Search

Short - term power demand forecasting plays an important role in the rational distribution of power utilization, reducing energy waste and enhancing the grid - connected operation of the power system. Using the single model of Lhe seasonal aulo regressive inLegraLed moving average Lo forecasl eleclriciLy demand will limil iLs prediclion accuracy. In order lo improve Lhe prediclion accuracy of Lhe SARIMA model, Lhe SARIMA - GS - SVR combined forecasting model is proposed in this study. The grid search algorithm is used to bring the residual predicted by SARIMA into the support vector regression model for paran1eter training, and the best parameters for optimization are brought into the SVR to predict the residuals. The obtained residual prediction results and the SARIMA prediction results are added together for comprehensive analysis. SARJMA, SVR, GS - SVR and SARJMA - GS - SVR forecasLing models are eslablished, and California's hislorical elecLriciLy demand dala is Laken as an example Lo predicL Lhe 24 - hour elecLricily demand in California on a cerLain day. In order Lo re11ecl Lhe overall superiorily of Lhe model, Lhe exponential smoothing method is selected as an irrelevant benchmark model for experimental comparison. The results show that compared with the SARJMA model, the prediction accuracy of the SARIMA - CS - SVR model is increased by 29. 181 2 %, and the three error index values of the SARJMA - CS - SVR model such as MAE, MAPE and RMSE are significantly lower than the other four models. [ABSTRACT FROM AUTHOR]

Anomaly detection through data analysis helps to accurately identify abnormal behaviors, improving service quality and decision-making capabilities. However, due to the temporal and spatial dependence of multi-dimensional time series data and the randomness of ab normal events, the existing methods still have certain limitations. Regarding the issue above, this paper proposed a multi-dimensional time series data anomaly detection method MBCLE, which combined new statistical methods and bidirectional convolutional LSTM. The method introduced a stacked median filter to handle point anomalies in the input data and smooth data fluctuations, and designed a neural network predictor combining Bi-ConvLSTM and Bi-LSTM for data modeling and prediction. It smoothed the prediction errors using bidirectional recurrent exponentially weighted moving average ( BrEWMA) . The method used dynamic threshold to calculate the threshold to detect contextual anomalies. The experimental results show that MBCLE has good detection performance and each step contributes to the performance improvement. [ABSTRACT FROM AUTHOR]

The traditional RGB-based action recognition method is vulnerable to background factors such as illumination and occlusion, which makes the recognition accuracy and real-time performance low. In order to solve the above problems, a 2 D skeleton action recognition method incorporating a hierarchical attention mechanism was proposed. Firstly, the human skeleton joints in the video were extracted using the human pose estimation algorithm OpenPose. Secondly, the missing and abnormal points in the extracted skeleton data were pre-processed by the mean-completion method and exponential smoothing method. Finally, a network model based on CNN-LSTM fused with hierarchical attention mechanism, CNN-HALSTM, was constructed to realize action classification. The results show that the recognition accuracy of this method in interactive motion and KTH data sets is 96.73% and 98.35%, and the model parameters are significantly reduced, which has the advantage of real-time performance and is better than other methods of the same type. [ABSTRACT FROM AUTHOR]

An optimized relevance vector machine（RVM）life prediction method is proposed for the remaining life of auxiliary power unit（APU）. Firstly，an improved kernel function is proposed by taking into account both efficiency and accuracy. Furthermore，the beetle antennae search（BAS）algorithm is applied to optimize the kernel parameters of RVM. Secondly，through analyzing the historical A13 message and maintenance record data of an airline 􀆳 s APU，the exhaust gas temperature（EGT）parameters are extracted，corrected， and noise reduced，and the turbine performance degradation pattern library of EGT is established with polynomial regression. Finally，it is proved that compared with the traditional RVM，the efficiency and accuracy of the proposed algorithm in the APU turbine life prediction are improved by 40%，20%， respectively. In addition，the optimal initial step size and input dimension in the model are determined based on sensitivity analysis. [ABSTRACT FROM AUTHOR]

