, and clinical benefit. Compared with using only an ADC-based model, adding a T2-based model to the combined model would reduce the model's predictive performance. The ADCCombinedScore model showed the best predictive performance among all and was transformed into a nomogram for better use in clinics. The constructed models in our study can be used as a predictor in differentiating PCa and BPH, thus helping clinicians make better clinical treatment decisions and reducing unnecessary prostate biopsies.

Land Surface Temperature Estimation from Landsat-9 Thermal Infrared Data Using Ensemble Learning Method Considering the Physical Radiance Transfer Process

Accurately estimating land surface temperature (LST) is a critical concern in thermal infrared (TIR) remote sensing. According to the thermal radiance transfer equation, the observed data in each channel are coupled with both emissivity and atmospheric parameters in addition to the LST. To solve this ill-posed problem, classical algorithms often require the input of external parameters such as land surface emissivity and atmospheric profiles, which are often difficult to obtain accurately and timely, and this may introduce additional errors and limit the applicability of the LST retrieval algorithms. To reduce the dependence on external parameters, this paper proposes a new algorithm to directly estimate the LST from the top-of-atmosphere brightness temperature in Landsat-9 two-channel TIR data (channels 10 and 11) without external parameters. The proposed algorithm takes full advantage of the adeptness of the ensemble learning method to solve nonlinear problems. It considers the physical radiance transfer process and adds the leaving-ground bright temperature and atmospheric water vapor index to the input feature set. The experimental results show that the new algorithm achieves accurate LST estimation results compared with the ground-measured LST and is consistent with the Landsat-9 LST product. In subsequent work, further studies will be undertaken on developing end-to-end deep learning models, mining more in-depth features between TIR channels, and reducing the effect of spatial heterogeneity on accuracy validation.

Placental MRI segmentation based on multi-receptive field and mixed attention separation mechanism

ObjectiveTo reduce the occurrence of massive bleeding during placental abruption in patients with placenta accrete, we established a medical imaging based on multi-receptive field and mixed attention separation mechanism (MRF-MAS) model to improve the accuracy of MRI placenta segmentation and provide a basis for subsequent placenta accreta. MethodsWe propose a placenta MRI segmentation technology using the MRF-MAS framework to develop a medical image diagnostic technique. The model first uses the multi-receptive field feature structure to obtain multi-level information, and improves the expression of features at differing scales. Note that the hybrid attention mechanism combines channel attention and spatial attention, separates the input feature sets and computes the attention separately, and finally reorganizes the feature maps. To show that the model can improve the accuracy of segmenting the placenta, we adopt mean Intersection over Union (IoU), Dice similarity coefficient (Dice) and area under the receiver operating characteristic curve (AUC) with U-Net, Mask RCNN, Deeplab v3 for comparison. ResultsThe four models achieved different outcomes based on our placenta dataset, with our model IoU and Dice up to 0.8169 and 0.8992, which are 5.51% and 3.03% higher than the average of the three comparison models. ConclusionThe model proposed by us is helpful to assist the imaging diagnosis and at the same time provides a quantitative reference for the precise treatment of placenta accreta, assists the Equationtion of the clinical operation plan of the physician, and promotes the precision medicine of placenta accreta.

Machine learning‐based prediction of the seismic response of fault‐crossing natural gas pipelines

AbstractHerein, we utilized machine‐learning (ML) and data‐driven (regression) techniques to tackle a critical infrastructure engineering problem—namely, predicting the seismic response of natural gas pipelines crossing earthquake faults. Such a 3D nonlinear problem can take up to 10 h to solve by performing finite element analysis (FEA), considering the length of the pipeline and a large number of pipe and soil elements. However, the ML and data‐driven techniques can learn the projection rule of input‐output and predict the pipeline response instantaneously given a set of input features. In addition, the well‐trained ML model can be implemented for regional‐scale risk and rapid post‐event damage assessments. In this study, the input for ML comprised approximately 217K nonlinear FEAs, which covered a wide range of combinations of soil, structural and fault properties and yielded critical pipe strain responses under fault‐rupture displacements. We adopted various regression models and physics‐constrained neural networks, which can accurately and rapidly predict the tensile and compressive strains for a broad range of probable fault‐rupture displacements. Performances of various ML and conventional statistical models were systematically examined. Not surprisingly, neural networks exhibited the best performance for this multi‐output regression problem, in whichR2 > 0.95 was achieved for a wide range of fault displacement (FD) levels. Further, we used the trained neural network with 14.5 million Monte‐Carlo‐generated input samples to predict the maximum tensile and compressive strain curves of pipelines. This new dataset aimed at filling the missing input‐output points from the 217K FEAs, and improved the accuracy of the prediction of probability of failure for natural gas pipelines under FD hazards.

Machine learning-based classification of cardiac relaxation impairment using sarcomere length and intracellular calcium transients

