7 results about "Structural risk minimization" patented technology

The invention discloses a sugarcane planting yield prediction and evaluation method and system based on edge calculation, and relates to the technical field of edge calculation. The method comprises the following steps: dividing a selected target planting area into N field units, and deploying an edge computing node on each field unit to obtain meteorological environment data; the technical key points are as follows: support vector regression is adopted as a local prediction model, optimization is performed in combination with insensitive loss and regularization terms, a prediction model with high generalization ability can be stably trained under a small sample condition so as to cope with selected planted crops, and by utilizing a structural risk minimization principle of SVR, a prediction model with high generalization ability can be obtained. Dynamic balance between fitting effectiveness and robustness is realized after corresponding processing in the target function; meanwhile, the SVR only reserves key historical samples as support vectors, real-time reasoning can be carried out, and then a personalized prediction system in which one field unit corresponds to one model is effectively constructed.

This application discloses a method, apparatus, electronic device, and program product for screening high-net-worth users, relating to the field of user profiling technology. The method includes: acquiring feature data of users to be screened, including user occupation information, terminal price information, luxury goods interaction information, car owner identity information, and average monthly consumption contribution; inputting the feature data into a user net worth screening model to obtain the net worth screening result for the users to be screened, the user net worth screening model being trained based on the XGBoost ensemble learning algorithm and through structural risk minimization; and determining whether the users to be screened are high-net-worth users based on the net worth screening result. This application, by using a user net worth screening model to determine whether a user belongs to high-net-worth users based on multi-dimensional feature data, effectively improves the accuracy of identifying high-net-worth users from communication users.

The invention discloses a domain adaptive method for solving subject differences in a motor imagery brain-computer interface, and belongs to the technical field of transfer learning in the motor imagery brain-computer interface. According to the method, the problem of low classification accuracy of the MI-BCI system caused by electroencephalogram signal difference between subjects is solved. According to the method, the information of source domain and target domain samples is fully utilized, and the recognition precision and robustness of the motor imagery task are remarkably improved in combination with data processing and a classification algorithm. Through efficient data preprocessing and a feature extraction method in domain adaptive manifold embedding, the consistency of data feature mapping of a training set and a target domain is ensured. The classifier optimization based on the structural risk minimization principle further enhances the classification performance. Through a feature fusion and voting mechanism, the credibility of label classification is effectively improved, and the method can effectively avoid the problem of low classification accuracy of an MI-BCI system caused by electroencephalogram signal difference between subjects. The method can be applied to electroencephalogram signal classification in a motor imagery brain-computer interface.

The embodiment of the present invention discloses a robust state estimation method for grid voltage sag, which relates to the technical field of power quality monitoring of power systems and can improve the robustness and accuracy of grid voltage sag state estimation. The present invention includes: after establishing a voltage sag disturbance signal model for state estimation, collecting voltage sag monitoring waveform data and inputting it into an improved nonlinear filtering system; using the improved nonlinear filtering system, performing dynamic phasor measurement of the voltage sag parameters of each monitoring point; establishing a structural risk-minimized robust state estimation model for voltage sag, using the instantaneous voltage phasor of each monitoring point as the basic quantity measurement for state estimation, and simultaneously using the instantaneous voltage phasor of non-monitoring points as the state quantity; using the structural risk-minimized robust state estimation model for voltage sag, adopting a quasi-real-time incremental support vector machine algorithm driven by a hybrid model and data, to obtain a robust state estimation result for distribution network voltage sag.

A domain self-adaptive method for solving subject difference in motor imagery brain-computer interface belongs to the technical field of transfer learning in motor imagery brain-computer interface. The method solves the problem of low classification accuracy of the MI-BCI system caused by the difference of the electroencephalogram signals between subjects. The method makes full use of the information of the source domain and the target domain samples, combines data processing and classification algorithms, and significantly improves the recognition accuracy and robustness of the motor imagery task. Through efficient data preprocessing and feature extraction method in domain self-adaptive manifold embedding, the consistency of the feature mapping of the training set and the target domain data is ensured. The classifier optimization based on the principle of structural risk minimization further enhances the classification performance. Through feature fusion and voting mechanism, the reliability of label classification is effectively improved. The method can effectively avoid the problem of low classification accuracy of the MI-BCI system caused by the difference of the electroencephalogram signals between subjects. The method can be applied to the electroencephalogram signal classification in motor imagery brain-computer interface.

The invention discloses an optimal kernel combination method for large-scale multi-label data stream classification in the technical field of machine learning, and the method comprises the steps: introducing a random Fourier feature RFF technology, and enabling a plurality of predefined kernel functions to be explicitly mapped into fixed-dimension feature vectors; linear increase of storage and calculation overhead along with data volume caused by kernel matrix implicit calculation is effectively avoided; the learning robustness is improved by adopting a fixed label threshold strategy; an improved online updating mechanism is provided, the problem of structural risk minimization of a multi-label kernel classifier and a combination coefficient thereof is directly solved, the combination coefficient is updated through index gradient descent, and an upper bound is constructed by using a Jensen inequality to realize independent and efficient online gradient descent updating of each single-kernel classifier, so that the classifiers and the combination coefficient are synchronously optimized, and the robustness of the multi-label kernel classifier is improved. Complementary advantages of different kernels are fully mined, multiple rounds of learning of the same sample are supported, and the method has higher prediction accuracy and expansibility and is suitable for large-scale data stream scenes such as text classification and image annotation.

The present invention discloses a source domain selection method for multi-source EEG migration. First, the tangent space features and Grassmann manifold features are extracted to minimize the difference in the marginal probability distributions between the source domain and the target domain. After obtaining the manifold features, the structural risk minimization and the minimization of the conditional probability distribution difference between the source domain and the target domain are used as the objective function to train the classification model for each source domain. Each classifier predicts the target domain respectively, and the prediction results of different source domains are integrated by voting. After the first iteration, a classifier is trained for each source domain respectively, and finally a multi-source classifier is generated by voting, thus meeting the conditions of LSA. At this time, an LSA is performed to obtain the transferability estimation values of different source domains, and k source domains are selected. In subsequent iterations, it is only necessary to repeat the iteration of training the classifier for the remaining source domains, improving the operation efficiency.