Number Of Nodes In Hidden Layer Research Articles

Short-Term PV Power Prediction Based on Optimized VMD and LSTM

Because of intermittence and fluctuation of photovoltaic (PV) power, it is difficult to enhance prediction accuracy. To sustain high-efficient operation of power system, this paper proposes a hybrid method to predict the short-term PV power. It consists of components separation of PV power, parameters optimization and re-construction of prediction result. Firstly, the methods based on the identifying of feature frequency and mutual information maximum are proposed to optimize the mode number and penalty factor of VMD, respectively. The optimized VMD (OVMD) is used to decompose the complicated fluctuation components of PV power into single component. Then, the improved PSO (IPSO) based on non-linear inertia weight of anti-sine function is proposed to optimize the number of hidden layer nodes, learning rate and iteration number of LSTM network. The optimized LSTM is used to predict each single component of OVMD decomposition. Thirdly, the prediction result of each single component is re-constructed to obtain the final PV prediction power. The experiment result indicates that the prediction accuracy of the proposed method (OVMD-IPSO-LSTM) outperformances the other typical methods. By the improvement of the traditional method (VMD and PSO) and the parameter optimization of LSTM, this hybrid method makes a contribution to the prediction of short-term PV power.

An Improved QZSS Satellite Clock Offsets Prediction Based on the Extreme Learning Machine Method

The Japanese Quasi-Zenith Satellite System (QZSS) has been developed as a GPS (Global Positioning System) complementary system to improve positioning accuracy in the Asia-Pacific region. However, the accuracies of ultra-rapid predicted clocks are not high enough for real-time applications, so it is still a challenge problem. This article focuses on the clock predictions of the new QZSS constellation. Based on the QZSS clock periodic characteristics, an improved clock prediction method combining the spectrum analysis model (SAM) and extreme learning machine (ELM) is proposed with abbreviation as iELM. The key parameters of iELM are selected carefully including the number of hidden layer nodes and activation function. Further, the input length of the iELM network is optimized and discussed thoroughly. For the purpose of assessing the performance of the proposed algorithm, the clock prediction accuracies are compared among GBU-P (the ultra-rapid predicted orbits/clocks provided by GFZ (Deutsches GeoForschungsZentrum)), SAM and iELM methods. It is demonstrated that iELM keeps in a high level of accuracy below 1.0ns with the predicted time increasing from 0 to 18h, and a little larger during the next six hours. And the QZO (Quasi-Zenith Orbit) satellites perform better than GEO (Geostationary Earth Orbit) satellite. Furthermore, precise point positioning (PPP) for both static and kinematic modes are experimentally studied for 13 IGS (International GNSS (Global Navigation Satellite System) Service) MGEX (Multi-GNSS Experiment) stations in the longitude range between 100°E and 180°E. In the static PPP mode, the iELM method is verified to be effective for the GPS/QZSS constellation as positioning accuracy is improved by 28.3%, 57.7% and 47.4% on average in the east, north and up component with respect to GPS/QZSS GBU-P results. Nearly 70.0% stations achieve sub-decimeter accuracy in the vertical component. As for the kinematic PPP, the iELM method based on GPS/QZSS observations performs much better than the others for shorter convergence time and better positioning accuracy. Compared with GPS/QZSS GBU-P, iELM takes at most a half time to get convergence, and the accuracy is enhanced by 27.6%, 23.7% and 13.9% on average in the east, north and up direction respectively.

A Location Prediction Method Based on GA-LSTM Networks and Associated Movement Behavior Information

Due to the lack of consideration of movement behavior information other than time and location perception in current location prediction methods, the movement characteristics of trajectory data cannot be well expressed, which in turn affects the accuracy of the prediction results. First, a new trajectory data expression method by associating the movement behavior information is given. The pre-association method is used to model the movement behavior information according to the individual movement behavior features and the group movement behavior features extracted from the trajectory sequence and the region. The movement behavior features based on pre-association may not always be the best for the prediction model. Therefore, through association analysis and importance analysis, the final association feature is selected from the pre-association features. The trajectory data is input into the LSTM networks after associated features and genetic algorithm (GA) is used to optimize the combination of the length of time window and the number of hidden layer nodes. The experimental results show that compared with the original trajectory data, the trajectory data associated with the movement behavior information helps to improve the accuracy of location prediction.

Prediction of Shear Characteristics of Unsaturated Soil Based on BP Neural Network

The prediction model of shear strength parameters of unsaturated soil based on indoor test data is established by using BP neural network. Five kinds of network models with different number of hidden layer nodes are trained and studied, and the best network model is selected to conduct the prediction. The results show that the optimal BP network model is a single hidden layer structure of 8-16-2. Using this model to predict, the correlation coefficient and regression coefficient between the predicted value and the measured value are high, and the predicted result is reliable, so the method has certain practicability.