Impaired relaxation of cardiomyocytes leads to diastolic dysfunction in the left ventricle. Relaxation velocity is regulated in part by intracellular calcium (Ca2+) cycling, and slower outflux of Ca2+ during diastole translates to reduced relaxation velocity of sarcomeres. Sarcomere length transient and intracellular calcium kinetics are integral parts of characterizing the relaxation behavior of the myocardium. However, a classifier tool that can separate normal cells from cells with impaired relaxation using sarcomere length transient and/or calcium kinetics remains to be developed. In this work, we employed nine different classifiers to classify normal and impaired cells, using ex-vivo measurements of sarcomere kinematics and intracellular calcium kinetics data. The cells were isolated from wild-type mice (referred to as normal) and transgenic mice expressing impaired left ventricular relaxation (referred to as impaired). We utilized sarcomere length transient data with a total of n = 126 cells (n = 60 normal cells and n = 66 impaired cells) and intracellular calcium cycling measurements with a total of n = 116 cells (n = 57 normal cells and n = 59 impaired cells) from normal and impaired cardiomyocytes as inputs to machine learning (ML) models for classification. We trained all ML classifiers with cross-validation method separately using both sets of input features, and compared their performance metrics. The performance of classifiers on test data showed that our soft voting classifier outperformed all other individual classifiers on both sets of input features, with 0.94 and 0.95 area under the receiver operating characteristic curves for sarcomere length transient and calcium transient, respectively, while multilayer perceptron achieved comparable scores of 0.93 and 0.95, respectively. However, the performance of decision tree, and extreme gradient boosting was found to be dependent on the set of input features used for training. Our findings highlight the importance of selecting appropriate input features and classifiers for the accurate classification of normal and impaired cells. Layer-wise relevance propagation (LRP) analysis demonstrated that the time to 50% contraction of the sarcomere had the highest relevance score for sarcomere length transient, whereas time to 50% decay of calcium had the highest relevance score for calcium transient input features. Despite the limited dataset, our study demonstrated satisfactory accuracy, suggesting that the algorithm can be used to classify relaxation behavior in cardiomyocytes when the potential relaxation impairment of the cells is unknown.

Uncertainty quantification for spindle axial thermal error of CNC machine tools considering hysteresis effect

Most of the existing research on thermal error regression modeling of CNC machine tools establishes deterministic models, presents point predictions for thermal errors and focuses on the scenario where there is no hysteresis effect between input and output. There is little research on quantifying the uncertainties of spindle axial thermal error when there exist significant hysteresis effects between the temperature points and spindle axial thermal error. In this regard, this paper presents a novel uncertainty modeling method for spindle axial thermal error. Firstly, to screen out key temperature points, the whole temperature points are sorted according to the minimal redundancy maximal relevance (mRMR) criterion, and the candidate input feature sets are generated. Then, the gated recurrent unit (GRU) model is introduced to handle the hysteresis effects. Finally, the deterministic GRU model is transformed into the prediction interval-based GRU (PI-GRU) model, which can produce prediction intervals (PIs) of the spindle axial thermal error. The correctness and effectiveness of this method are verified by the spindle axial thermal error experiments. In contrast with the Gaussian process regression (GPR) method, the proposed method deals with the hysteresis effects more appropriately and performs better in uncertainty prediction accuracy.

Resnet-based power system frequency security assessment considering frequency spatiotemporal distribution characteristics

Under the background of the accelerated transformation of the modern power system, renewable energy will gradually replace the role of traditional energy in the power system. Because of the randomness, uncertainty, and low inertia of renewable energy, the power system frequency security assessment is becoming a serious issue under the influence of large-scale renewable energy integrations. To achieve the rapid assessment of power system frequency security, a resnet-based frequency security assessment method considering frequency spatiotemporal distribution characteristics is proposed. In this paper, to provide more comprehensive frequency response information, the spatiotemporal characteristics of frequency are introduced and integrated into the input feature set. At the same time, the Resnet network and multi-task-learning framework are applied to construct a frequency security assessment model. Based on the experimental results on the modified IEEE39 bus system, the assessment accuracy is higher than 99%, which shows the excellent assessment performance of the proposed model.

Lithium Battery SOH Estimation Based on Manifold Learning and LightGBM

In order to accurately identify the state of health (SOH) and remaining useful life (RUL) of lithium-ion batteries, this paper proposes an SOH estimation algorithm for lithium-ion batteries based on stream learning and LightGBM. To address the problem of inconsistent data length, which makes it difficult to establish the state mapping relationship between degraded data and health state, the health factors in this paper are extracted from capacity degradation features, entropy features, and correlation coefficient features. Then, the landmark isometric mapping (L-ISOMAP) manifold learning algorithm is used to dimensionally reduce the input feature set and map the high-dimensional features to the low-dimensional space to solve the dimensional explosion problem. Finally, a LightGBM prediction model is developed to perform SOH prediction on different datasets, and the superiority of the multidimensional model is evaluated. The experimental results show that the goodness-of-fit is 0.98 and above, and the MSE values are below 4 × 10−4. Comparing several prediction models, the LightGBM model has the best performance and better results in several indexes, such as MSE and RMSE. Under different working conditions, the proposed model in this paper has a goodness-of-fit of more than 0.98 in dataset B, which proves that the proposed model has a strong generalization ability.

Joint load prediction of multiple buildings using multi-task learning with selected-shared-private mechanism

Short-term load forecasting for residential buildings is of great significance in ensuring the safe and economic operation of the power grid. However, most existing prediction methods focus solely on the temporal characteristics of individual buildings and ignore the spatial correlation between different buildings in a neighborhood. To tackle this problem and realize joint prediction for multiple buildings, a novel multi-task learning model with a selected-shared-private mechanism is proposed in this paper. Firstly, the temporal and spatial characteristics hidden in electricity consumption patterns are analyzed and represented by auto-correlation and cross-correlation, respectively. Then, a correlation-oriented combination strategy is proposed to build input feature set for the prediction model, and the temporal convolutional network is adopted to extract features. Furthermore, a novel selected-shared-private mechanism is designed for multi-task learning to improve the prediction accuracy, which can selectively utilize information from related tasks while learning private features. The proposed model is compared with other methods on the public dataset, and the results demonstrate that the proposed model can make satisfactory joint predictions for multiple buildings with more than 35% accuracy improvement over the support vector machine.