8results about How to "Maximize performance" patented technology

The present application relates to the field of artificial intelligence, and provides a hardware-aware ONNX graph optimization method and a hardware-aware graph optimization and compilation engine, which breaks the barrier between hardware-independent optimization and hardware-dependent compilation, optimizes and compiles an original ONNX model calculation graph based on the hardware characteristics of an AI chip through a unified and closed-loop optimization framework, realizes deep collaborative optimization from the ONNX calculation graph to the AI chip code, and thus maximizes the performance of the AI chip.

This invention discloses a pellet preparation process based on fibrous bentonite slurry-enhanced pelletizing, belonging to the field of pellet preparation technology in the iron and steel industry. The process includes: wetting iron concentrate to 2.5%~3.5% moisture content; mixing with a first portion of bentonite, followed by sequential high-intensity mixing and milling to obtain pelletizing material; feeding the material into a pelletizer; after the mother pellets are prepared, atomized spraying of fibrous bentonite slurry replaces the addition of water during pelletizing, allowing the pellets to grow to qualified green pellets. The fibrous bentonite slurry is prepared by mixing a second portion of bentonite and water at a mass ratio of 1:5~20, and stirring for 4~6 minutes at a revolution speed of 120~130 r / min and a rotation speed of 280~290 r / min. This invention achieves uniform distribution of bentonite within the pellets through graded moisture control and precise spraying, reducing the total bentonite usage from 1.6% to 1.2% while ensuring pellet quality, significantly reducing bentonite consumption and achieving cost reduction, efficiency improvement, energy conservation, and emission reduction in pellet production.

The application relates to the technical field of kitchen appliances, and discloses a range hood. The range hood comprises a fan assembly and a pipeline assembly, the fan assembly comprises a machine box and a fan arranged in the machine box, an air inlet of the machine box is arranged on the machine box, the pipeline assembly is connected with the machine box, an inner cavity of the pipeline assembly is communicated with the air inlet of the machine box, a flow guide outer expansion structure is arranged at the connecting position of the pipeline assembly and the machine box, the area of the flow guide outer expansion structure is adjustable, and the angle of the flow guide outer expansion structure relative to a horizontal plane is adjustable. The range hood provided by the application has the advantages that the airflow turning angle is large, the flow loss is reduced, the tail edge separation vortex is weakened, the effective flow area is increased, the performance of the fan can be further improved, the flow guide outer expansion structure can be adjusted according to different airflow flow states, the flow guide outer expansion structure is matched with the performance of the fan, the performance of the fan is maximally exerted, and the adaptability to the flow working condition is further improved.

The application discloses a kind of fusion network reconstruction and real-time scheduling collaborative optimization method, device and equipment, it is related to power system dispatching control technical field, and the section flow prediction result is obtained by sensitivity matrix, and section flow prediction result is determined to exceed limit state, and can quickly determine limit section;Then in the case where section exceeds limit, with the target of eliminating overload and maximizing new energy consumption, network reconstruction optimization model is established and power supply network is reconstructed, obtains the power supply network after reconstruction, can quickly remove the hard constraint of line overload and section limit, finally constructs dispatching optimization model, and real-time AGC dispatching optimization is carried out to the power supply network after reconstruction using dispatching optimization model, can realize economic operation and consumption maximization, optimize resource allocation within the safe topological boundary, guide dispatching system to preferentially reduce traditional unit output, through the collaborative optimization of network reconstruction and real-time scheduling, effectively improve the adjustment speed.

The invention relates to a cooperative control method and a control circuit for an SRAM (Static Random Access Memory) integrated zero-area charge pump, which are characterized in that strict isolation in a time domain is realized by dividing an SRAM reading window and a charge pump high-noise working window in a shared master clock period; a self-adaptive timing calibration mechanism based on BIST (built-in self-test) is introduced, a programmable delay unit is scanned in a chip test or power-on stage, the bit error rate of an SRAM is quantized to determine the individualized optimal timing configuration of each chip, the configuration is solidified in an on-chip nonvolatile memory, and the on-chip nonvolatile memory is used for realizing on-chip test or on-chip power-on. Residual noise and time sequence uncertainty caused by process mismatch and PVT drift are effectively suppressed, and the SRAM reliability and the large-scale mass production yield are improved.

The invention belongs to the technical field of coatings, and particularly relates to a high-temperature-resistant coating and a preparation method thereof.The high-temperature-resistant coating is prepared from, by weight, 45-55 parts of organic silicon resin, 8-10 parts of flexibilizer, 6-8 parts of damp-heat-resistant composite additive, 15-20 parts of curing agent, 0.5-0.8 part of defoaming agent and 30-35 parts of solvent; the damp-heat-resistant composite additive is prepared by performing surface modification on potassium titanate whiskers and talcum powder through a silane coupling agent, then reacting with aluminum dihydrogen phosphate to form a ceramic coating layer, performing gradient drying and curing, and finally compounding with vanadium pentoxide, molybdenum trioxide, zinc oxide and boron nitride. According to the additive, the interface bonding strength, the high-temperature stability and the damp-heat aging resistance of a coating are remarkably improved, and the obtained coating is particularly suitable for protecting sealing parts of heavy oil thermal recovery equipment and the like in extreme high-temperature and high-humidity environments and has excellent comprehensive performance.

This utility model relates to the technical field of cosmetic lotion pumps, and discloses a magnetized small molecule lotion pump for cosmetics. It includes: a lotion pump threadedly connected to the top of a lotion bottle; the lotion pump includes a pump cap housing and a pump nozzle; the pump cap housing contains a pressing component and a magnetizing component, the pressing component and the magnetizing component are internally connected, and the pressing component changes the internal pressure by pressing, forming a negative pressure to generate adsorption force; this utility model utilizes the unique closed-loop arrangement of the first permanent magnet component (four sets of permanent magnets connected in a positive-negative-positive-negative configuration) and a spiral channel design, built into the pump cap housing, to allow the lotion to fully receive the magnetic field during flow, effectively changing the molecular structure of the active ingredients in the cosmetic and enhancing their activity. After magnetization, the lotion molecules can more easily penetrate the stratum corneum of the skin and reach deeper into the skin's basal layer, significantly improving the penetration rate of the active ingredients.

The invention relates to the technical field of remote sensing image application, and particularly provides a universal whole-process remote sensing image semantic segmentation model construction method, which is used for performing sample screening based on feature codes stored by remote sensing images in a sample acquisition area in combination with a hierarchical clustering algorithm so as to obtain diversified sample images which are uniformly distributed; labeling a sample image and inputting the sample image into the remote sensing large model until a whole sample data set label is obtained; according to the method, the coupling of deep learning sample construction and model training is realized, and the efficiency and the performance of the semantic segmentation model taking the remote sensing large model as the backbone are maximized.