Analysis and Application of Grey Wolf Optimizer-Long Short-Term Memory

Long short-term memory (LSTM) is widely applied in both academic and industrial fields. However, there is no reliable criterion on selecting hyperparameters of LSTM. Currently, although some widely used classic methods such as random search and grid search have obtained success to some extent, the problems in local optimum and convergence still exist. In this research, we propose to use grey wolf optimizer (GWO) to search for the hyperparameters of LSTM. Through the method, the superiority of metaheuristic in global optimization and the strength of LSTM in predicting are combined. In this model, number of hidden layer nodes and learning rate of LSTM are set as preys, and grey wolf pack has a simple but efficient mechanism to search for the optimal hyperparameters. The benchmark tests on several basic functions were utilized, and the results were verified by a comparative study with random search, support vector regression and several other regression methods. Specifically, we applied this algorithm in predicting the degradation trend of the airborne fuel pump. As a result, the ergodicity and convergence of the algorithm are proved mathematically based on Markov processes theory. The benchmark tests show that the GWO-LSTM model holds for predicting data with low overall slope and high partial fluctuation. The application in airborne fuel pump shows that, trained by dataset with 5700 points, the proposed model could predict sequence of 300 points with root mean square error 0.617 after 30 iterations of optimizing, which is 2.512 previously. The result further demonstrates that the proposed algorithm is applicable to make prediction with high accuracy. Overall, the effectiveness of GWO-LSTM model is verified from theoretical proof to benchmark tests and then to actual product application.

Multi-objective particle swarm optimisation for mental task classification using hybrid features and hierarchical neural network classifier

Recognition of mental tasks using electroencephalograph (EEG) signals is of prime importance in man machine interface and assistive technologies. Considerably low recognition rate of mental tasks is still an issue. This work combines power spectral density (PSD) features and lazy wavelet transform (LWT) coefficients to present a new approach to feature extraction from EEG signals. A simple but novel neural network classifier called hierarchical neural network is proposed for the task recognition. A novel methodology based on multi objective particle swarm optimisation (MOPSO) to select discriminative features and the number of hidden layer nodes is proposed to improve the classification accuracy. The extracted features are presented to the hierarchical classifier to discriminate left-hand movement, right-hand movement and word generation task. The results are verified on standard brain computer interface (BCI) database and our own B-alert experimental system database. The benchmarking indicates that the proposed work outperforms the state of the art.

Intrusion Detection of UAVs Based on the Deep Belief Network Optimized by PSO.

With the rapid development of information technology, the problem of the network security of unmanned aerial vehicles (UAVs) has become increasingly prominent. In order to solve the intrusion detection problem of massive, high-dimensional, and nonlinear data, this paper proposes an intrusion detection method based on the deep belief network (DBN) optimized by particle swarm optimization (PSO). First, a classification model based on the DBN is constructed, and the PSO algorithm is then used to optimize the number of hidden layer nodes of the DBN, to obtain the optimal DBN structure. The simulations are conducted on a benchmark intrusion dataset, and the results show that the accuracy of the DBN-PSO algorithm reaches 92.44%, which is higher than those of the support vector machine (SVM), artificial neural network (ANN), deep neural network (DNN), and Adaboost. It can be seen from comparative experiments that the optimization effect of PSO is better than those of the genetic algorithm, simulated annealing algorithm, and Bayesian optimization algorithm. The method of PSO-DBN provides an effective solution to the problem of intrusion detection of UAV networks.

Medical rolling bearing fault prognostics based on improved extreme learning machine

The problem studied in this article:the random selection of the input weight and the implicit layer bias of the extreme learning machine leads to the instability of the medical rolling bearing fault prediction result of the algorithm. It requires more hidden layer nodes to ensure the accuracy of the algorithm, duing to this, Ensemble Error Minimized Extreme Learning Machine (EEM-ELM) is proposed. The EEM-ELM uses various error limit learning machines (EM-ELM) trained on different training sets as member classifiers. Member classifier factors are also used, including predictive entropy, which verifies the set’s accuracy and average and output weights as weights. All of these form a composite classifier by weighted linear combination. This method skillfully solves the problem of optimal hidden layer neurons number selection. The normalized energy and permutation entropy of each IMF component obtained by VMD decomposition fault signal are established as feature vectors, and the improved EEM-ELM algorithm is used as the fault diagnosis model for bearing fault classification algorithm. The fault diagnosis model is applied to the classification of bearing fault signals. The analysis of experimental data proves that the classification result of the proposed EEM-ELM algorithm is better than the ELM algorithm. At the same time, the accuracy rate is higher than each member classifier because of proper weighting processing. Apart from this, since the EEM-ELM algorithm integrates the error minimization limit learning machine, the EEM-ELM algorithm does not need to select the optimal hidden layer node number. The EEM-ELM algorithm only needs to specify the maximum number of training set samples that each EM-ELM-based classifier can tolerate misclassification to achieve high stability and high accuracy classification.

