6results about How to "Achieve robustness" patented technology

The application discloses a sample model training method and system based on data driving, relates to the technical field of model training, and comprises the following steps: obtaining an initial sample set and inputting the initial sample set and a noise vector into a generator together to generate an initial prediction sample; constructing a composite feature tensor based on the initial sample, monitoring the difference between the prediction sample and a reference label in a feature space of a discriminator and the change trajectory of a total loss of the generator, and outputting a training termination condition tensor; feeding back the tensor to the generator as a condition constraint to update the prediction sample, and inputting the prediction sample and the initial sample set into the discriminator together to perform adversarial training, wherein the total loss of the generator is a weighted sum of an adversarial loss and a regularization loss, and network parameters are updated in reverse propagation according to the total loss; in the process of continuous updating of the network parameters, feature resampling is triggered to dynamically correct training data, and a final generator network is output when the difference between two consecutive rounds of training reaches a convergence threshold, so that the method realizes closed-loop correction of adaptive determination of training convergence.

This invention provides a calibration-free path reconstruction method for buried cables based on Topo-NeRF and active sensing. Addressing the issues of traditional magnetic field inversion relying on calibration and being prone to mismatch in environments with strong interference such as reinforced concrete, the method preprocesses raw data to obtain a spatiotemporal feature matrix. This matrix undergoes feature extraction, spatial alignment, and drift compensation to eliminate extrinsic parameter calibration, resulting in multi-source fused data. This multi-source fused data is combined to characterize the cable's magnetic field using a continuous three-dimensional topological manifold. An implicit neural radiation field is constructed and coupled with differentiable electromagnetic rendering. The cable body and branch nodes are separated within a persistently cohomologically constrained topological bottleneck layer. The initial reconstruction results are combined with conditional diffusion and ground-penetrating radar dielectric priors to enhance low signal-to-noise ratio weak fields, ensuring consistency between the reconstruction results and the underground physical structure. Based on reconstruction uncertainties, topological entropy is calculated, and online acquisition trajectories are planned to form a closed-loop active sensing system. Finally, a three-dimensional path model is output, enabling rapid detection of underground cables.

The application discloses a channel optimization method for configuring an electroencephalogram monitoring device, and belongs to the technical field of biomedical signal processing and artificial intelligence. The method obtains high-density multi-channel electroencephalogram data and a depression risk label of a subject group, pre-processes and segments the electroencephalogram data, extracts multi-domain features such as time domain, frequency domain, nonlinear entropy value and inter-channel connectivity of each physical channel, and trains a machine learning classification model in a cross-validation framework. Further, a channel-level grouping permutation importance strategy is used to calculate channel contribution, and a minimum optimal channel combination is determined based on contribution ranking and a performance threshold to generate an algorithm deployment package containing a channel mask index and a lightweight classification model. The application significantly reduces the number of electroencephalogram channels, reduces hardware complexity, power consumption and computing load under the premise of ensuring the performance of depression risk assessment, and is suitable for engineering deployment of low-density electroencephalogram monitoring devices.

This invention discloses a stepped steer-by-wire redundancy control method and system for steering actuator failure, applicable to intelligent electric vehicles with independent steering and drive for all four wheels. When steering actuator failure is detected, the location of the failed wheel is determined and its steering angle is locked. Based on the state of the remaining healthy steering wheels, the maximum yaw moment they can provide is estimated in real time using a tire model. The total required yaw moment is calculated based on driver input and vehicle status. By comparing the two values, a compensation mode is automatically selected: if the steering force is sufficient, a pure steering compensation mode is entered, optimizing only the steering angle of the healthy wheels; if the steering force is insufficient, a hybrid compensation mode is entered, first controlling the healthy wheels to operate in the optimal efficiency state, then calculating the missing steering force step size and compensating through differential torque. This invention pre-sets differentiated strategies for multi-wheel failure scenarios, achieving stepped safety protection from single-wheel to multi-wheel failure, improving vehicle safety, and is low-cost and robust.

The application discloses an energy management method for a new energy three-electricity system of an engineering vehicle, and relates to the technical field of new energy three-electricity systems.The method comprises the following steps: S100, analyzing the driving requirements of the engineering vehicle; S200, intelligently deciding the power; and S300, executing in real time based on feedforward.The application constructs a three-layer closed-loop control architecture of accurate requirement perception, intelligent power decision and safe feedforward execution, and constructs an energy management method with intelligent decision capability and strong execution force.The method not only eliminates the risk of engine stalling due to the superposition of electric-hydraulic load from the root by feedforward constraint based on real-time hydraulic power, but also realizes intelligent scheduling and fine management of battery and fuel energy by means of fixed electric quantity following and regenerative braking recovery strategy.Meanwhile, the safety execution is ensured by battery power boundary protection and anti-integration fan mechanism, so as to comprehensively guarantee the stability and service life of the three-electricity system.

The application discloses a human posture recognition system based on a wearable sensor and belongs to the technical field of sensor measurement. The human posture recognition system collects foot bottom pressure through a pressure sensor module, filters and denoises and performs smooth processing through a signal processing module, and wirelessly transmits sensor signals to an upper computer through a wireless transmission module. The upper computer adopts a hidden Markov algorithm and a random forest fusion algorithm to recognize postures. The system comprises a pair of flexible pressure sensor arrays, the sensor arrays are worn on the foot bottom to collect foot bottom pressure signals, the signal processing module is completed by the upper computer, the wireless transmission module selects a Bluetooth module, and the upper computer performs real-time recognition after training the model. The system improves the recognition rate, is high in adaptability, and is suitable for medical treatment, human-computer interaction, sports training and other fields.