10results about How to "Reduce divergence" patented technology

The invention provides a multi-agent strategy network training method and device, a strategy generation method and device, equipment and a medium. The multi-agent strategy network training method comprises the steps that local observation states and global observation states of all agents and rewards for executing action strategies in the execution process of the action strategies generated based on a currently updated strategy network are collected; determining a target obstacle function value corresponding to each safety evaluation network based on the local observation state and a pre-constructed random discrete diagram control barrier function; determining the security constraint condition of each agent based on the approximate value of the obstacle function output by each security evaluation network; determining an advantage value of each agent action based on a state value prediction value output by the task value evaluation network and the collected rewards; and under the constraint of a security constraint condition, guiding the strategy network to perform parameter updating by taking the advantage value as an updating direction signal. According to the method, the training oscillation or divergence phenomenon caused by safety signal distortion is reduced, and a smoother and more stable convergence process is realized.

The embodiment of the invention provides an online reinforcement learning method and system, a storage medium and a degree product, and the method comprises the steps: controlling a robot to execute a target operation task through a strategy model, so as to obtain corresponding first training data; receiving first interaction information from a human expert to construct corresponding second training data; performing time difference loss calculation based on the first training data and the second training data to train a value evaluation model, and obtaining value evaluation information output by the value evaluation model; performing advantage calculation based on the second training data and the corresponding value evaluation information to obtain advantage information; performing fusion loss calculation based on the advantage information to train a strategy model; the fusion loss comprises advantage weighting strategy loss and near-end strategy optimization loss. According to the embodiment of the invention, the success rate of the robot on the target operation task can be improved under the conditions of lower human teaching cost and fewer environment interaction samples.

The application provides a fiber-optic Fabry-Perot sensor based on light beam convergence and a manufacturing method thereof, and relates to the technical field of semiconductors. The sensor comprises a diaphragm layer, an intermediate layer and a bottom layer which are arranged in layers; the diaphragm layer is provided with a first groove, the intermediate layer is attached to the diaphragm layer in a manner that a first layer surface thereof faces the first groove, so as to form a Fabry-Perot cavity based on the first groove; the bottom layer is provided with a second groove, the intermediate layer is attached to the bottom layer in a manner that a second layer surface thereof faces the second groove, so as to form a nano-containing cavity based on the second groove; the second layer surface of the intermediate layer is formed with a micro-nano anti-reflection structure comprising array-distributed anti-reflection units located in the nano-containing cavity; and the bottom layer is formed with a micro-nano collimation structure comprising array-distributed collimation units located in the nano-containing cavity and in a symmetrical positional relationship with the micro-nano anti-reflection structure. The application at least reduces the manufacturing complexity.

The invention discloses an integrated photovoltaic curtain wall system, and relates to the technical field of photovoltaic curtain walls. The system comprises a supporting frame, the supporting frame comprises a plurality of stand columns installed on a building main body structure and a plurality of supporting arms fixed to the stand columns, the stand columns are arranged at equal intervals, the supporting arms are perpendicular to the stand columns, and the supporting arms are arranged along the stand columns at equal intervals; the photovoltaic assembly comprises a first keel, a second keel connected to the lower end of the first keel and a photovoltaic panel installed on the first keel, and the outer end of the first keel and the outer end of the second keel are connected to the vertically adjacent supporting arms respectively. The lower portion of the first keel and the upper portion of the second keel incline in the direction away from the building main body structure. The heat dissipation effect of the photovoltaic panel is improved in a high-temperature environment in summer, the photoelectric conversion efficiency of the photovoltaic panel is improved, and meanwhile, the influence of high-temperature air gathered at the rear part of the photovoltaic panel on the room temperature of the internal space of a building can be reduced.

This invention discloses a rinsing device for non-ferrous metal materials, comprising a circulation tank, an air compressor pump, a delivery pipe, a water pump, a connecting pipe, a water tank, a rinsing pipe, an air supply pipe, an air inlet box, an air inlet pipe, a support rod, a tension steel wire, a check ball, a threaded cylinder, and an installation assembly. The delivery pipe is embedded in the side of the circulation tank, one end of which is connected to the water pump. One end of the water pump is embedded in the connecting pipe, and the other end of the connecting pipe is connected to the water tank. Several internally impacting rinsing pipes are connected between the water tank and the circulation tank. In this invention, water flows at high speed inside the rinsing pipes. The workpiece is fixed inside the rinsing pipes by the installation assembly. After the water flow accelerates inside the rinsing pipes, it impacts the surface of the workpiece. The rinsing pipes can isolate the water flow from atmospheric pressure, increasing the water flow velocity. Simultaneously, the rinsing pipes can reduce water flow dispersion, ensuring the water flow completely impacts the workpiece, thus greatly improving the rinsing effect.