An optimized time series combined forecasting method based on neural networks

In the field of time series forecasting, combining forecasts from multiple models significantly improves the forecasting precision as well as often produces better forecasts than each constituent model. The linear method is the main method in the current literatures for it is simpler and more efficient, and usually gives good results. However, the selection of the basic unit models and the determination of combination weights always bring difficulties to combined forecasting model. In addition, there is usually more time consuming in the combining model. To address these problems, this paper proposes an optimized time series combined forecasting method based on neural networks. The principle of the proposed method adheres to two primary aspects. (a) A multi-threaded grid search method is proposed to quickly determine the number of hidden layer nodes in a neural network for a scenario. (b) A method based on sample dynamic partitioning to explore the weight generation mode is proposed to determine the combined forecasting. Empirical results from six real-world time series show that the superiority of our approach in forecasting accuracies.

Fault Diagnosis Method of Photovoltaic Array Based on BP Neural Network

Photovoltaic arrays are prone to various failures due to long-term work. In order to quickly and accurately diagnose the type of failure of the PV array and implement online monitoring of the PV array, this paper proposes the BP neural network for PV array fault diagnosis, and proposes a network search method when training BP neural network. And the K-cross-validation method is used to select the number of hidden layer nodes. The BP neural network fault diagnosis model designed and trained by this method is proved to have high precision.

Parametric model for microwave filter by using multiple hidden layer output matrix extreme learning machine

Smart tuning of a filter depends very much on an accurate parametric model. Here, the authors develop a parametric model for a microwave filter based on an improved extreme learning machine (ELM). First, a coupling matrix is extracted from scattering parameters in undesirable states. The automatic decomposition and modular training strategy greatly reduces the network complexity. Next, by increasing the size of the output matrix of the hidden-layer, the number of hidden layer nodes can be changed. Finally, the model parameters are optimised by combining the particle swarm optimisation (PSO) algorithm and the differential evolution (DE) algorithm. The presented parametric model benefits from the information of the fusion mechanism with a hybrid optimisation algorithm and succeeds in avoiding slow convergence and premature influence on the system. It also achieves the desired training accuracy within the target time. Compared with prediction based on back-propagation (BP) neural networks and least-squares support vector machines (LS-SVMs), the proposed method can be generalised and has a better modelling accuracy.

Pressure prediction of a spark ignition single cylinder engine using optimized extreme learning machine models

In this study, the cyclic of a spark ignition engine using octane fuel is modeled using extreme learning machine, an emergent technology related to single-hidden layer feedforward neural networks (SLFNs). The experimental engine case study was operated with five different engine speeds from 1000 to 3000 rpm, and crankshaft angle from −360° to 360° without exhaust gas recirculation. The mean effective pressure was used to indicate the cyclic variability for the mean of 100 consecutive cycles. In this study the extreme learning machine (ELM), the regularized extreme learning machine and the outlier robust extreme learning machine were applied to predict the conditions of a combustion parameter used to reflect pressure information for entire cycle in a single-cylinder compression ignition naturally aspirated engine. Prediction by ELM models is normally faster than mathematical models employed to solve a set of differential equations by iterative numerical methods. The essence of ELM is that the hidden layer of SLFNs need not be tuned. Nevertheless, the selection of an appropriate ELM topology is crucial in terms of simplicity, velocity and accuracy. The suitable determination of the number of hidden layer nodes (neurons), type of activation function, and sparse connection structure of weights and biases were obtained using a modified biogeography-based optimization approach (BBO), a population-based metaheuristic algorithm inspired on the mathematical model of organism distribution in biological systems. The experimental dataset were used to train ELM models, and the reliability of these models was assessed and compared for two case studies based on performance criteria related to accuracy, sparsity and complexity using a cross-validation procedure. After training, experimental results show that the pressure can be modeled with reasonable accuracy. The results analysis indicated that the proposed optimized ELM and its variants optimized by BBO approaches have potential for prediction the mean effective pressure showed reasonable consistency with the experimental results.

