scholarly journals CL-Net: ConvLSTM-Based Hybrid Architecture for Batteries’ State of Health and Power Consumption Forecasting

Mathematics ◽  
2021 ◽  
Vol 9 (24) ◽  
pp. 3326
Author(s):  
Noman Khan ◽  
Ijaz Ul Haq ◽  
Fath U Min Ullah ◽  
Samee Ullah Khan ◽  
Mi Young Lee
Keyword(s):  
Power Consumption ◽  
Short Term Memory ◽  
Mean Squared Error ◽  
Renewable Energy Sources ◽  
The State ◽  
Hybrid Architecture ◽  
State Of Health ◽  
Short Term ◽  
Term Memory ◽  
Long Short Term Memory

Traditional power generating technologies rely on fossil fuels, which contribute to worldwide environmental issues such as global warming and climate change. As a result, renewable energy sources (RESs) are used for power generation where battery energy storage systems (BESSs) are widely used to store electrical energy for backup, match power consumption and generation during peak hours, and promote energy efficiency in a pollution-free environment. Accurate battery state of health (SOH) prediction is critical because it plays a key role in ensuring battery safety, lowering maintenance costs, and reducing BESS inconsistencies. The precise power consumption forecasting is critical for preventing power shortage and oversupply, and the complicated physicochemical features of batteries dilapidation cannot be directly acquired. Therefore, in this paper, a novel hybrid architecture called ‘CL-Net’ based on convolutional long short-term memory (ConvLSTM) and long short-term memory (LSTM) is proposed for multi-step SOH and power consumption forecasting. First, battery SOH and power consumption-related raw data are collected and passed through a preprocessing step for data cleansing. Second, the processed data are fed into ConvLSTM layers, which extract spatiotemporal features and form their encoded maps. Third, LSTM layers are used to decode the encoded features and pass them to fully connected layers for final multi-step forecasting. Finally, a comprehensive ablation study is conducted on several combinations of sequential learning models using three different time series datasets, i.e., national aeronautics and space administration (NASA) battery, individual household electric power consumption (IHEPC), and domestic energy management system (DEMS). The proposed CL-Net architecture reduces root mean squared error (RMSE) up to 0.13 and 0.0052 on the NASA battery and IHEPC datasets, respectively, compared to the state-of-the-arts. These experimental results show that the proposed architecture can provide robust and accurate SOH and power consumption forecasting compared to the state-of-the-art.

scholarly journals State of Health Estimation for Lithium-Ion Batteries Based on Healthy Features and Long Short-Term Memory

IEEE Access ◽  
2020 ◽  
Vol 8 ◽  
pp. 28533-28547 ◽  
Author(s):  
Yitao Wu ◽  
Qiao Xue ◽  
Jiangwei Shen ◽  
Zhenzhen Lei ◽  
Zheng Chen ◽  
...  
Keyword(s):  
Lithium Ion Batteries ◽  
Short Term Memory ◽  
Lithium Ion ◽  
State Of Health ◽  
Short Term ◽  
Term Memory ◽  
Long Short Term Memory

scholarly journals A Hybrid Model for Power Consumption Forecasting Using VMD-Based the Long Short-Term Memory Neural Network

2022 ◽  
Vol 9 ◽  
Author(s):  
Yingjun Ruan ◽  
Gang Wang ◽  
Hua Meng ◽  
Fanyue Qian
Keyword(s):  
Neural Network ◽  
Energy Consumption ◽  
Power Consumption ◽  
Hybrid Model ◽  
Short Term Memory ◽  
Series Data ◽  
Short Term ◽  
Term Memory ◽  
Long Short Term Memory ◽  
Individual Features