To realize the online real-time scrolling optimization of the heat transfer learning coefficient and phase transformation rate learning coefficient in a coiling temperature control(CTC) model， the coiling temperature adaptation between strip segments was studied. Firstly， an equilateral triangle of learning coefficient was built， where the initial learning coefficent adopted by strip segment was in its center of gravity. Based on the learning coefficient on the triangle’s vertices， the coiling temperatures were predicted by the strip temperature model and then the first-order partial derivative of each learning coefficient to coiling temperature can be also obtained. Secondly， the instantaneous value of incremental learning coefficient can be calculated on the basis of the computed partial derivative value and temperature deviation between the predicted temperature and the measured one. There after， the instantaneous values were learned according to the learning rate， followed by the data validation and smoothing. Finally， the new incremental learning coefficients were delivered to the CTC model to update the cooling schedule of each strip segment located in laminar cooling zone. The results in practice show that the adaptive learning method can respond quickly to the change of rolling speed and the coiling temperature along the strip can be controlled more accurately. [ABSTRACT FROM AUTHOR]

[Objective] Estimating soil total potassium content quickly and accurately is very important to precision agriculture. [Method]Based on soil spectral data and soil total potassium content data obtained from laboratory analysis, the relationship between soil spectral and soil total potassium content was studied. The smooth spectral data of soil total potassium were treated with the first derivative, second derivative and reciprocal logarithmic to select the spectral indicators with the highest correlation with soil total potassium content. Finally, a model was established to predict soil total potassium content. [Result] Spectral variables based on first derivative are the best spectral index for estimating soil total potassium content. The predicted accuracy of hyperspectral inversion model(y=2E+06x2+11328x+16. 372) was the best with R2 of 0. 64 and RMSE of 4. 850 g/kg. [Conclusion] It is feasible to rapidly estimate the soil total potassium content of Guangdong Province by using this model. [ABSTRACT FROM AUTHOR]

Existing studies have failed to capture the complex social location transition patterns of suspects and lack the capacity to address the issue of data sparsity. Therefore, we propose a location prediction model called SSLP (spatio-temporal semantics location prediction) to enhance the location prediction performance. Firstly, the similar suspect groups of the target suspects are extracted using the distributed proximity of the suspects in different semantic periods and semantic positions.Then, their mobility data are applied to estimate transition frequencies and temporal visiting probabilities for unobserved locations based on a KDE (kernel density estimating) smoothing method. Finally, by a Bayesian-based formula, the spatio-temporal prediction for the individual suspect can be realized. In the experiments with the location recording data set consisting of 158 suspects and their 17 539 location records from January to June 2013 in W city, SSLP model outperforms baseline algorithms by 40%-50%, validating its adaptability for data sparsity problem. [ABSTRACT FROM AUTHOR]

The smoothing-type algorithm has been successfully applied to solve various optimization problems. Based on a new class of smoothing functions, in this paper, we propose a smoothing-type algorithm with a mixed line search for solving the nonlinear complementarity problem. Under suitable conditions, the proposed algorithm is well-defined and maintains global convergence and local superlinear convergence. Preliminary numerical results are also reported and demonstrate the efficiency of the proposed algorithm. [ABSTRACT FROM AUTHOR]

The objective function of the nonsmooth optimization problems was locally Lipschitz and the feasible set of that consisted of a set of equality constrained smoothing and convex function. The nonsmooth function was conversed into smooth function by being applied with the smoothing approximate techniques. Moreover, it modeled the Lagrange neural network by a class of differential equations, which could be implemented easily. The methodology was based on the Lagrange multiplier theory in optimization and seeked to provide solutions satisfying the necessary conditions of optimality. It proved that any equilibrium point of the network was a subset to the critical point set of primal problems, when the objective function of primal problems was convex, the minimum set coincided with the equilibrium point set of the network. Finally, it presented a simulation experiment to illustrate above theoretical finding. [ABSTRACT FROM AUTHOR]