Mathematical modeling for total organic carbon content prediction with logging parameters by neural networks: A case study of shale gas well in South China

The lower Cambrian Niutitang Formation in northwestern Hunan, South China, has already reached its high or over matured stage and is formed with hydrothermal activity and deposition. Thus, it is extremely difficult to predict the total organic carbon (TOC) content accurately by common methods with well-logging data. To solve this problem, we use artificial neural networks for predicting the TOC of the black shales in our study case. We got the input vectors through principal component analysis and based on the relationships and the logging response mechanism between TOC and logging data. In the back-propagation algorithm, some important parameters including the sample databases, the number of hidden layer nodes, transfer function, and weight value adjustment were all optimized correctly in the networks. Then, we built the mathematical model through a large number of learning sample datum and the error function between the actual and expected outputs, and we found that the results are good according to many performance indicators. In testing samples, mean absolute and relative errors are all reduced probably due to the datum ranges and features being focused, but the accuracy also drops when the numbers of participating samples are reduced. Through redefining the [Formula: see text] sample database, we gained more accurate values for the high-TOC source rock. Finally, we think that the results suggest that the method is suitable for shale gas resource exploration under similar geologic conditions and data characteristics.

Dynamic modeling of NOX emission in a 660 MW coal-fired boiler with long short-term memory

With the rapid development of renewables, increasing demands for the participation of coal-fired power plants in peak load regulation is expected. Frequent transients result in continuous, wide variations in NOX emission at the furnace exit, which represents a substantial challenge to the operation of SCR systems. A precise NOX emission prediction model under both steady and transient states is critical for solving this issue. In this study, a deep learning algorithm referred to as long short-term memory (LSTM) was introduced to predict the dynamics of NOX emission in a 660 MW tangentially coal-fired boiler. A total of 10000 samples from the real power plant, covering 7 days of operation, were employed to train and test the model. The learning rate, look-back time steps, and number of hidden layer nodes were meticulously optimized. The results indicate that the LSTM model has excellent accuracy and generalizability. The root mean square errors of the training data and test data are only 7.6 mg/Nm3 and 12.2 mg/Nm3, respectively. The mean absolute percentage errors are within 3%. Additionally, a comparative study between the LSTM and the widely used support vector machine (SVM) was conducted, and the result indicates that the LSTM outperforms the SVM.

Improve Efficiency of Extreme Learning Machine based on Chaos Particle Swarm Optimization Method

Extreme Learning Machine (ELM) and Regularized Extreme Learning Machine (RELM) have advantages of fast training speed and good generalization. However, ELM/RELM often needs numerous number of hidden layer nodes to get better performance. The superabundant nodes in hidden layer maybe lead to low running speed. Thus it is not feasible to use ELM in some fields that require high speed algorithms. Therefore, in this paper, we propose an Improved ELM/RELM Optimized based on Chaos Particle Swarm Optimization (CPSO-ELM/RELM) to reduce the number of hidden layer nodes, but still maintain a desirable accuracy. At the same time, it lowers the running speed compared with other algorithms. To verify the application of this method, we design numerous experiments for ELM and RRELM. Their simulation shows that the approach improves the speed of the algorithms, and the accuracy is still high. This makes it possible to use improved CPSO-ELM/RELM in some system with high real-time requirements.

Back propagation neutral network based signal acquisition for Brillouin distributed optical fiber sensors.

This manuscript proposes a method based on back propagation (BP) neural network and the spectral subtraction method to quickly obtain sensing information in Brillouin fiber optics sensors. BP neural network's characteristics which can realize any complex nonlinear mapping help to determine the frequency shift section(s) information. The training function, transfer function and number of hidden layer nodes of BP neural network are determined with experimental data. The experimental results show that comparing with traditional Lorentz fitting algorithm and edge detection with Sobel operator, the BP neural network is about 1/12 in terms of time complexity with the Lorentz algorithm, about 1/9 with the edge detection based on Sobel operator; while the respective accuracy on determine the frequency shifted section(s) has improved by 79.4% and 27.9%.