The application relates to the technical field of real-time target detection and discloses a weakly supervised RT-DETR target detection method based on improved pseudo labels. The method generates pseudo labels by weakly supervising label data sets and enhances the pseudo labels, and an improved model comprising a lightweight backbone network, a pseudo label guided cross-scale interaction module and a decoder with a pseudo label position constraint mechanism is constructed. The model is trained end to end by adopting an iterative pseudo label updating strategy. The pseudo label guided cross-scale interaction module dynamically modulates the feature fusion process by using pseudo label information, thereby enhancing the feature discrimination ability of the model for multi-scale targets. The decoder with the pseudo label position constraint mechanism introduces the geometric prior of the pseudo label as a soft constraint into query optimization, thereby improving the positioning accuracy and training stability of the model under weak supervision and reducing the dependence on accurate labeled data.

The utility model discloses an integrated light-emitting diode (LED) light source, which relates to the field of LED light sources, and comprises an LED light source main body, the LED light source main body consists of a lamp bead mounting substrate and LED lamp beads, one end of the lamp bead mounting substrate is fixedly provided with a light condensing component, and the LED lamp beads are positioned in the light condensing component. The light condensation assembly is composed of an annular installation plate, a heat conduction silicone grease layer, an aluminum conical barrel and a conical reflection lens, the heat conduction silicone grease layer is fixedly installed at one end of the annular installation plate, the aluminum conical barrel is fixedly installed at the other end of the annular installation plate, the light condensation assembly is arranged on the lamp bead installation base plate, and meanwhile the LED lamp beads are located in the light condensation assembly; the light condensation assembly can focus light emitted by the LED lamp beads, light divergence and loss are reduced, the light is focused in a specific area, energy waste is reduced, and meanwhile the heat dissipation efficiency of the LED light source body can be improved through the annular mounting plate, the heat conduction silicone grease layer and the aluminum conical barrel on the light condensation assembly.

The utility model discloses a splice type base plate for a refrigerator car, which relates to the technical field of refrigerator cars and comprises a refrigerator carriage and a fixing component, one end of the refrigerator carriage is fixedly connected with a refrigeration component, the other end of the refrigerator carriage is hinged with a carriage door, and the inner bottom wall of the refrigerator carriage is fixedly connected with a base plate. The bottom plate is formed by splicing and combining a plurality of splicing plates left and right, supporting protrusions are arranged on the splicing plates, under the action of first connecting strips, second connecting strips, clamping heads and clamping grooves, the multiple splicing plates are clamped in the left-right direction and the front-back direction, the mounting efficiency is greatly improved, the bottoms of goods are supported through the supporting protrusions, and the goods are more convenient to mount. A ventilation groove is formed between every two adjacent supporting protrusions, so that cold air or water can flow at the bottom of goods and in the refrigerator van, the goods are blocked through the arranged sliding frame and the blocking rod assembly, the goods are fixed, and the stability in the transportation process is improved.

The application discloses a non-divergence interpolation method for fluid motion control equation in sea surface oil spill simulation, which comprises the following steps: solving initial velocity field; performing non-divergence interpolation; updating particle position; if the running time of simulation meets the requirement, ending, otherwise turning to step A to continue simulation. The application can realize oil spill dynamic simulation based on fluid mechanics NS equation, which is a physical-based method, thus effectively improving the real feeling of oil spill behavior. Since the application combines the continuity equation of fluid mechanics, the divergence of velocity field is gradually reduced through multi-step iteration solving method; therefore, the application can guarantee that the interpolation result meets the continuity equation, so that the calculation result is more stable. Since the application meets the continuity equation, the number of iterations required for calculation is less.

The application discloses a complex equipment technical support fuzzy comprehensive evaluation method, which decomposes safety into 25 aspects, adopts the way of scoring by experts at the present level to score in percentage system, then establishes a fuzzy system and fuzzy rules to obtain the fuzzy score of equipment safety, and on the basis, according to the percentage score data of experts at the present level and the comprehensive evaluation score of safety of equipment after completing a task by experts at the upper level, a kind of radial basis neural network based on positive and negative sine mixed convergence oscillation is established, the weight of the neural network is adaptively trained by using the existing historical data and the data of the fuzzy system, and finally the trained network and the fuzzy system are used to evaluate the safety of the equipment technical support. The method combines the fuzzy system and the neural network well, and solves the problem of low accuracy of the single fuzzy system.