Energy consumption prediction is a popular research field in computational intelligence. However, it is difficult for general machine learning models to handle complex time series data such as building energy consumption data, and the results are often unsatisfactory. To address this difficulty, a hybrid prediction model based on modal decomposition was proposed in this paper. For data preprocessing, the variational mode decomposition (VMD) technique was used to used to decompose the original sequence into more robust subsequences. In the feature selection, the maximum relevance minimum redundancy (mRMR) algorithm was chosen to analyse the correlation between each component and the individual features while eliminating the redundancy between individual features. In the forecasting module, the long short-term memory (LSTM) neural network model was used to predict power consumption. In order to verify the performance of the proposed model, three categories of contrast methods were applied: 1) Comparing the hybrid model to a single predictive model, 2) Comparing the hybrid model with the backpropagation neural network (BPNN) to the hybrid model with the LSTM and 3) Comparing the hybrid model using mRMR and the hybrid model using mutual information maximization (MIM). The experimental results on the measured data of an office building in Qingdao show that the proposed hybrid model can improve the prediction accuracy and has better robustness compared to VMD-MIM-LSTM. In the three control groups mentioned above, the R2 value of the hybrid model improved by 10, 3 and 3%, respectively, the values of the mean absolute error (MAE) decreased by 48.9, 41.4 and 35.6%, respectively, and the root mean square error (RMSE) decreased by 54.7, 35.5 and 34.1%, respectively.

scholarly journals A Three-Stage Data-Driven Approach for Determining Reaction Wheels’ Remaining Useful Life Using Long Short-Term Memory

Electronics ◽  
2021 ◽  
Vol 10 (19) ◽  
pp. 2432
Author(s):  
Md Sirajul Islam ◽  
Afshin Rahimi
Keyword(s):  
System Parameter ◽  
Short Term Memory ◽  
Mean Squared Error ◽  
Remaining Useful Life ◽  
Health Index ◽  
Short Term ◽  
Term Memory ◽  
Squared Error ◽  
Useful Life ◽  
Long Short Term Memory

Reaction wheels are widely used in the attitude control system of small satellites. Unfortunately, reaction wheels failure restricts the efficacy of a satellite, and it is one of the many reasons leading to premature abandonment of the satellites. This study observes the measurable system parameter of a faulty reaction wheel induced with incipient fault to estimate the remaining useful life of the reaction wheels. We achieve this goal in three stages, as none of the observable system parameters are directly related to the health of a reaction wheel. In the first stage, we identify the necessary observable system parameter and predict the future of these parameters using sensor acquired data and a long short-term memory recurrent neural network. In the second stage, we estimate the health index parameter using a multivariate long short-term memory network. In the third stage, we predict the remaining useful life of reaction wheels based on historical data of the health index parameter. Normalized root mean squared error is used to evaluate the performance of the various models in each stage. Additionally, three different timespans (short, moderate, and extended in the scale of small satellite orbit times) are simulated and tested for the performance of the proposed methodology regarding the malfunction of reaction wheels. Furthermore, the robustness of the proposed method to missing values, input frequency, and noise is studied. The results show promising performance for the proposed scheme with accuracy in predicting health index parameter around 0.01–0.02 normalized root mean squared error, the accuracy in prediction of RUL of 1%–2.5%, and robustness to various uncertainty factors, as discussed above.

scholarly journals Data Preparation and Training Methodology for Modeling Lithium-Ion Batteries Using a Long Short-Term Memory Neural Network for Mild-Hybrid Vehicle Applications

2020 ◽  
Vol 10 (21) ◽  
pp. 7880
Author(s):  
Daniel Jerouschek ◽  
Ömer Tan ◽  
Ralph Kennel ◽  
Ahmet Taskiran
Keyword(s):  
Neural Network ◽  
Lithium Ion Batteries ◽  
Short Term Memory ◽  
Mean Squared Error ◽  
Lithium Ion ◽  
Dense Layer ◽  
Intermediate Step ◽  
Short Term ◽  
Term Memory ◽  
Long Short Term Memory

Voltage models of lithium-ion batteries (LIB) are used to estimate their future voltages, based on the assumption of a specific current profile, in order to ensure that the LIB remains in a safe operation mode. Data of measurable physical features—current, voltage and temperature—are processed using both over- and undersampling methods, in order to obtain evenly distributed and, therefore, appropriate data to train the model. The trained recurrent neural network (RNN) consists of two long short-term memory (LSTM) layers and one dense layer. Validation measurements over a wide power and temperature range are carried out on a test bench, resulting in a mean absolute error (MAE) of 0.43 V and a mean squared error (MSE) of 0.40 V2. The raw data and modeling process can be carried out without any prior knowledge of LIBs or the tested battery. Due to the challenges involved in modeling the state-of-charge (SOC), measurements are used directly to model the behavior without taking the SOC estimation as an input feature or calculating it in an intermediate step.

Sign in / Sign up

Export Citation Format

Share Document