Minghua Xie's research while affiliated with Changsha Medical University and other places

... Support Vector Regression (SVR) is a generalization of Support Vector Machine (SVM) that incorporates regression functions into SVM to solve regression problems [39,40]. As a supervised machine learning algorithm, SVR has a high capability in regression modeling [41,42]. SVR is a kernel-based technique in which the kernel function projects the input data into higher-dimensional feature space to find the hyperplane with the lowest error margin and the best fit to the regression line [43,44]. ...

... The maximum harvested power of the MIS-SSHI circuit can reach 3.7 times that of SEH. Wu et al. [27] also presented a multi-input interface circuit using a single inductor, which integrates a series SSHI circuit and a voltage doubler. It can harvest 3 times as much power from two PZTs as a SEH circuit. ...

... However, its efficiency is lower due to the discarded capacitor charge. In contrast, the synchronized switch harvesting on inductor (SSHI), such as the P-SSHI [4], S-SSHI [5], Hybrid SSHI [6], or Triple bias-flip SSHI [7], adopts an LC resonator with an external inductor. The LC resonator in the SSHI scheme changes the polarity of the capacitor charge at , flipping the capacitor voltage from positive to negative [8][9][10][11][12][13][14]. ...

... In the decoupling strategy of the NNIS, the key is the design and construction of the neural network, but the relevant research and literature have not been discussed too much. A typical error in the back-propagation feed-forward neural network (BPNN) is selected in many documents to identify the inverse system (Bu et al., 2019b;Xie and Xie, 2020). There is no detailed description on how to select the parameters and algorithm in the BPNN. ...

... For this part of the experiments aimed at investigating the importance of the features, as it has been advised [7], we applied linear models, support vector machines with linear kernel and simple linear regression models with additional regularization. All the planned experiments were conducted under a recommended cross-validation regime [26], which allows for credible performance analysis. To evaluate the tested models, we employed several widely adopted metrics in the following experiments: the coefficient of determination (denoted as R 2 ), which is commonly used to compare the performance of different models [19], [21], the mean square error (MSE) or its' square root [20], the mean absolute percentage error (MAPE) [21] which should provide a good intuition on the relative scale average model's prediction error. ...

... Moreover, referring to Zhou et al., 25 one can convert the input and input increment constraints imposed in (36) into the LMIs (29)(30). ...

... It makes the feature importance better matched with its influence on the kernel space. Our previous work [10] has analyzed the necessity of feature weighting and verified it by using the grid search (GS) method to select the optimal combination of weights. Grid search is an exhaustive search method because its computational cost is very high. ...

... At the heart of SVR is mapping data to a higher-dimensional feature space, wherein a linear optimal hyperplane is sought. To realize this aim, SVR harnesses the kernel trick to implicitly compute the dot product in the feature space within the original data domain, sidestepping computational intricacies [40]. The choice of kernel function is pivotal to SVR's efficacy, with prevalent kernels encompassing linear, polynomial, and radial basis function kernels. ...