Nonlinear Process Monitoring Based on Global Preserving Unsupervised Kernel Extreme Learning Machine

Recently, the unsupervised extreme learning machine (UELM) technique as a nonlinear data mining approach has been employed to diagnose nonlinear process faults. However, during the dimensionality reduction of process observation data, UELM only aims at mining the local structure feature information of process data and lacks of the ability of preserving the global structure analysis of process data, which would perform unsatisfactorily for monitoring nonlinear process. Furthermore, the choice of the optimal number of hidden layer nodes in UELM method is still a challenging problem. To handle these two tough problems, a new fault detection approach on the basis of global preserving unsupervised kernel extreme learning machine (GUKELM) technique is proposed to monitor nonlinear process effectively. In GUKELM technique, the data global structure preserving framework is naturally incorporated into standard UELM to capture the local structure information as well as to preserve the global structure feature information of nonlinear process observation data during the dimensionality reduction. Meanwhile, to tackle the strong nonlinearity of process observation data, the kernel trick is utilized to successfully solve the challenging issue of setting the optimal number of hidden layer nodes. Based on extracted process data low dimensional feature information from GUKELM method, support vector data description algorithm is adopted to construct the monitoring statistic to detect process fault. Finally, the feasibility and effectiveness of the proposed process monitoring strategy is illustrated through a numerical nonlinear system and the continuous stirred tank reactor process which is a typical nonlinear process.

Particle Swarm Optimization-Based Deep Neural Network for Digital Modulation Recognition

Modulation recognition is a major task in many wireless communication systems including cognitive radio and signal reconnaissance. The diversification of modulation schemes and the increased complexity of the channel environment put higher requirements on the correct identification of modulated signals. Deep learning (DL) is considered as a potential solution to solve these problems due to the superior big data processing and classification capabilities. This paper proposes an efficient digital modulation recognition method based on deep neural network (DNN) model. Furthermore, we present the particle swarm optimization (PSO) algorithm to optimize the number of hidden layer nodes of the DNN so as to solve the problem that the traditional DNN is trapped in local minimum values and the number of hidden layer nodes needs selecting manually. In this paper, we utilize the proposed PSO-DNN method to learn characteristics extracted from the modulated signal added by additive white Gaussian noise (AWGN) and to train the network, which can improve the performance of recognition under the condition of low signal-to-noise ratio (SNR). The experimental results demonstrate that the recognition rate on this algorithm has improved by 9.4% and 8.8% compared with methods that adopt conventional DNN and support vector machine (SVM) when SNR equals 0 and 1 dB, respectively. Besides, another experiment compared with the genetic algorithm (GA) also proves that our proposed algorithm is more effective in optimizing the DNN. The proposed method is easy to be implemented so that it has a broad development prospect in modulation recognition.

Neural-network based boost pressure prediction for two-stage turbocharging system of diesel engine

Abstract A boost pressure prediction research based on neural network is carried out for a diesel engine with two-stage adjustable turbocharging system. First, based on the co-simulation using GT-POWER and MATLAB/Simulink model, the dynamic influence of cycle fuel injection quantity and bypass valve opening of the turbine on boost pressure of diesel engine is studied. The boost pressure shows strong nonlinear relationship with the two aforementioned affecting parameters. Second, a nonlinear autoregressive neural network prediction model (NARXNN) with external input is designed to predict the boost pressure of the diesel engine. The cycle fuel injection quantity, bypass valve opening of turbine, and current boost pressure are used as the inputs of the neural network in order to predict the future boost pressure. Then the neural network prediction model is identified based on the mean absolute percentage error (MAPE). The identification results show that the minimum MAPE (3.52%) is achieved when the number of hidden layer nodes is 15. At last, the neural network prediction model for boost pressure is verified by simulation. The one-step prediction simulation results of boost pressure show that the established prediction model can accurately predict the boost pressure, and the MAPE of the verification data is 3.98%. And the multistep prediction simulation results show that the pressure prediction error in the 50-step prediction domain is within 0.05bar, which indicates that the prediction model has high accuracy.

Selecting the Optimal Hidden Layer of Extreme Learning Machine Using Multiple Kernel Learning

Extreme learning machine (ELM) is emerging as a powerful machine learning method in a variety of application scenarios due to its promising advantages of high accuracy, fast learning speed and easy of implementation. However, how to select the optimal hidden layer of ELM is still an open question in the ELM community. Basically, the number of hidden layer nodes is a sensitive hyperparameter that significantly affects the performance of ELM. To address this challenging problem, we propose to adopt multiple kernel learning (MKL) to design a multi-hidden-layer-kernel ELM (MHLK-ELM). Specifically, we first integrate kernel functions with random feature mapping of ELM to design a hidden-layer-kernel ELM (HLK-ELM), which serves as the base of MHLK-ELM. Then, we utilize the MKL method to propose two versions of MHLK-ELMs, called sparse and non-sparse MHLK-ELMs. Both two types of MHLK-ELMs can effectively find out the optimal linear combination of multiple HLK-ELMs for different classification and regression problems. Experimental results on seven data sets, among which three data sets are relevant to classification and four ones are relevant to regression, demonstrate that the proposed MHLK-ELM achieves superior performance compared with conventional ELM and basic HLK-ELM.